Adverse drug reaction (ADR) is an important public health problem associated with a considerable amount of morbidity and mortality. Studies have shown that 5 – 10% of all hospital admissions are attributed to ADR. In Europe, about 197,000 persons die annually from the ADR-related problem. ADRs are the fourth to sixth largest cause of death in the United States. Despite being a leading cause of death, ADRs are largely preventable. The current pharmacovigilance practice rely mainly on spontaneous reporting systems and pharmacoepdemiological studies for detection and reporting of ADRs. However, these monitoring strategies do not capture all the ADR events occurring both in the hospital setting and in the general population. Therefore, additional strategies are needed to improve the monitoring and reporting of ADRs. Underreporting of ADRs is a major challenge for global pharmacovigilance and has a huge impact on the safety of medicines in public health. Although modern techniques are been developed including a web-based system, the existing monitoring strategies are inadequate and need to be integrated with fast-growing technologies.
In this project, we will design and develop a home-based monitoring of potential Adverse Drug Reactions (ADRs) using interactive mobile applications or/and web application. Using this app, both patients and their caregivers (doctors, nurses, etc.) will monitor the patient's responses to new medications. Patients will be able to track and record their reactions to various medications, observed changes. On the other hand, doctors and nurses can analyze these patients tracked data to predict possible adverse reactions to medications and take appropriate actions to prevent further damages.
Research area, student roles & skills
Research area: Our research group is generally interested in designing and evaluating interactive systems such as Web Applications and Mobile applications for promoting health and wellness. These are active Human-Computer Interaction (HCI) research areas. We are mostly interested in using these systems to solve problems that are of global importance especially in the area of health and wellness (e.g, promoting a healthy lifestyle, recovery, medication adherence, and disease management) and safety.
Student roles: The student will join a team of faculty, postdoctoral researchers and graduate students working on various projects at the Human-Computer Interaction Lab. The student would work with graduate students on projects that involve developing and evaluating interactive systems. There is flexibility for the students to focus on a particular application; web application, mobile application, social computing, games, or some combinations. The student would have access to collections of computer science papers and other relevant literature and resources (software and hardware) needed for the project. We hope to publish at least a fully refereed paper at the end of the project and with additional work to be carried out remotely after the end of the Globalink visit we could possibly generate more publications.
Skills required: The student needs to have some design skills, be interested in programming or learning programming. Previous experience with web and mobile application design and development will be of advantage. Familiarity with or interest in pursuing graduate studies in one of the following areas would be highly desirable: human-computer interaction, mobile computing, social computing, game design, visualization, data analytics, machine learning, text mining, information retrieval.
602. Privacy Risks and Secure Embedding Learning in RAG Systems
Retrieval-Augmented Generation (RAG) systems have become increasingly important for enhancing large language models (LLMs) through external knowledge retrieval. By converting documents and user queries into vector embeddings, RAG systems enable efficient semantic retrieval and context-aware text generation. However, recent studies have shown that embedding vectors are not fully privacy-preserving and may leak sensitive information through embedding reconstruction or inversion attacks. In such attacks, adversaries attempt to recover original text, private documents, or user queries from stored or intercepted embeddings, posing serious risks to healthcare, finance, cybersecurity, and enterprise AI systems.
This research project aims to investigate embedding reconstruction risks in modern RAG systems and develop robust privacy-preserving defense mechanisms against embedding inversion attacks. The project will analyze how embedding models, vector databases, retrieval pipelines, and contextual retrieval mechanisms contribute to information leakage. Various attack scenarios, including direct inversion, context-aware reconstruction, and transferable black-box inversion attacks, will be studied to evaluate their effectiveness against state-of-the-art embedding models and RAG architectures.
To mitigate these threats, the research will explore privacy-preserving representation learning techniques such as differential privacy, embedding perturbation, adversarial representation learning, secure vector storage, and retrieval-aware embedding sanitization. The project will further investigate the tradeoff between retrieval utility and privacy preservation to ensure that defense mechanisms maintain retrieval quality and generation accuracy while reducing sensitive information leakage.
The expected outcome of this research is a scalable and secure RAG framework capable of protecting embedding representations against reconstruction attacks while preserving the effectiveness of semantic retrieval and downstream language generation. This work will contribute toward the development of trustworthy and privacy-preserving AI systems suitable for real-world deployment in sensitive and high-impact domains.
Research area, student roles & skills
Research area: My research focuses on developing secure and privacy-preserving AI systems with applications in cybersecurity, healthcare, smart and connected environments, and trustworthy intelligent systems. I have worked on behavioral biometric authentication, federated learning frameworks, adversarial attack and defense mechanisms, and the secure deployment of LLMs and Retrieval-Augmented Generation (RAG) systems. I have published in journals such as IEEE Access and MDPI, and in conferences including FPS, IFIP SEC, SECRYPT, IEEE BigData, and IEEE COMPSAC. I also have experience supervising students under the Mitacs GRI program, supporting their research training and project development.
Student roles: The student will play an important role in supporting the development of secure and privacy-preserving Retrieval-Augmented Generation (RAG) systems against embedding reconstruction and inversion attacks. Their work will begin with an in-depth literature review on RAG architectures, embedding models, vector databases, representation learning, and privacy vulnerabilities in large language model (LLM)-based retrieval systems. The student will assist in designing and implementing experimental frameworks to evaluate how embedding vectors leak sensitive information and how adversaries can reconstruct original text or private data from embeddings.
A major responsibility will involve developing scripts and pipelines to simulate embedding inversion and reconstruction attacks on modern RAG systems. The student will help evaluate privacy leakage risks using reconstruction accuracy, embedding similarity, retrieval consistency, and generation quality metrics. They will also investigate defense mechanisms such as differential privacy, embedding perturbation, adversarial representation learning, secure vector storage, and retrieval-aware embedding sanitization to improve the privacy and robustness of RAG systems.
Additional responsibilities include maintaining reproducible codebases, analyzing experimental results, preparing visualizations and technical reports, and contributing to research publications and presentations. The student will participate in regular research meetings and collaborative discussions throughout the project.
Through this project, the student will gain hands-on experience in RAG systems, embedding security, adversarial machine learning, privacy-preserving AI, and trustworthy language model development for real-world applications.
Skills required: The student should have strong programming skills in Python and knowledge of machine learning, deep learning, and Retrieval-Augmented Generation (RAG) systems. Proficiency with PyTorch or TensorFlow is essential, along with familiarity with large language models (LLMs), embedding models, vector databases, and information retrieval techniques. A solid foundation in probability, statistics, and linear algebra is required. Knowledge of AI security, adversarial machine learning, embedding inversion attacks, and privacy-preserving AI techniques will be considered a strong asset. Experience with open-source LLMs, RAG frameworks, and evaluation tools is desirable.
603. Privacy in Machine Learning
Supervisor: Julia Rubin
University: University of British Columbia (Vancouver campus)
A Membership Inference Attack (MIA) is a type of ML attack where an adversary tries to determine whether specific data was used in training a machine learning model. Model Inversion Attack (MOA) is type of privacy attack, where an adversary uses access to a machine learning model to infer sensitive information about the data it was trained on. Despite solid prior work on theoretical guarantees and safeguards against both attacks, from an empirical point of view, most of the research has focused on evaluating attacks in a synthetic scenario that may not necessarily represent realistic settings.
The goal of this project is to (a) explore the applicability of classic membership inference and model inversion attacks on practical datasets and (b) examine these techniques from a more pragmatic perspective that better aligns with real-world applications, such as estimating model memorization vs. generalization capacity or supporting data provenance analysis.
The research will start with a systematic review of existing techniques, in which we will replicate and analyze results from relevant prior work. We will then select most promising methods and examine how these methods perform under different contexts, such as those outlined above.
Research area, student roles & skills
Research area: Machine learning (ML) systems learn data patterns to build models used for a wide variety of problems. However, it is well-known that ML models memorize features about the data they are trained on, including privacy-sensitive features. This project will look at the privacy properties of ML systems.
Student roles: The student will help analyze existing work on privacy attacks, adapt it to realistic datasets, and evaluate the effectiveness of the developed solutions. The tasks will include developing software and experimental infrastructure, running experiments, analyzing and reporting results, and reading and discussing research manuscripts.
At the end of the project, the student will be familiar with the ML privacy domain and will be able to analyze and discuss major ML privacy considerations. The student will also be familiar with data analysis and will learn to build high-quality ML systems. The student will join weekly project meetings with the group and the PI, will be exposed to ways research groups operate, and will have the chance to collaborate internally and with external partners.
Skills required: The student is expected to: - have solid programming experience and good command of Java, C++, C, or a similar language; - be able to learn new programming languages and frameworks; - have deep understanding of a variety of machine learning techniques; - have deep understanding of algorithms and data structures; - have a good understanding of mobile application environments (at least as an informed mobile user); - be detail-oriented and thorough.
Additional desirable skills (optional): - experience with security and privacy work.
604. Privacy preservation techniques for digital forensics
Supervisor: Mary Oluwasola Adedayo
University: University of Winnipeg
Location: Winnipeg, Manitoba
Start date: 2027-05-31 (flexible)
Disciplines: Computer Science, Electronic Systems, Engg-Computer, Engg-Electrical, Engg-Software, Engg-Systems and Technology, Engineering, Management Information Systems, Mathematics, Physics, Science and Technology, Studies Science and Technology
Privacy preservation of digital evidence is often considered as a contradicting goal to that of uncovering all relevant information about an incident in digital forensic investigations. Although there have been many recent advances in the development of privacy protection methods in general, the application of these methods in digital forensics is lagging. This is due to the challenge of ensuring that the what, why, where, when, who and how questions of digital forensics can still be answered while keeping some information protected. In our earlier work, we explored the application of cryptographic techniques in digital forensics from a theoretical standpoint but many other privacy techniques are yet to be explored.
The goal of this project is to explore the practical application of privacy preservation techniques in digital forensics. We will experiment with simplified implementations of homomorphic encryption, searchable encryption, differential privacy, selective hashing and federated learning for digital evidence collection, and management of data sharing among different jurisdictions. While this work may be extended to many areas of digital forensics, the focus of this project will be limited to application areas including mobile and IoT devices, smart home devices and computers. This will involve a review of existing works and implementation of privacy-preserving techniques and a practical evaluation of their potential use for data privacy in digital forensics. Through this study, we will identify the challenges associated with using specific privacy-preserving techniques and design and test simulated experiments to evaluate their use in digital forensics.
Research area, student roles & skills
Research area: Digital forensics focuses on identifying, preserving, analyzing, documenting, and presenting evidence from digital sources. Despite the large amount of personal information that can often be encountered during analysis, methods for addressing data privacy in digital investigations are still not well defined from a practical view. Many of the existing privacy protection mechanisms also need to be explored from a digital forensics perspective. Our earlier work focused on how cryptographic techniques can be applied in digital forensics investigation but more research is required into the practical analysis of this approach and the application of other privacy protection mechanisms and security algorithms.
Student roles: The student will be required to do background research to understand the field of digital forensics and some of the research advances in privacy protection techniques. The student may be required to implement certain algorithms in an applicable programming language (e.g., Java, Python, C, or C++). The student will be required to learn to use some open-source digital forensics tools and apply them in the analysis of digital evidence. Sample data will mainly be obtained from different digital forensics research data corpora and used to test proposed solutions as appropriate. The student will be involved in writing up results and preparing them for publication. The student should be able to work well with others and be comfortable with working in a research area where the existing literature may be limited.
Skills required: We seek students with prior knowledge of computer security and privacy and an interest in digital forensics. The student must have some mathematics background, demonstrate excellence in courses such as Computer security, Data structures and algorithms, Operating systems, Computer architecture, and system software, and be capable of working with or learning to use the Linux operating system as necessary. The student must be able to work with minimal direct supervision and be willing to ask questions as necessary. Strong communication skills (written and oral) and some experience working on technical projects and writing research articles are required.
605. Privacy-Preserving Analytics for Cross-National Aging Survey Data Using the Gateway to Global Aging Data (G2Aging)
Population aging is a defining global demographic trend, and large-scale longitudinal surveys such as the Health and Retirement Study (HRS) and its international counterparts provide invaluable data for understanding aging processes, health trajectories, and socioeconomic determinants of well-being in later life. The Gateway to Global Aging Data (G2Aging), hosted at the University of Southern California, harmonizes data from over 16 studies spanning 44 countries, creating unprecedented opportunities for cross-national aging research. However, these datasets contain highly sensitive personal health, cognitive, financial, and social information about older adults a population disproportionately vulnerable to data breaches, identity theft, and privacy violations. As data sharing and secondary analysis expand, ensuring robust privacy protection without sacrificing analytical utility becomes critical. This project will investigate and develop privacy-preserving data analytics methodologies tailored to harmonized aging survey data. Interns will conduct a systematic cross-national comparison of data governance frameworks applicable to aging survey data across G2Aging partner countries. They will then design, implement, and evaluate privacy-preserving analytical pipelinesincluding differential privacy mechanisms, data perturbation techniques, and synthetic data generation applied to G2Aging harmonized datasets. The project will benchmark the trade-offs between privacy guarantees and the accuracy of key aging research outcomes, such as cognitive decline trajectories, multimorbidity prevalence, and economic security indicators. Findings will contribute to both the academic literature on data privacy in health research and practical recommendations for data custodians managing sensitive aging survey data internationally.
Research area, student roles & skills
Research area: This research sits at the intersection of data privacy, information security, and population aging. The project focuses on developing and evaluating privacy-preserving analytical techniques for sensitive longitudinal health survey data, drawing on harmonized datasets from the Gateway to Global Aging Data (G2Aging) platform. Research areas include differential privacy, secure computation methods, data anonymization, and cross-national comparative analysis of privacy regulations governing aging-related health data.
Student roles: The two interns will work collaboratively under the supervision of Dr. Ernest Johnson on complementary aspects of the project. Intern 1 will focus on the cross-national regulatory and policy analysis component. This involves conducting a structured comparative review of data privacy and governance frameworks (e.g., PIPEDA in Canada, GDPR in Europe, HIPAA-adjacent frameworks in the U.S.) as they apply to aging survey datasets available through G2Aging. The intern will develop a taxonomy of privacy requirements and identify gaps and inconsistencies across jurisdictions that affect data sharing for aging research. Intern 2 will focus on the technical privacy-preserving analytics component. This intern will design and implement differential privacy and synthetic data generation techniques on selected G2Aging harmonized variables (e.g., health status, cognitive scores, income). The intern will conduct utility-privacy trade-off experiments, measuring how privacy mechanisms affect the accuracy of standard gerontological analyses. Both interns will participate in weekly supervisory meetings, present interim findings at the research group seminar, contribute to a joint working paper, and prepare a conference presentation. Interns will gain experience with real-world harmonized survey data, advanced privacy techniques, cross-national policy analysis, and collaborative interdisciplinary research.
Skills required: Strong programming skills in Python or R. Familiarity with statistical analysis and data manipulation (e.g., pandas, NumPy). Basic understanding of data privacy concepts (differential privacy, k-anonymity) is an asset. Experience with survey data or longitudinal datasets is desirable. Strong written and oral communication skills in English.
606. Privacy-Preserving Artificial Intelligence for Healthcare Analytics Using Federated Learning
Supervisor: abbas Yazdinejad
University: University of Regina
Location: Regina, Saskatchewan
Start date: 2027-05-17 (flexible)
Disciplines: Computer Science, Health Studies, Public Health, Electronic Systems
Healthcare organizations often possess valuable data that could improve disease prediction and clinical decision-making. However, privacy regulations and institutional policies frequently prevent direct sharing of sensitive patient information. Federated Learning (FL) offers a promising solution by enabling institutions to collaboratively train machine learning models while keeping data local.
This project investigates privacy-preserving federated learning techniques for healthcare analytics. The student will explore methods for distributed model training, secure aggregation, differential privacy, and explainable artificial intelligence. Using publicly available healthcare datasets, the project will evaluate how federated learning can improve predictive performance while preserving data confidentiality.
The project combines artificial intelligence, cybersecurity, and health informatics to address real-world challenges in secure data sharing. Students will gain experience with machine learning, deep learning, distributed AI systems, and privacy-preserving technologies while contributing to cutting-edge research with applications in healthcare and public health.
Research area, student roles & skills
Research area: My research focuses on trustworthy artificial intelligence, federated learning, privacy-preserving machine learning, and healthcare analytics. I develop AI systems that enable multiple organizations to collaboratively train machine learning models without sharing sensitive data. The research integrates cybersecurity, explainable AI, and privacy-enhancing technologies to support secure decision-making in healthcare environments. Applications include disease prediction, public health surveillance, and medical data analytics while maintaining privacy, security, and regulatory compliance.
Student roles: The student will assist in data preparation, literature review, model development, experimentation, and result analysis. Responsibilities include implementing machine learning and federated learning algorithms using Python, conducting performance evaluations, visualizing results, and documenting findings.
The student will participate in weekly research meetings, collaborate with graduate students and researchers, and contribute to the preparation of technical reports, presentations, and potential research publications. The internship provides hands-on experience in privacy-preserving AI, healthcare analytics, and scientific research methodologies.
Skills required: Students should have a background in Computer Science, Artificial Intelligence, Data Science, Software Engineering, Health Informatics, or a related field. Experience with Python programming and basic machine learning concepts is required. Familiarity with deep learning frameworks such as PyTorch or TensorFlow is desirable. Knowledge of data analysis, statistics, or healthcare applications is considered an asset but is not mandatory.
607. Privacy-Preserving Health Assistants Using On-Device Large Language Models
Supervisor: Hamid Mansoor
University: University of Manitoba (Winnipeg campus)
Artificial intelligence systems are increasingly being used to support health and wellness applications. However, many current AI systems rely on cloud-based processing, requiring personal data to be transmitted to remote servers. This can create concerns related to privacy, security, latency, and user trust, particularly when sensitive health information is involved.
This project explores how large language models (LLMs) can be deployed directly on smartphones and other personal devices to provide personalized health support while keeping data under the user's control. Students will investigate techniques such as model compression, quantization, and on-device inference that allow advanced AI capabilities to operate within the computational and battery constraints of mobile devices.
A second focus of the project is personalization. While AI systems often perform well when trained on large populations, individuals have unique behaviors, preferences, and health goals. Students will explore privacy-preserving approaches that allow AI systems to learn from personal data without transmitting sensitive information to centralized servers. This may include investigating federated learning techniques, where models are improved collaboratively across many devices while keeping user data local.
Students will gain hands-on experience working with modern large language models, mobile AI frameworks, machine learning systems, and digital health applications. Depending on their interests, they may contribute to model optimization, mobile application development, performance evaluation, personalization algorithms, or experimental studies comparing different deployment approaches.
The project contributes to the development of next-generation health technologies that combine the capabilities of modern AI with strong privacy protections, enabling personalized health assistants that can operate directly on users' devices while maintaining responsiveness, efficiency, and trust.
Research area, student roles & skills
Research area: My research focuses on digital health, mobile and wearable computing, human-computer interaction, and artificial intelligence. I develop systems that combine smartphone and wearable sensor data with machine learning and large language models to support personalized health monitoring, behavior change, and health decision-making. My work emphasizes user-centered design, explainable AI, and practical health technologies that can be deployed on everyday devices.
Student roles: Students will work as part of a research team investigating how large language models and other artificial intelligence technologies can be deployed directly on smartphones and personal devices while preserving user privacy. Their responsibilities may include implementing and evaluating machine learning models, developing software prototypes, analyzing experimental results, and investigating techniques for efficient on-device AI deployment.
Depending on their interests and experience, students may contribute to model compression, quantization, federated learning, mobile application development, performance benchmarking, or data analysis. They may also help evaluate tradeoffs between model accuracy, personalization, battery consumption, memory usage, and response time across different deployment approaches.
A major focus of the project will be understanding how personalized AI systems can adapt to individual users while keeping sensitive information on-device. Students may investigate methods that allow AI models to learn from personal data without transmitting raw information to centralized servers, helping to improve privacy and user trust. They will gain experience working with modern large language models, mobile AI frameworks, and state-of-the-art machine learning systems.
The project emphasizes research, experimentation, and independent problem solving. Students will be encouraged to explore new technologies, learn unfamiliar tools when needed, read current research literature, and contribute ideas to the design and evaluation of novel AI systems. They will participate in regular research meetings, collaborate with graduate students, and communicate their findings through presentations and technical reports. Students may also contribute to preparing research papers, posters, and presentations describing project findings. By the end of the internship, students will have gained hands-on experience conducting research at the intersection of artificial intelligence, mobile computing, privacy-preserving machine learning, and digital health.
Skills required: Applicants should have a background in Computer Science, Software Engineering, Data Science, Artificial Intelligence, or a related field. Experience with programming in Python is required. Familiarity with machine learning, data analysis, mobile application development, wearable technologies, or data visualization is beneficial but not required.
More importantly, I am looking for students who are adaptable, and comfortable learning independently. Successful applicants should be willing to explore new technologies, read technical documentation and research papers, and quickly acquire skills needed for unfamiliar tools and frameworks. Strong problem-solving abilities and the ability to work independently and as part of a team are valuable.
608. Privacy-Preserving Personalized Federated Learning for Biometric Authentication
The rapid advancement of AI has increased the demand for privacy-preserving and decentralized learning systems. Traditional centralized ML frameworks require collecting data from multiple sources into a central server, creating privacy, communication, and regulatory challenges when handling sensitive data. Federated Learning (FL) addresses these issues by enabling decentralized model training where data remains on local devices and only model updates are shared with a central server.
Despite its advantages, FL faces several critical challenges, particularly in biometric authentication systems where clients typically possess only positive-class data corresponding to their own biometric. In such scenarios, each client learns highly personalized and non-IID data distributions, making conventional aggregation methods such as FedAvg ineffective. Direct aggregation of client classifiers often produces unstable global models because different clients represent fundamentally different biometric classes. Existing approaches, including regularization-based FL methods and data-sharing strategies, either fail to scale effectively or introduce additional privacy risks through indirect information leakage.
This research proposes a privacy-preserving personalized FL framework for biometric authentication based on representation learning and split federated learning principles. The proposed system separates the model into two components: a client-side feature extractor and a server-side classifier. Clients locally generate biometric embeddings and share only privacy-preserving feature representations with the server instead of raw biometric data. The server collaboratively trains a global model using the received embeddings and continuously updates the shared feature extractor through federated optimization. Techniques such as embedding perturbation will be explored to protect shared embeddings from information leakage and reconstruction attacks. Personalized federated learning techniques will also be incorporated to improve user-specific authentication performance while maintaining a robust shared embedding space across distributed clients.
The expected outcome of this research is a scalable and adaptive FL-based biometric authentication framework capable of handling heterogeneous positive-class biometric data while preserving user privacy and supporting real-world deployment.
Research area, student roles & skills
Research area: My research focuses on developing secure and privacy-preserving AI systems with applications in cybersecurity, healthcare, smart and connected environments, and trustworthy intelligent systems. I have worked on behavioral biometric authentication, federated learning frameworks, adversarial attack and defense mechanisms, and the secure deployment of LLMs and Retrieval-Augmented Generation (RAG) systems. I have published in journals such as IEEE Access and MDPI, and in conferences including FPS, IFIP SEC, SECRYPT, IEEE BigData, and IEEE COMPSAC. I also have experience supervising students under the Mitacs GRI program, supporting their research training and project development.
Student roles: The student will play an important role in the development of a privacy-preserving personalized Federated Learning (FL) framework for biometric authentication systems. Responsibilities will include conducting literature reviews on FL, embedding perturbation, personalized FL, split learning, representation learning, and biometric authentication systems. The student will assist in designing and implementing experimental FL environments with heterogeneous and extreme non-IID data distributions commonly observed in biometric applications.
A major responsibility will involve developing and evaluating client-side feature extraction models and server-side classification frameworks for privacy-preserving authentication. The student will help investigate embedding perturbation techniques for protecting shared biometric representations against information leakage and reconstruction attacks, along with personalized learning strategies to improve user-specific authentication performance while maintaining a generalized embedding space across distributed clients. They will also explore techniques for handling positive-class-only biometric data, adaptive client participation, and privacy-preserving embedding sharing in FL environments.
Additional responsibilities include implementing and testing FL algorithms, analyzing experimental results, preparing technical reports and research presentations, and contributing to academic publications and software documentation. The student will participate in regular research meetings and collaborative discussions throughout the project.
Through this project, the student will gain hands-on experience in federated learning, behavioral biometrics, representation learning, privacy-preserving AI, and distributed deep learning systems for real-world applications.
Skills required: The student should have strong programming skills in Python and a solid understanding of ML and biometric authentication techniques. Proficiency with key libraries such as TensorFlow and PyTorch is essential. A strong foundation in mathematics- particularly in probability, statistics, and linear algebra- is also required. Familiarity with Federated Learning and split-learning concepts is expected, including aggregation protocols, optimization methods, and regularization techniques. The student should be prepared to work with FL libraries such as TensorFlow Federated (TFF), PySyft, and Flower.
609. Privacy-Preserving and Trust-Aware Machine Learning for Longitudinal In-Home Behavioural Sensing
Supervisor: Rashid Khokhar
University: Algoma University (Sault Ste. Marie campus)
Location: Sault Ste. Marie, Ontario
Start date: 2027-05-04 (flexible)
Disciplines: Computer Science, Engg-Computer, Engg-Software, Statistics, Mathematics, Engg-Systems and Technology
The increasing use of ambient sensing systems in smart home and assisted living environments enables continuous analysis of longitudinal behavioural signals such as movement patterns, voice activity, and device interactions. While these systems offer significant potential for supporting caregiving and health monitoring, they introduce critical challenges related to privacy preservation, trust in system outputs, and efficient computation under resource constraints.
This project investigates privacy-preserving and trustworthy machine learning methods for longitudinal in-home behavioural data. A key technical challenge arises from strong temporal dependencies in behavioural sequences, where observations over time are correlated rather than independent. This complicates the application of privacy-preserving learning methods, as well as the preservation of model utility.
The project will explore privacy-preserving learning techniques adapted for sequential data and assess their impact on behavioural prediction performance. In parallel, it will address the reliability of caregiver-facing alert systems, where the usefulness of a system depends not only on detection accuracy but also on the perceived trustworthiness of generated alerts. High false positive rates can significantly reduce user confidence and long-term adoption. To address this, the project will develop a lightweight reliability-aware alerting mechanism that uses model confidence and short-term behavioural consistency to prioritize or suppress alerts before they are presented to caregivers.
Finally, the project will examine efficient on-device inference strategies for ambient sensing pipelines, focusing on reducing computational cost while ensuring that sensitive behavioural data is processed locally wherever possible.
The expected outcome is a prototype framework that integrates privacy-aware learning, reliability-aware alert generation, and efficient edge-based inference for behavioural sensing applications. The project aims to contribute practical insights into building deployable intelligent systems for real-world assisted living environments.
Research area, student roles & skills
Research area: My research spans data privacy, cloud computing, cybersecurity, social network analysis, and AI/machine learning. I have addressed challenges in safeguarding personal data in cloud environments and developed secure, reliable systems. I proposed a cloud solution incorporating an entropy-based trust computation approach for data-as-a-service environments. Additionally, I have worked on heterogeneous information networks, focusing on healthcare data privacy, where I developed a differentially private algorithm for releasing de-identified health data with reduced information loss and enhanced scalability. My work also includes contributions to user credibility systems in social networks, blockchain-based solutions, as well as supply chain cybersecurity risk analysis.
Student roles: The student will contribute to the design and implementation of a machine learning framework for privacy-preserving and trustworthy analysis of longitudinal behavioural data in ambient sensing environments. The work will begin with a structured literature review on sequential behavioural modeling, privacy-preserving machine learning, and trust-aware decision systems. This will help establish a foundation for understanding current methods and identifying practical challenges in real-world deployment scenarios. The student will then implement baseline machine learning models for behavioural pattern analysis using sequential data. This includes preprocessing time-series inputs, training predictive models, and evaluating performance using standard metrics such as precision, recall, and F1-score. A core component of the project will involve exploring privacy-preserving learning techniques adapted for temporally correlated data. The student will study how such methods affect model performance and behavioural prediction quality, and will conduct comparative experiments to understand the trade-offs between privacy and utility. In parallel, the student will develop a lightweight reliability-aware alerting component. This module will incorporate model confidence scores and short-term behavioural consistency checks to reduce unnecessary or low-confidence alerts, improving the practical usefulness of the system for caregiver-facing applications. Finally, the student will explore efficient inference strategies suitable for running models locally, with a focus on reducing computational overhead and ensuring that sensitive data is processed within the system pipeline whenever possible. Throughout the project, the student will document experiments, analyze results, and contribute to technical reporting and potential dissemination. The role is designed to provide hands-on experience in applied machine learning, privacy-aware system design, and real-world AI deployment considerations.
Skills required: - Strong foundation in machine learning and basic deep learning concepts - Proficiency in Python and familiarity with ML frameworks such as PyTorch or TensorFlow - Basic understanding of probability, statistics, and time-series data analysis - Interest in privacy-preserving machine learning and data security concepts - Experience with sensor data or sequential data modeling is an asset - Ability to preprocess data, implement models, and evaluate results using standard metrics - Strong analytical thinking and willingness to learn applied AI system design - Good communication and technical writing skills
610. Proactive Dialogue Agents for Subjective Tasks (Tutoring, Emotional Support) and Goal-Oriented Multi-turn Interactions
This project focuses on improving proactive dialogue agents in subjective tasks such as tutoring, emotional support, or negotiation: settings where the goal is not to retrieve facts, but to guide users through complex, often unclear situations. Unlike traditional chatbots that simply respond to prompts, proactive agents must take initiative, ask meaningful questions, and steer conversations toward helpful outcomes. These abilities are essential in domains where understanding is subtle, yet they remain underdeveloped in current AI systems.
A major limitation lies in how these agents are trained. Many systems rely on self-play simulations between LLMs, but the resulting dialogues often lack variability - they repeat similar surface patterns - and depth, failing to explore realistic challenges or user reactions. For example, a simulated student might always ask straightforward questions, missing the hesitation or confusion real learners express. This makes it hard for agents to generalize to diverse, real-world users. This project will improve simulation quality by increasing behavioral diversity and role consistency.
A second challenge is that agents often assume they fully understand the user. But in subjective tasks, meanings are rarely clear-cut. Grounding-aware dialogue policies will be developed so agents can actively check and refine shared understanding - by asking clarifying questions, confirming goals, or adjusting when misunderstandings arise. For instance, if a user says, “I feel stuck,” the agent might ask, “Do you mean professionally, personally, or both?”
We also aim to move beyond single-turn evaluation. Instead, we will define goal-based metrics that assess whether the agent helps users reach meaningful outcomes over multiple turns.
Potentially, the project will explore reinforcement learning from AI feedback, using both verbal suggestions and reward signals to train more effective, adaptive agents.
Together, these contributions aim to build AI systems that are strategic, responsive, and robust in complex human conversations.
Research area, student roles & skills
Research area: My specialized research area is Natural Language Processing (NLP). NLP is one of the most important and useful fields of artificial intelligence, dealing with automatic text processing. Its widespread application is due to the ubiquity of language and text as a main human communication channel, e.g., emails, blogs, social media, web search, chatbots, medical reports, etc. The objective of my research is to devise new methods for machine language understanding, which include language representation, summarization, and classification from unstructured and unconventional text streams, mainly in an unsupervised fashion.
Student roles: • Explore and synthesize research papers on proactive conversational agents and subjective dialogue tasks (e.g., tutoring, emotional support, negotiation) from top NLP and AI conferences. Develop a critical understanding of challenges like grounding, planning, and long-term goal evaluation. • Conduct reviews of self-play simulation techniques and grounding-aware dialogue strategies in multi-turn LLM conversations. • Get familiar with existing datasets such as ESConv, CIMA, and CraigslistBargain. Preprocess these datasets as needed for training and evaluation purposes. • Investigate and compare existing evaluation protocols (turn-level vs. goal-based) and implement custom metrics to assess the success and efficiency of proactive agents. • Develop a small, trainable model that selects high-level dialogue actions for the LLM during multi-turn conversations. Its goal is to help the LLM act strategically in subjective tasks like tutoring or emotional support. The model will be fine-tuned first on human-annotated data, then further improved through reinforcement learning based on feedback from simulated conversations. • Implement and test grounding-aware dialogue policies that enable clarification and intent-alignment behaviors during multi-turn interactions. • Conduct controlled experiments with various prompting, planning, and reward strategies to improve adaptability and conversational depth. • Regularly report progress to the supervisor through short presentations and written updates. • Collaborate with other researchers or students working on adjacent projects (e.g., dialogue modeling, evaluation, responsible AI) to become familiar with scientific practices in NLP research.
Skills required: Expertise in any of these fields: Computer Science, Computer Engineering, Information Science and Technologies. Prior programming experience in Python. Experience with Large Language Models prompting and Fine-tunning. Experience with PyTorch. Knowledge of basic multivariate calculus, linear algebra, probability, and statistics. Familiarity with the essentials of data mining, machine learning, or artificial intelligence. Being previously introduced or worked on a project involving NLP.
Compilers are complex, fundamental system software. Debugging compilers has long been recognized as a time-consuming and painstaking task due to the complexities within their algorithms and the bug-triggering input programs (Inputs to language implementations are also programs). Consequently, compiler developers have extensively adopted program reduction techniques to streamline the debugging process. Specifically, when given a bug-triggering program, a program reducer can autonomously produce a minimized program that still triggers the same bug but with bug-irrelevant code elements removed.
Existing program reduction algorithms exhibit shortcomings in efficiency (reduction time), effectiveness (size of reduced programs), or both. For example, language-agnostic program reducers work for a variety of languages but can be slow and produce large results (e.g., Delta Debugging, Hierarchical Delta Debugging, and Perses). On the other hand, language-specific program reducers are tailored and optimized to reduce programs in a particular language, often resulting in smaller outcomes at the cost of reduced generalizability.
This project aims to develop a novel, language-agnostic approach to address existing issues concerning effective, efficient, and general program reduction. Specifically, the project has three main components. First, this proposal seeks to exploit the formal syntax of the program under reduction in a comprehensive manner, to expand the boundaries of syntax-guided, language-agnostic program reduction. Second, the project aims to harness the capabilities of Large Language Models (LLMs) for language-specific program reduction. Last, this project will introduce a standardized evaluation procedure and a benchmark suite comprising multilingual programs. These resources will enable a comprehensive assessment of program reduction techniques from various perspectives.
The project will benefit various development communities of compilers. Due to the limited time of the internship, the internship project will be a part of the whole project. More information about the project can be found at https://github.com/uw-pluverse/perses.
Research area, student roles & skills
Research area: My research interests span software engineering and programming languages. I focus on how to improve software reliability and developers’ productivity by designing and developing novel, practical techniques, models and tools. My research philosophy values practicability. I choose real-world, important problems to work on, and aim for novel solutions. I have been working on validating critical system software for over eight years. My work has been highly impactful (e.g., discovering over 1600+ confirmed bugs in real-world production compilers such as GCC and LLVM), and has been published at top-tier prestigious international conferences such as PLDI, OOPSLA and ICSE.
Student roles: The student will work closely with me and collaborate with one or two of my Ph.D. students who also work on this project. There will be three stages of this project from the student’s viewpoint. At the beginning, the student will be doing literature survey to familiarize him/her with the related work. Guided by the PI, the student will also start to study and work on the existing code base. This stage will only take two or three weeks, and the student is expected to be prepared for the main proposed work both theoretically and technically.
Then the student will start implementing the proposed work for around eight weeks. During this time, the student is expected to work closed with the PI and his students. We will adopt the regular software development process, which includes version control, pull requests and code reviews. We value this project, and expect the code quality of the deliverables to be high-quality and easy to maintain.
In the last week or two, the student is expected to wrap up the work, write a technical report on the project. We hope that the technical report can be published, or at least can be a part of a publication.
Throughout the internship, the student will have weekly meetings with the PI, in which, the student can discuss the progress, research ideas and technical challenges. The PI loves programming, and values code hygiene. One of his example project can be found at https://github.com/perses-project/perses.
Skills required: The student is expected to self-motivated and have solid understanding of fundamental concepts of computer science. The student should be interested in programming languages and software engineering. Specifically,
First, the student should have taken a course on compilers, and know basic concepts of compiler internals, e.g., lexing, parsing, abstract syntax tree, control-flow graph, compiler optimizations.
Second, due to the nature of this project to validate language tools, the student should be good at programming, and is expected to be proficient with one prevalent programming language, for example, Kotlin, Java, C, C++, Rust.
612. Programmable Pattern Matching for Proof Assistants
Pattern matching provides a convenient way to write functional programs that exhaustively handle all inputs. In dependently typed programming languages, pattern matching corresponds to the "proof by cases" technique, allowing a programmer to write a function by specifying the output for each possible shape the input might have. However, pattern matching in dependently typed languages is often quite limited: the cases must directly correspond to the constructors of a datatype, and support for overlapping or catch-all cases tends to be poor.
Recent work has developed a theory of pattern matching based on coverage semantics, where the exact set of patterns which is allowed is treated as a parameter of the language. Sheaf theory is used to provide the criteria for which patterns lead to a well-defined semantics for matching. Coverage semantics capture the status quo for pattern matching, but also enable more exotic forms of matching, such as matching on arbitrary isomorphisms.
This project aims to develop a prototype of "pattern matching as a library," based on an internal variant of coverage semantics. In a proof assistant such as Agda, Idris, or Lean, we will develop a library for pattern matching which is parameterized over the set of allowed patterns. This will provide a testbed for novel pattern-matching variants, without requiring major changes to the implementation of a proof assistant. To use the library, the programmer initially provides a predicate defining allowed pattern-sets, along with a proof that this meets the sheaf-theoretic criterion for leading to well defined semantics. The function for pattern matching will take a list of patterns, a result for each pattern, and a proof that the set of patterns satisfy the earlier provided predicate.
Research area, student roles & skills
Research area: Broad research area: Programming Languages, Type Theory
Dependently typed programming languages allow developers to write programs that are mathematically guaranteed to meet their specifications. If a dependently typed program compiles, it must be bug free. Such languages have been used to great success in the mechanization of mathematical proofs, but have yet to see widespread adoption by programmers, in part due to their difficulty to use. This research develops theoretical and practical tools which make dependently typed languages easier to use. This in turn makes it easier to write programs which are guaranteed to be correct.
Student roles: The main duty of the student will be writing dependently typed implementations of existing algorithms from the literature. The project will begin with a period of background reading and learning to use a proof assistant. Depending on the student's background and preference, Agda, Idris, or Lean can be chosen. The student will work with the supervisor to learn just enough category theory and sheaf theory to understand the existing theoretical work.
Once the initial learning phase is complete, the student will work to mechanize the basic definitions of sheaf theory: presheaves, matching families, amalgamations, covers, and sheaves. Then, by implementing these in a constructive type theory, obtaining a procedure for pattern matching is straightforward. The student can explore the metaprogramming facilities and macro system of the proof assistant they are working with to make the library more usable, as well as using proof automation or typeclasses to generate the validity proofs for patterns.
Skills required: The student should have experience with strongly-typed functional programming (e.g. Haskell/ML/Elm/Scala), including features like algebraic datatypes, lambdas, higher-order functions, pattern matching, and monads. Also, the student should have some familiarity with logic and mathematical proofs, especially induction.
Students are not required to have experience with proof assistants (such as Rocq/Coq, Agda, Idris, Lean, Isabelle), but any experience working with proof assistants, type theory, dependent types, or formal logic would be beneficial. Likewise, knowledge of abstract mathematics (abstract algebra, set theory, category theory, topology, etc.) is beneficial, but is not required.
613. Project Title EMI-RAPL: Trustworthy Mechatronics and Robotics Intelligence through Reasoning Agentic LLMs and Physics-Informed Learning
This project develops EMI-RAPL, a principled unification of agentic LLMs, graph-based reinforcement learning, and physics-based modeling for trustworthy intelligence in mechatronics, robotics, and manufacturing. Rather than treating data-driven AI and physics-based methods as competing approaches, EMI-RAPL establishes a symbiotic feedback loop: agentic systems reason over structured abstractions while physics models enforce feasibility, safety, and interpretability under uncertainty and limited data. The research is structured around two primary thrusts. The first thrust focuses on physics-grounded intelligent modeling and process control, addressing real-world challenges in WAAM process parameter identification using an integrated Sinusoidal CLS-Taguchi framework, adaptive PID control for multi-zone industrial furnace systems using neural networks and differential evolution, and adaptive extended Kalman filtering for heterogeneous multi-rate sensor fusion. These components embed physical laws as inductive bias during training and runtime safety layers for agentic controllers. The second thrust focuses on agentic knowledge-driven optimization and robotic planning. This includes a generalized three-stage RLHF framework applicable to peg-in-hole robotic assembly, bearing remaining useful life prediction, drilling chatter detection, and welding sequence optimization. A hybrid genetic algorithm with Jacobian refinement is developed for robot inverse kinematics across serial manipulators with four or more degrees of freedom. Both thrusts are validated on physical testbeds including Motoman MH-6 and SDA-20 robots, dual-tank fluid systems, and multi-zone industrial furnaces. The project will produce open-source simulation tools OS-WeldSim and OpenAM-SimCCX to democratize high-fidelity finite element analysis for academic institutions and SMEs. Results will be disseminated through peer-reviewed publications in venues including ECML, Pattern Recognition, and IEEE journals, with technology transfer through industrial workshops and the Manufai manufacturing startup, which has already deployed doctoral research on welding sequence optimization in commercial software.
Research area, student roles & skills
Research area: This research area advances intelligent mechatronic and robotic systems by integrating agentic large language models, graph-based reinforcement learning, and physics-informed modeling. The program spans five application pillars: physics-grounded process control for wire arc additive manufacturing (WAAM) and multi-zone furnace systems; knowledge-driven robotic planning including hybrid genetic algorithms for inverse kinematics and RLHF-based peg-in-hole assembly; intelligent vision-based quality inspection for Industry 4.0; mechatronic design and human-robot collaboration using FEM-guided adaptive grippers; and open-source physics-based simulation democratizing high-fidelity finite element analysis for academic institutions and small manufacturers worldwide.
Student roles: The student will contribute directly to the implementation and validation of the EMI-RAPL framework across selected mechatronics and robotics application areas. Their core technical responsibilities will include assisting in the development of the three-stage RLHF pipeline, including supervised fine-tuning, reward model construction, and proximal policy optimization, applied to one or more target tasks such as peg-in-hole robotic assembly, bearing RUL prediction, or welding sequence optimization. The student will implement domain-specific reward functions that encode physical feasibility, safety constraints, and expert preferences, and will run controlled experiments comparing agentic versus non-agentic baselines. In the physics-grounded control thrust, the student will assist in implementing and validating the adaptive PID framework for multi-zone furnace systems, including neural network training, differential evolution hyperparameter tuning, and performance benchmarking against fixed-gain controllers. The student will also contribute to the adaptive EKF sensor fusion module by implementing asynchronous multi-rate data handling and EWMA noise adaptation, validating results on the dual-tank testbed with ultrasonic and camera sensors. Additional responsibilities include supporting the hybrid genetic algorithm for robot inverse kinematics, implementing warm-start initialization strategies, Hamming-distance diversity mechanisms, and Jacobian refinement steps, and evaluating solution accuracy across manipulator configurations. The student will also assist in the preparation and documentation of open-source simulation tool components. Beyond technical work, the student will engage in knowledge mobilization by assisting in manuscript preparation, creating presentations for academic and industrial audiences, and documenting all experimental protocols and results. These activities will develop the student's scientific communication and professional writing skills. Through this project the student will gain expertise in agentic RL, physics-informed modeling, sensor fusion, and robot control, alongside essential soft skills in collaboration, project management, and technical communication, preparing them for careers in industrial AI, robotics, or advanced manufacturing.
Skills required: The student should have foundational knowledge in mechanical or electrical engineering, robotics, or computer science, with basic programming experience in Python. Familiarity with control systems, kinematics, or machine learning concepts is an asset. Prior exposure to reinforcement learning, sensor fusion, or finite element analysis is beneficial but not required. The student should be motivated to work at the intersection of AI and physical engineering systems, comfortable with mathematical modeling, and eager to learn agentic AI frameworks. Students from mechatronics, electrical engineering, mechanical engineering, or related disciplines with interest in Industry 4.0 and smart manufacturing are encouraged.
614. Project Title Participatory Design of Accessible Software Interfaces for Older Adults and Users with Cognitive Disabilities
Supervisor: Sanaa Alwidian
University: Ontario Tech University (Oshawa campus)
Despite the existence of accessibility standards such as WCAG 2.2 and Canada's Accessibility Act, a significant proportion of software interfaces remain inaccessible to older adults and users with cognitive disabilities such as mild cognitive impairment and acquired brain injury. This gap reflects not only a compliance shortfall but a deeper methodological problem, where users with cognitive accessibility needs are rarely involved as co-designers in the software engineering process, and existing guidelines do not adequately translate into actionable design practices for software development teams.
This project addresses that gap through a participatory design study conducted in partnership with a local community organization serving older adults and individuals with cognitive disabilities. Rather than treating users as test subjects after design decisions are made, this project positions them as active co-designers from the earliest stages of requirements elicitation through iterative prototyping. Over 12 weeks, the intern will facilitate three cycles of co-design workshops with 12–18 participants, translating findings into interactive prototypes of two target applications, namely, a health information portal and a community services directory. Prototypes will be evaluated through structured usability sessions measuring task success rate, error frequency, and user satisfaction.
The central contribution of this project is a validated pattern language of 10–15 cognitive accessibility design patterns, grounded in empirical co-design and usability data, and published as an open-access repository for use by software engineering teams across industry and government. Findings will also be prepared for submission to a leading accessibility or HCI venue such as ASSETS or CHI.
This work directly supports Canada's aging population and its legislative commitments to inclusive digital access, while advancing the methodological integration of participatory design into mainstream software engineering practice.
Research area, student roles & skills
Research area: I am the director of the ARiSE research lab (https://arise-lab.ca/). My research focuses on human-centric requirements and software engineering, with an emphasis on developing methodologies and frameworks that place human needs, values, and experiences at the center of the software development process. I investigate how requirements are elicited, represented, and validated in complex socio-technical contexts, and how software systems can be designed to better serve diverse user populations, including those with accessibility needs. My work spans empirical software engineering, participatory design, and AI-enabled requirements engineering, contributing to the development of more reliable, inclusive, and user-responsive software systems.
Student roles: The intern will play a central and hands-on role in all phases of this research project, functioning as a junior researcher embedded within the ARiSE Lab at Ontario Tech University. Rather than supporting pre-existing work, the intern will co-own the execution of the project from inception to dissemination, developing both technical and research competencies throughout. In the early weeks, the intern will conduct a focused review of cognitive accessibility literature and existing design guidelines, synthesizing gaps that motivate the study. They will assist in finalizing the ethics protocol, preparing accessible workshop materials, and establishing contact with the community partner organization. During the co-design workshop phase, the intern will take a lead facilitation role, guiding participants through sensitizing, generative, and evaluative design activities. This requires sensitivity, adaptability, and clear communication when working with older adults and individuals with cognitive disabilities. The intern will document sessions through notes, photos, and recordings, and conduct thematic analysis of findings between workshop cycles. Based on workshop insights, the intern will design and develop interactive prototypes using Figma and web technologies (HTML/CSS/JavaScript), iteratively refining them in response to participant feedback. They will then plan and conduct structured usability evaluation sessions, administer standardized questionnaires, and analyze both quantitative and qualitative data. In the final phase, the intern will synthesize all findings into a pattern language of cognitive accessibility design patterns and contribute to building an open-access web repository documenting each pattern with rationale, examples, and evidence. They will also co-author a draft research manuscript targeting a leading accessibility or HCI venue. Throughout the internship, the intern will participate in weekly lab meetings, present progress updates, and deliver a final oral presentation to the research group.
Skills required: Applicants should be in their 3rd or 4th year of an undergraduate degree in Computer Science, Software Engineering, or Human-Computer Interaction. Required skills include basic programming (HTML/CSS/JavaScript or mobile development) and a genuine interest in user research and accessibility. Familiarity with usability evaluation methods, wireframing tools such as Figma, or accessibility standards (WCAG) is an asset. Strong interpersonal and communication skills are essential, as the project involves direct engagement with older adults and individuals with cognitive disabilities through co-design workshops. Experience with qualitative research methods or community-based research is considered a strong advantage.
615. Protecting Confidential Cloud Applications from Memory Side-Channel Attacks
Supervisor: Aastha Mehta
University: University of British Columbia (Vancouver campus)
Cloud providers increasingly offer Confidential Virtual Machines (CVMs), which protect applications and their data even if the underlying cloud software is compromised. These technologies make it possible to run sensitive workloads, such as AI services, healthcare applications, and financial systems, on untrusted cloud infrastructure.
However, protecting data is not enough. Even when data remains encrypted, attackers may still learn sensitive information by observing how an application uses system resources such as memory. These so-called side-channel attacks can reveal information about program behavior, user inputs, or confidential data.
This project investigates practical techniques for reducing such information leakage. In particular, the project explores user-level paging, a mechanism that allows applications to control how memory is managed without requiring changes to the operating system. By carefully controlling memory access patterns, user-level paging may help hide sensitive program behavior while remaining compatible with existing applications.
The goal of the project is to design and evaluate a practical defense against memory side-channel attacks in Confidential Virtual Machines. The student will develop a prototype system, measure its effectiveness and performance overhead, and explore how user-level memory management can improve the security of modern cloud applications.
Research area, student roles & skills
Research area: I am broadly interested in systems security, data privacy, operating systems, and distributed systems. My research involves building systems that address security and privacy challenges by design, and evaluating the security and performance properties of the system empirically. My current research directions include designing data security mechanisms in serverless and edge cloud platforms, and mitigations for software and hardware side channels in cloud applications. I am looking for students with strong software engineering skills. Experience with operating systems programming is desired.
Student roles: The student will meet the project group weekly and produce weekly reports to track the research progress. The student will participate in these four steps:
First, the student will implement a prototype for memory access obfuscation on well-known interfaces like Linux's userfaultfd, using obfuscation techniques like Oblivious RAM (ORAM).
Second, the student will apply the prototype of the system to common cloud workloads, such as web servers and machine learning.
Third, the student will measure the system's performance impacts. This requires the student to set up a stable evaluation environment and write reusable evaluation scripts.
Finally, based on the performance evaluation, the student will explore strategies to reduce the overhead, which is expected to be significant. At the mechanism level, this involves optimizing the user-level paging and randomization mechanisms. At the policy level, this involves investigating principled trade-offs between security and performance through algorithmic design choices.
Skills required: The student must have completed an undergraduate-level course on operating systems and computer architecture by the time of the internship. The student must also be familiar with C/C++ programming. Experience in system programming and CTF challenges is highly desirable.
616. Provable AI results through formalization of subatomic physics in Lean
Supervisor: Wouter Deconinck
University: University of Manitoba (Winnipeg campus)
The Canadian subatomic physics community, through the EIC Canada Collaboration led by UM, holds leadership roles in the EIC program and in the international ePIC collaboration (with 850+ members) which will build the first detector at the EIC: the ePIC detector, budgeted at a total cost of US$300M. In particular, the UM group leads various software and computing efforts in the ePIC collaboration. These includes simulation and reconstruction software, AI knowledge platform development, and data production campaign operations.
As part of this project, the student will expand the solid mathematically rigorous formal foundation of subatomic physics concepts at the EIC by encoding derivations and proofs in a functional programming language called Lean, potentially using AI code generation assistants as a first-class coding partner (but with full understanding of the proofs). Having the fundamental concepts of the field of subatomic physics encoded in Lean allows us to insist on formal proofs for any statements that an AI code generator emits, such as new methods to analyzer data collected at the EIC and connect it to the underlying physical theories. In particular, we aim to apply this basis in Lean to a mathematical description of the unfolding techniques used to correct for bin migration effects in observed quantities with uncertainties.
This project will require the student to have a keen understanding of the mathematics underlying the physics at the EIC, to be able to assess correctness of mathematical derivations, and to have a familiarity with type hierarchies in programming languages.
Research area, student roles & skills
Research area: The Electron-Ion Collider is a US$3B new particle collider facility to be built in the current decade at Brookhaven National Lab, Long Island, New York, by the US Department of Energy. At the EIC, for the first time, polarized electrons will collide with polarized protons and light ions, and unpolarized heavy nuclei at intensities a thousand times higher than at previous (unpolarized) colliders. The EIC is at the center of a global 1400+ scientist effort to answer questions about the origin of mass and spin of the proton, and the dynamics of dense gluon systems.
Student roles: The student will work in a highly dynamic subatomic physics collaboration, with global participants in all major world regions. Alongside a global team of physics and mathematics experts, the student will formalize the proofs that are the underpinnings of quantum field theory, quantum chromodynamics, deep-inelastic scattering, and the experimental extraction of knowledge from observables.
Skills required: - strong understanding of subatomic physics concepts and their underlying mathematical foundation, - strong collaborative work ethic, supported by portfolio of git activity, - C++ and/or Python experience, - experience in a functional programming language (such as Lean).
617. Quadruped Robot Perception and Control
Supervisor: Geoff Fink
University: Thompson Rivers University (Kamloops campus)
This research project aims to improve the autonomy of quadruped robots by developing novel perception and control algorithms. More specifically, we will enhance the robot’s position and velocity estimates within dynamic environments. We will do this with two main approaches. The first approach is to improve mapping and visual-inertial sensor fusion by including robot and environmental dynamics in simultaneous localization and mapping (SLAM). The second approach uses a team consisting of a ground-based robot (quadruped) and an aerial robot (quadrotor) to perform coordinated state estimation.
Research area, student roles & skills
Research area: Quadruped robots are increasingly used for a wide range of both indoor and outdoor applications. They have remarkable agile capabilities; however, most are still remotely piloted because they lack a high level of self-awareness. Quadruped robots must be robust while traversing a wide variety of new terrain. The terrain may vary in geometry and its physical properties, such as terrain impedance or friction. Reliable state estimation is a significant aspect for the deployment of quadrupeds because most locomotion planners and control strategies rely on an accurate estimate of the pose and velocity of the robot.
Student roles: The student will work as part of an interdisciplinary team on perception and control algorithms for quadruped robots. This includes analysing and implementing computer vision algorithms, implementing control algorithms, improving the quadruped software framework, simulating quadruped robots, running experiments with quadruped robots, and writing reports.
Skills required: This is a large project where students can focus on any one or more of the following areas: control, software engineering, robotics, simulation, computer vision, or computer/electronic engineering.
618. Quantum Algorithms for Visual Analytics of Large Graphs
Supervisor: Debajyoti Mondal
University: University of Saskatchewan (Saskatoon campus)
Our lives today are deeply influenced by a large variety of networks. Activities such as online shopping, medical check-ups, interaction with mobile apps, social websites, and phone calls are a few examples that produce vast quantities of network data. Network summarization is a powerful technique for discovering hidden patterns, correlations, and anomalies in such networks, which helps users to make better decisions. A number of systems to analyze networks with millions of entities have already been developed, but given today's massive datasets, one needs to develop even faster algorithms.
In this project, we aim at overcoming this difficulty by leveraging quantum algorithms to process graphs by leveraging quantum-powered search and data structures. The goal is to transform the traditional processing into modules that can be gradually replaced with faster quantum processing. We will combine Grover's search methods and graph data structures to design such network analytics techniques. Quantum processing of networks has a huge potential to improve our capability of handling big data. Thus this project is expected to inspire many new directions for future research in the quantum powered data science area.
Research area, student roles & skills
Research area: Debajyoti Mondal is an associate professor in the Department of Computer Science at the University of Saskatchewan, Canada. His primary research interest is in visualization algorithms, graph drawing, visual analytics of large networks. His work spans interactive network layouts, cartographic representations, geometric representations, metro maps, etc. A major part of the visualization process is to map the abstract information to some geometric data for rendering, which inspires his research in computational geometry and graph algorithms.
Student roles: The student will work closely with his advisor. The initial phase of the project is learning intensive, where the student will learn the existing streaming algorithms, tools, and modeling techniques. The student will also independently review papers on quantum algorithms and graph visualizations from premier venues. The next phase would be a development phase, where the student will develop and implement algorithms. These algorithms will be integrated into a visualization system. In the final phase, the students will conduct a user evaluation to the evaluate performance of the developed system. The student will regularly meet the advisor to discuss the progress and challenges.
In this project, the student will gain unique skills in developing quantum powered graph algorithms and systems. The software library developed by the student will be a great asset for building future software. The student will be able to develop unique skills to visualize large datasets, which will place the student in a favorable position in the job market. Specifically, the student will gain expertise in the areas of data processing techniques, user interactions, and smart visualization systems.
Being in a research environment, the student will also be familiar with the research contacted by the graduate students of the Visualization Lab (Saskatchewan) and Human-Computer Interaction Lab (Saskatchewan), which may encourage the students to plan a graduate career in this area.
Skills required: Strong knowledge of theoretical computer science, data structure, algorithms, proof techniques, and discrete mathematics. Good knowledge of computational geometry or graph theory (graph terminologies and properties) and any knowledge of quantum computing would be an asset.
619. Quantum Computer Vision
Supervisor: Li Cheng
University: University of Alberta (Edmonton campus)
This project focuses on analyzing and interpreting computer vision tasks from quantum computing perspectives. Its aim is to explore the integration of quantum computing principles with computer vision and applications. As a member of the Vision and Learning Lab at the University of Alberta (https://vision-and-learning-lab-ualberta.github.io/), you are expected to work with a graduate student/Postdoc researcher, get familiar with state-of-the-art quantum computer vision techniques, and gain hands-on research experience on benchmark datasets. There are opportunities to be involved in exciting research topics in computer vision and to publish research findings at top-tier conferences and journals.
Research area, student roles & skills
Research area: Computer vision and applications
Student roles: You are expected to work with a graduate student/Postdoc researcher, get familiar with state-of-the-art quantum vision techniques, and gain hands-on research experience on benchmark datasets. There are ample opportunities to be involved in exciting research topics in computer vision and to publish research findings at top-tier conferences and journals.
Skills required: Necessary computer vision and image processing background and working knowledge of quantum computing, calculus and linear algebra; Good programming skill in python.
620. Quantum Computing for Aerial Robots Tracking and Motion Planning
Supervisor: Farrokh Janabi-Sharifi
University: Toronto Metropolitan University
Location: Toronto, Ontario
Start date: 2027-05-31 (flexible)
Disciplines: Computer Science, Engg-Electrical, Engg-Computer, Engg-Mechanical, Engg-Systems and Technology
Objectives:
- Developing QC-based motion planning algorithm for drones with aerial manipulator operating in cluttered environments.
- Validating the method in simulation environment such as Gazebo or Pybullet and QC-simulator.
- Benchmarking the results against the conventional motion planning approaches.
- Implementation of the results on actual quantum computers accessible through partner organizations.
Expected Outcomes:
1. An efficient QC-based motion planning algorithm for aerial robots in highly cluttered environments.
2. A simulation setup for aerial robot that could be used to test the method(s).
3. Comparative analysis of the proposed QC-based approach with traditional variants of motion planning algorithms for drones/aerial manipulators.
4. Insights into the feasibility and limitations of implementing the method on quantum computers with limited computational resources.
5. Quantum hardware implementation and real-world experiments.
Research area, student roles & skills
Research area: Despite recent advances in computing devices and perception technology, development of a robust and effective motion plan remains as a pending key issue for many autonomous aerial robot (drone) applications. The general problem is often addressed in two levels of global and local planning which can be posed as a constrained optimization problem that is computationally expensive to solve with classical methods (due to its NP-completeness nature). The aim of the project is to develop and verify a robust QC-based motion planning algorithms for aerial robots, particularly for aerial manipulators. We have already some on-going work in this regard.
Student roles: The student will start by getting familiar with the literature of motion planning and quantum-based planning. The intern will then work to develop a simulation platform. In addition to developing the method, the student will also get in contact with quantum computing partners for further training and assistance with the simulation and implementation of the quantum-based method. Beside working with the industry, the intern will work with the graduate students and postdoctoral fellows of t he lab. The intern will meet the supervisor on a weekly basis and report to them. Also the intern will make presentation to the group at least once a month. In addition to quantum simulation of the method, the intern will work with quantum industry to test the algorithm in real quantum computers. He/she will document the method, code, simulation and experiments.
Skills required: - Previous research experience - Familiarity with robots’/drones’ kinematics and dynamics - Previous experience with motion planning, optimization, and quantum computing - Familiarity with robot simulation systems and implementation on the hardware (C or C++ or Python) - Previous experience in ROS/ROS2 is an asset - Good communication skills - Previous experience in experiments and data processing
621. Quantum Computing for Bioengineering Applications
The aim of this exploratory project is to use quantum computing to reformulate some of the computationally intensive combinatorial design problems in bioengineering including 1) design of new pathways to novel chemicals, 2) design of new mutants with improved activity for enzymes, 3) design of genome modifications for genome editing using metabolic models. The successful application to even one of these problems can lead to enormous impact. For the first problem, we have published a paper that uses simulated annealing to solve this problem but the algorithm still takes too long to explore the diversity of the chemical space. In the second problem, while the field of machine learning has been applied to better explore the space of protein sequences, there is still need to confirm the activity of the designed sequence with molecular modeling techniques including techniques such as DFT that are accelerated via quantum computing. For the last problem, we have mixed integer linear optimization problem that can lead to identification but is not yet scalable to microbial communities and larger networks such as the human metabolic network and hence the application of accelerated optimization tools from Quantum computing can allow the exploration of genome editing for larger models. The students will have a choice of working on any of these projects and will work closely with experimentalists in all these areas so that the computational experts can interact with wet lab experts and identify new designs.
Research area, student roles & skills
Research area: Recent advances in machine learning has led to many bioengineering tools including detailed molecular modeling. Our group develops computational methods for engineering enzymes, pathways, microbial strains and communities for applications in environment, medicine and industrial biotechnology. Our methods combine both deterministic modeling along with machine learning based applications to develop a comprehensive bioengineering platform. We also combine these computational methods with experimental techniques that allow the validation of these methods. Recently, our group demonstrated that augmenting sequence information with other features lead to improved metabolic network models and we used these models to engineer metabolism for practical applications.
Student roles: The student will develop new quantum computing workflows, will prepare new algorithms using QC, and test new workflows on the three design problems. The student will also identify deficiencies in the data sets that prevent training, evaluate the importance of acceleration from QC compared to classical techniques and once the advantage is shown use the QC codes to solve much larger design problems and identify more biologically relevant solutions. In addition, the student will reproduce existing algorithms in the literature to establish a baseline against which the newly developed algorithms can be compared. Student will develop metrics that measure the successful acceleration and designs and identify new experimental designs that can enable improved prediction and models. Finally the student will also closely interact with experimentals to generate targeted data that can enable the improved performance of these methods. Student will also write a report documenting the research and contribute to a research article summarizing the results of the research. The student will also present their research at weekly group meetings and at local , national and international conferences and interact with collaborators from other Universities.
Skills required: The ideal will student will have strong background in industrial engineering, computational biology, biochemistry, quantum computing, physics and machine learning. The undergraduate program can be in computer science or engineering or physics and the student will have experience with using quantum computers for optimization, bioinformatics and genome sequence analysis. The student must be willing to learning techniques and work with senior researchers in order to execute new ideas and algorithms into code and test them out. Finally, the student should be comfortable programming in Python and familiarity with Pytorch and developing code for GPUs.
622. Quantum Computing for Hyperspectral Image Classification
Supervisor: Saad Ahmed
University: Lakehead University (Thunder Bay campus)
The proposed project on Quantum inspired Hyperspectral Image Classification aims to develop a lightweight hybrid framework that combines classical deep learning with quantum machine learning techniques to analyze high-dimensional hyperspectral data. By using variational quantum circuits implemented through platforms such as PennyLane, the project focuses on encoding spectral information into quantum feature spaces that can capture complex patterns more efficiently than traditional approaches. The model will be evaluated on benchmark datasets like Indian Pines and Pavia University dataset, with particular attention to performance under limited training data conditions. The expected outcome is a parameter-efficient and interpretable model that highlights the potential of quantum-enhanced learning for applications in remote sensing, agriculture, and environmental monitoring.
Research area, student roles & skills
Research area: My research focuses on using Artificial Intelligence and Machine Learning to solve real-world problems involving complex image data, especially hyperspectral images. I work on building models that can understand both the spatial and spectral information in data, using techniques like deep learning and transformer-based approaches. A key part of my work is making these models easy to understand and trustworthy, so their decisions can be explained clearly. I also explore new ideas such as combining classical AI with quantum tools like PennyLane. My research is applied in areas such as agriculture (crop monitoring), healthcare (medical image analysis), and document forensics.
Student roles: The student will work as an active research team member and contribute to the development of a hybrid quantum-classical model for hyperspectral image classification. Their role will include reviewing relevant literature, preparing and preprocessing datasets, and implementing machine learning models using Python-based frameworks. The student will also run experiments, analyze results, and help improve model performance. In addition, they will document their work, participate in regular research meetings, and assist in preparing reports or a potential publication. The role emphasizes hands-on learning, problem-solving, and collaboration within the research lab.
Skills required: The ideal student should have a background in Computer Science, Data Science, or a related field, with a basic understanding of machine learning and deep learning concepts. Familiarity with programming in Python and experience using libraries such as TensorFlow or PyTorch will be helpful. Some exposure to image processing or working with datasets is beneficial, though not mandatory. An interest in learning advanced topics like hyperspectral imaging, explainable AI, or quantum-inspired methods (e.g., using tools like PennyLane) would be an advantage. Most importantly, the student should be motivated, willing to learn new concepts quickly, and comfortable working independently as well
623. Quantum Computing for enhanced Remote Sensing Data Analysis
Remote sensing is the science of acquiring information about the Earth’s surface from a distance, typically using satellite or airborne sensors. It provides valuable data for various applications like environmental monitoring, land use mapping and resource management. Image classification plays a crucial role in remote sensing as it enables the categorization of pixels or image segments into predefined classes, such as vegetation, water bodies, or urban areas. This helps in understanding land cover and land use patterns over large areas. Quantum computing can aid in these improvements by leveraging its parallel processing capabilities and advanced algorithms. Quantum algorithms can accelerate data processing tasks and improve pattern recognition accuracy.
Quantum machine learning algorithms, such as Quantum Support Vector Machines (QSVMs) or Quantum Neural Networks (QNNs), can process image data in parallel and extract complex features and relationships that may be difficult for classical algorithms to capture. By harnessing the power of quantum computing, image classification in remote sensing can achieve higher accuracy, faster processing times, and better insights into the Earth’s surface characteristics. The interns will be researching on the development of methodologies for processing data from different satellites and generate the information required for deriving answers for the problems faced by different societies of the world.
Research area, student roles & skills
Research area: My research interests include developing new computational frameworks for implementing quantum algorithms for several generic and specific engineering and scientific questions.
Student roles: The student intern is required to develop coding framework and implement the same to analyse the satellite remote sensing data for changes in the conditions on the earth. The will develop codes using IBM's test platform QSKIT. On successful completion of this step, the students will move on to one of the QC technology platforms such as quantum annealing process provided in DWave Systems or equivalent.
Skills required: The student interns are required to have strong computational background in Python, C/C++/Java and excellent skills in translating the algorithms into the codes. They should have a deep understanding about both the classical computing and Quantum Computing technologies. They should have take courses on algorithms and any introductory course on theoretical computer science.
624. Quantum Edge Detection for Disaster Imagery Analysis
Supervisor: Rakesh Mishra
University: University of New Brunswick (Fredericton campus)
This project aims to explore quantum-based edge detection algorithms for analyzing remote sensing imagery in post-disaster scenarios such as floods, earthquakes, or wildfires. Edge detection is a fundamental task in image processing used to extract object boundaries, structural damage, and landscape changes—critical for rapid damage assessment and emergency response. Traditional edge detection methods, while effective, can be computationally intensive and less adaptable to high-resolution, multispectral satellite data.
The goal of this research is to design and test a quantum edge detection algorithm, leveraging quantum parallelism and image encoding techniques like NEQR (Novel Enhanced Quantum Representation) or FRQI (Flexible Representation of Quantum Images). By simulating quantum operations (e.g., quantum Sobel filtering or quantum Fourier transforms), the student will develop a pipeline for identifying structural edges—such as collapsed buildings, flooded zones, or wildfire perimeters—in satellite or UAV imagery.
The student will begin by implementing a classical edge detection algorithm (e.g., Canny or Sobel) to benchmark performance and accuracy. Next, a quantum version will be developed using quantum simulation tools (IBM Qiskit, Pennylane, or other platforms). The performance of the quantum algorithm will be evaluated on disaster-affected image datasets in terms of detection accuracy, scalability, and computational efficiency.
This project represents a novel intersection of quantum computing and disaster remote sensing. It has the potential to contribute to next-generation real-time analysis tools for emergency management and Earth observation.
Research area, student roles & skills
Research area: My research specializes in remote sensing and multi-sensor data fusion, with a focus on integrating hyperspectral, LiDAR, and optical data for environmental and geospatial applications. I work extensively in computer vision, object detection, and the development of advanced machine learning and deep learning models for image analysis. A core area of interest is applying quantum algorithms to remote sensing challenges, particularly in data processing, classification, and feature extraction. My work bridges cutting-edge AI techniques with real-world applications in forestry, climate monitoring, and agriculture, enabling more accurate, scalable, and intelligent analysis of complex Earth observation data.
Student roles: Key responsibilities include: -Data Acquisition and Preprocessing: Collect disaster imagery from sources such as Sentinel-1/2 (floods, fires), MODIS, or UAV datasets. Preprocess data for edge detection by converting it into grayscale or selecting key spectral bands relevant to structural changes or surface damage. -Baseline Algorithm Development: Implement classical edge detection methods (Sobel, Canny, Laplacian) using OpenCV or Scikit-Image. Evaluate their effectiveness in delineating affected zones in disaster images to establish a reference for comparison. -Quantum Algorithm Implementation:Research quantum image encoding techniques (e.g., FRQI, NEQR) and apply quantum gates or quantum convolutional operations to simulate edge detection. Use platforms like IBM Qiskit or Pennylane to construct and simulate quantum circuits. -Performance Evaluation:Assess the performance of the quantum edge detection algorithm in terms of edge sharpness, completeness, computational complexity, and suitability for remote sensing data. Compare the results with classical methods and analyze potential benefits (e.g., scalability or computational speedup). -Documentation and Reporting: Maintain organized code and records of experiments. Produce visualization of edge-detected outputs and prepare a report summarizing methodology, results, and future directions. Contribution to a research abstract or poster is encouraged.
Skills required: The student should have strong Python programming skills and experience in classical image processing techniques (e.g., edge detection, convolution, feature extraction). Familiarity with remote sensing imagery and geospatial analysis is preferred. A basic understanding of quantum computing and exposure to platforms like Qiskit or Pennylane is desirable. Knowledge of linear algebra, Fourier transforms, or quantum image representation formats (e.g., FRQI or NEQR) would be beneficial. The student should be motivated to work across disciplines and contribute to innovative applications in disaster monitoring and quantum-enhanced imaging.
625. Quantum Integrated Sensing and Communication Using Quantum Channels
Supervisor: Ekram Hossain
University: University of Manitoba (Winnipeg campus)
Quantum integrated sensing and communication (Q-ISAC) extends the traditional integrated sensing and communication paradigm in the wireless systems into the quantum domain, introducing new opportunities by leveraging quantum resources to provide inherent physical-layer security for communication, as well as realizing the sensing precision to approach the Heisenberg limit. While integrated sensing and communication (ISAC) has matured in the classical domain, and also quantum communication (QC) and quantum sensing (QS) have developed into established research areas, the joint integration of QC and QS remains largely unexplored. This project will involve development of Q-ISAC methods using quantum communication channels. It will explore various paradigms for integrating QS and QC within quantum channels (e.g., quantum circuits shared with QS and QC, Shared Quantum State Resources for QS and QC, integration of QC to Quantum radar application).
Research area, student roles & skills
Research area: My research areas include 6G and beyond wireless communications, quantum information systems, and machine learning. More information about our research is available at: https://home.cc.umanitoba.ca/~hossaina
Student roles: The student will collaborate with a Post-Doctoral Fellow, and will be primarily responsible for developing simulation models for Q-ISAC.
Skills required: The student will need to have a strong background on communication systems design, a basic background on quantum information systems, and experience of programming using MATLAB and Python. Some background on optimization techniques will be a plus.
626. Quantum Integrated Sensing and Communication in Classical Communication Channel
Supervisor: Ekram Hossain
University: University of Manitoba (Winnipeg campus)
Quantum integrated sensing and communication (Q-ISAC) extends the traditional integrated sensing and communication paradigm in the wireless systems into the quantum domain, introducing new opportunities by leveraging quantum resources to provide inherent physical-layer security for communication, as well as realizing the sensing precision to approach the Heisenberg limit. While integrated sensing and communication (ISAC) has matured in the classical domain, and also quantum communication (QC) and quantum sensing (QS) have each developed into established research areas, the joint integration of QC and QS remains largely unexplored. This project will involve development of Q-ISAC methods using classical communication channels. Integrating quantum sensing (QS) into conventional communication infrastructures enables data transmission over standard classical channels while upgrading the sensing component with quantum technology. To incorporate QS with classical communication in a Q-ISAC platform, two paradigms will be considered: (i) Quantum Sensors in Conventional ISAC Systems with Shared Communication and Sensing Carrier; (ii) Quantum Sensing and Classical Communication in Dedicated Channels.
Research area, student roles & skills
Research area: My research area is wireless communication and sensing systems, quantum communication and computing. More information about our research can be found at: https://home.cc.umanitoba.ca/~hossaina
Student roles: The student will collaborate with a Post-Doctoral Fellow and will be primarily responsible for developing simulation models for Q-ISAC systems.
Skills required: The student needs to have background in communication technology and experience in programming using MATLAB and Python. Some background on optimization techniques will be a plus.
Our project aims to develop and test quantum machine learning (QML) algorithms for classifying and detecting disease signs in medical images, utilizing QML libraries such as PennyLane and TensorFlow Quantum. By using quantum computing, we seek to enhance the analysis of complex, high-dimensional medical data. We will integrate these QML algorithms with deep learning models to improve diagnostic accuracy and efficiency. The methodology involves creating quantum algorithms for image analysis, combining them with classical neural networks, and testing the hybrid models on datasets including MRI, X-rays, and CT scans. Performance will be measured in terms of accuracy, sensitivity, specificity, and computational efficiency. Iterative refinement will optimize the algorithms for real-world applications. Expected outcomes include the development of powerful QML algorithms that significantly boost deep learning models' performance in medical image analysis, leading to faster and more accurate disease detection. This project aims to improve patient outcomes and support healthcare professionals in diagnostics through advanced technology integration.
Research area, student roles & skills
Research area: My research focuses on artificial intelligence (AI), computer vision, and intelligent robotic systems. I develop advanced algorithms for machine learning and decision-making, and innovative solutions for image and video analysis in computer vision. Additionally, I design intelligent robotic systems that integrate AI and computer vision technologies, enabling robots to perform complex tasks autonomously. This includes developing algorithms for real-time navigation, environmental interaction, and decision-making, advancing automation and smart systems.
Student roles: Read about the topic, and develop and test the algorithms created. He will have the opportunity to collaborate with our graduate students and postdocs working on this subject.
Skills required: Python programming, basic mathematics and statistics such as algebra and matrix computation. Knowledge of deep learning libraries such as PyTorch and TensorFlow is considered a plus.
With the rise of research in Quantum Technology, numerous areas have emerged showcasing the potential of quantum advantage. One such promising domain is Machine Learning (ML), where quantum models augment classical models to enhance their performance. This project aims to explore the application of Quantum Machine Learning (QML) in the sphere of predictions across various domains, such as hurricane forecasting and stock market analysis.
The primary goal of this project is to leverage a suitable QML algorithm, like the Quantum Convolutional Neural Network (QCNN), to deliver accurate predictions for a chosen problem. QML algorithms, known for their efficiency in processing quantum data, offer a significant advantage over classical counterparts by utilizing quantum superposition
and entanglement, leading to potentially exponential speedups and improved accuracy. This research aims to demonstrate the quantum advantage in practical predictive tasks, potentially revolutionizing how predictions are made in critical areas by providing faster and more accurate insights.
Research area, student roles & skills
Research area: This project aims to explore the application of Quantum Machine Learning (QML) in the sphere of predictions across various domains, such as flood forecasting and stock market analysis.
Student roles: The student has to identify the knowledge gaps and write a state-of-the-art report before implementing the identified algorithms using either simulators or cloud platforms, as applicable for the problem at hand. The student has to write papers based on the work. At the end of the internship, the student will pass on the literature collected, codes developed, data collected and used, outputs in addition to the paper(s)/presentation(s).
Skills required: Python, Quantum libraries from IBM, dWave, and other well-known sources.
629. Quantum Variational Autoencoder for Wildfire Segmentation
Supervisor: Rakesh Mishra
University: University of New Brunswick (Fredericton campus)
This project aims to develop a Quantum Variational Autoencoder (QVAE) to segment wildfire-affected areas in remote sensing imagery. Wildfires have devastating environmental and societal impacts, and timely, accurate mapping of burned areas is essential for emergency response and ecological recovery. Traditional segmentation models such as classical autoencoders or convolutional neural networks require large datasets and high computational resources. Quantum machine learning (QML), especially variational quantum circuits, offers a new approach to model complex data distributions with fewer parameters and improved generalization.
The proposed research will explore a hybrid quantum-classical architecture, where a quantum component is embedded in the encoder or latent space of a variational autoencoder. The student will train this QVAE using pre- and post-wildfire satellite imagery (e.g., Sentinel-2 or MODIS) to learn compact latent representations of fire-affected regions and reconstruct segmentation masks for burn scars. The quantum layer will be implemented using IBM Qiskit, PennyLane, or TensorFlow Quantum, and evaluated on its ability to improve segmentation accuracy or model uncertainty.
The project will begin with a classical VAE as a baseline, followed by integration of the quantum layer, training, testing, and comparison. The final output will include a wildfire segmentation model, visual results, and performance analysis against classical methods. This innovative research merges generative modeling, quantum computing, and remote sensing to address a critical environmental challenge with high potential for academic publication or prototype development.
Research area, student roles & skills
Research area: My research specializes in remote sensing and multi-sensor data fusion, with a focus on integrating hyperspectral, LiDAR, and optical data for environmental and geospatial applications. I work extensively in computer vision, object detection, and the development of advanced machine learning and deep learning models for image analysis. A core area of interest is applying quantum algorithms to remote sensing challenges, particularly in data processing, classification, and feature extraction. My work bridges cutting-edge AI techniques with real-world applications in forestry, climate monitoring, and agriculture, enabling more accurate, scalable, and intelligent analysis of complex Earth observation data.
Student roles: Key responsibilities include: -Data Collection and Preprocessing: Acquire remote sensing imagery of wildfire events from public datasets (e.g., Sentinel-2, MODIS). Preprocess the imagery by aligning time-series, generating normalized burn ratio (NBR), and preparing labeled data for training. -Model Implementation: Develop a baseline classical variational autoencoder (VAE) for segmentation of wildfire-affected areas. Extend the model by integrating a quantum variational circuit into the encoder or latent space using platforms like Qiskit or PennyLane. The student will experiment with different quantum circuit configurations and data encoding schemes. -Training and Evaluation: Train both the classical and quantum VAE models on wildfire datasets. Evaluate model performance using segmentation metrics such as Intersection over Union (IoU), Dice coefficient, and uncertainty estimation. Compare quantum and classical models in terms of performance and computational efficiency. -Documentation and Reporting: Maintain clear and organized code, log training results, and produce visualizations of segmentation outputs. Prepare a final project report or presentation, and optionally contribute to a research poster or publication draft.
Skills required: The student should have solid proficiency in Python and familiarity with data analysis libraries (e.g., NumPy, Pandas, Matplotlib). Knowledge of machine learning concepts, particularly PCA and dimensionality reduction, is essential. A foundational understanding of remote sensing and hyperspectral imagery is highly desirable. Basic exposure to quantum computing and experience with platforms like IBM Qiskit or Pennylane is an asset. The student should be motivated to explore interdisciplinary research at the intersection of quantum computing, climate science, and geospatial data processing.
630. Quantum Walk Algorithms for Sampling Chordal Graphs
Graphs represent pairwise relationships between objects, with vertices as objects and edges as connections. They are widely used to model real-world problems in biology, social networks, circuit design, scheduling, telecommunications, and data analysis, helping us understand complex systems through mathematical models.
Among the many classes of graphs, chordal (also known as triangulated) graphs are among the most extensively studied. Chordal graphs—in which every cycle of four or more nodes has a chord—are central to database theory, sparse matrix computation, and graphical models, and their structural properties enable efficient algorithms for optimization, decomposition, and learning. Many algorithms exist for chordal graphs, and we often need input instances to test them.
For such testing to be meaningful, instances should be representative rather than biased. Ideally, we would sample chordal graphs uniformly at random, or from a tunable distribution controlling properties such as edge density. This is challenging: fast classical generators tend to be biased, while unbiased samplers are expensive. A natural strategy is a Markov chain that moves through chordal graphs toward the desired distribution, but its efficiency is governed by a mixing time that can be slow.
This project aims to develop novel quantum algorithms for sampling chordal graphs. While classical Markov chain samplers exist, no quantum counterpart has been explored. The project will design a chain that preserves chordality under local moves and converges to a uniform or tunable distribution, then accelerate it using quantum walks—the quantum analogue of random walks, known to offer a quadratic improvement in mixing time. The student will analyze the chain's connectivity and spectral gap and validate the sampler on small instances.
This project offers hands-on experience in quantum computing, quantum walks, Markov chain analysis, graph theory, and algorithm design, contributing to cutting-edge advances at the intersection of quantum computation and graph algorithms.
Research area, student roles & skills
Research area: • Graph theory and its applications
• Complex network analysis
• Quantum Computing
• Natural language processing (NLP)
Student roles: Research Assistant
Skills required: • Solid background in computer science, mathematics, or related field • Knowledge of graph theory and algorithm design • Familiarity with programming languages (e.g., Python or C++) • Basic understanding of quantum computing concepts (an asset, but not required) • Strong analytical and problem-solving skills • Ability to work independently and learn quickly • Interest in quantum computing and graph algorithms • Willingness to explore and contribute to cutting-edge research
This project has three themes and will focus on the following topics:
1. designing QuBO models algorithms for combinatorial optimization problems, including vertex cover, minimum satisfiability, and asymmetric TSP.
2. The use of VQE type algorithms for solving optimization problems arising in management of electricity grids (optimal power flow, unit commitment and such).
3. The use of VQE solvers for solving optimization problems arising in machine learning.
4. The use of Machine Learning for error mitigation so that the NISQ era computers are usable for computation.
5. Use machine learning to optimize quantum circuits for VQE for NISQ devices.
When creating these algorithms, we will focus on considering topology, reductions, and noise. Additionally, we aim to compare the quantum advantage of these algorithms to similar classical algorithms. For a list of related publications please visit my website, https://www.cs.uleth.ca/~gaur
I have supervised several MITACS Globalink interns as part of my research program. Several of them are now graduate students in CS at research-intensive universities in North America and Europe. If you are motivated, technically inclined and interested in reshaping the future of machine intelligence and computation, I encourage you to apply to this project.
Research area, student roles & skills
Research area: Our research focuses on developing and analyzing hybrid quantum algorithms for optimization problems. We have developed hybrid quantum algorithms for machine learning, optimization in electrical grids and for solving differential equations. We are working on a method to optimize quantum circuits used in VQE methods. This work is supported by grants from NSERC-DG, NSERC Quantum Alliance International, NSERC Alliance International, AB Innovates, and MITACS GRI.
Student roles: As a student, you will have the opportunity to learn various techniques for designing hybrid quantum algorithms. The focus in my group is to accelerate computation using quantum approaches. You will be able to choose an application area in machine learning, electricity grid optimization, linear optimization on focus on developing and testing quantum algorithms.
For this project, we will be using Python libraries such as Pennylane, Circ, or Qiskit to code various algorithms. We will then evaluate the performance of these algorithms through simulations and measure the solution quality to determine the benefits of a quantum-accelerated approach. To conduct these simulations, we will use the Compute Canada cluster, which will give you valuable experience in programming and utilizing a cluster. Upon completion of the project, you will be expected to submit a final report detailing your research problem, methodology, and results, as well as give a presentation to the group.
As a member of the quantum accelerated algorithms team, our interns get to collaborate with Ph.D. and MSc students within our group. Upon completion of the project, the intern will be expected to present their findings.
Below is a timeline of tasks, which may overlap:
- Background reading and literature review: 3-5 weeks - Problem identification and algorithm design: 2-3 weeks - Programming and experimental evaluation: 3-5 weeks - Final report and presentation: 1 week
Detailed plan will be provided to students accepted for internships.
Skills required: We are looking for an ideal candidate who is a student in Computer Engineering or Computer Science, with a strong interest in research in theoretical computer science. It is essential to have coursework in Data Structures, Algorithms, Complexity, and Theory of Computing. Additional experience in quantum information and computation, linear programming, and combinatorial optimization would be an asset. If you are a student in Engineering Physics or a related discipline and have an interest in Quantum Computation, we encourage you to apply. Programming experience with Qiskit, PennyLane or such technologies would also be an asset.
632. Quantum equivariant diffusion model for 3D de novo molecule design
Introduction
Although state-of-the-art models such as equivariant diffusion models have achieved significant progress in generating realistic molecular structures, they still encounter limitations in handling the high-dimensional complexity and ensuring time efficiency. We hypothesize that quantum machine learning can potentially solve these problems. Therefore, we propose a Quantum Equivariant Diffusion Model (QEDM) for 3D de novo molecule design, which aims to enhance the quality of generated molecules while reducing both training and inference times. This innovative approach leverages the unique capabilities of quantum computing to overcome the inherent limitations of classical methods, promising substantial improvements in the efficiency and accuracy of molecular design.
Research goal
The overarching goal of this study is to build a QEDM to generate 3D de novo molecules with desired properties.
Proposed analysis
We will build our QEDM on the quantum platform (https://quantum.ibm.com/) provided by International Business Machines corporation (IBM). The QEDM will leverage a two-step process: a forward diffusion process and a backward denoising process. In the forward process, we will employ a non-unitary depolarizing channel to progressively add noise to the quantum-encoded representation of a molecular structure, transforming it into a maximally mixed state. Each molecule is initially represented by its atomic coordinates and types, which will be encoded into quantum states using amplitude encoding. The noise schedule will be designed using a cosine function to ensure a gradual and controlled transformation. For the backward process, we will implement a parameter-sharing quantum circuit that denoises the quantum states. This circuit will include a timestep embedding circuit to incorporate temporal information and a denoising circuit that processes the noisy quantum state along with the timestep embedding state. The quantum circuits will be designed with gates that implement rotations and translations consistent with the physical symmetries of the molecule, ensuring equivariance.
Key Words: Physics, computational biology, computer
Research area, student roles & skills
Research area: My group (https://phulab.org/) develops and applies novel artificial intelligence (mainly deep learning) and large-scale statistical algorithms for various topics related to health data science. We also collaborate very closely with local, national and international life science scientists and clinicians on different omics projects. We are actively working in the following four research areas.
(1) Deep Learning and Visual Analytics Algorithms for Omics Data, Imaging, and Electronic Medical Records;
(2) Deep Learning for Drug Discovery
(3) Integrative Analysis Frameworks for Omics Data, Imaging, and Electronic Medical Records
(4) Translational Medicine and Precision Health
Student roles: • Implement the quantum machine learning algorithms proposed by the principal investigator in the project • Test the algorithms using the collected data sets and improve the algorithms • Document the algorithm details and analysis results • Make a public presentation on the internship work • Write a manuscript or report for a potential journal or conference submission
Skills required: • Senior undergraduate students in Physics, Chemistry, Computer Science, Statistics, Biostatistics, mathematics, engineering or a related quantitative field • Basic programming skills using Python (required), MATLAB (optional). • Strong background and experience in quntum computing. • Experience in handling big data sets • Understand the basic principles of machine learning and deep learning, particularly reinforcement learning
633. Quantum optimization for machine learning and predictive analytics
Supervisor: Shohini Ghose
University: Wilfrid Laurier University (Waterloo campus)
The increasing complexity and scale of modern datasets are creating significant computational challenges for machine learning and predictive analytics. Many core tasks including feature selection, hyperparameter optimization, clustering, scheduling, resource allocation, and model training involve solving difficult optimization problems that can become computationally prohibitive using conventional methods. This project aims to investigate how emerging quantum computing technologies can accelerate and improve optimization processes that underpin machine learning and predictive analytics.
The research will focus on the development and evaluation of quantum optimization algorithms, including the Quantum Approximate Optimization Algorithm (QAOA), quantum annealing approaches, and hybrid quantum-classical techniques. These methods will be applied to representative predictive analytics problems in domains such as finance, logistics, and environment. Particular emphasis will be placed on identifyingefficient mappings from real-world optimization tasks to quantum computational frameworks.
A key objective is to establish rigorous benchmarks comparing quantum, classical, and hybrid approaches with respect to solution quality, computational efficiency, scalability, and robustness to noise. The project will leverage both quantum simulators and available cloud-based quantum processors to assess practical performance under realistic operating conditions.
The anticipated outcomes include quantum-enhanced methods for predictive analytics, open-source software implementations, and a deeper understanding of the opportunities and limitations of quantum computing for data-driven decision making. By advancing the intersection of quantum computing, machine learning, and optimization, this research will contribute to the development of next-generation computational tools capable of addressing complex challenges beyond the reach of current classical approaches. The project will provide valuable training opportunities for students in the rapidly growing field of quantum technologies.
Research area, student roles & skills
Research area: My research is in the area of quantum information science, which is the study of how to harness the laws of quantum physics to revolutionize computing, communication and information processing, and to perform novel tasks such as teleportation. I also study quantum cryptography and explore fundamental questions such as the connections between chaos theory and quantum physics. I am a theoretical physicist, but I have been working closely with experimental groups. My team has also implemented new algorithms on the IBM quantum computer and D-Wave's quantum device to apply quantum optimization applied to finance and environment.
Student roles: The student will contribute to the development and evaluation of quantum optimization methods for machine learning and predictive analytics. Working closely with the research team, the student will gain hands-on experience in quantum computing, machine learning, and scientific software development. The student's responsibilities will include conducting literature reviews on quantum optimization algorithms and hybrid quantum-classical approaches. They will assist in implementing and testing algorithms using quantum computing software frameworks, including platforms such as Qiskit and cloud-accessible quantum hardware. The student will also contribute to data analysis, benchmarking studies, and visualization of results. Through regular meetings and mentorship, the student will learn best practices in computational research, scientific programming, data management, and reproducible research methods. In addition to technical training, the student will develop valuable professional skills by participating in team discussions, presenting progress updates, and contributing to research reports and publications. The project will provide exposure to interdisciplinary research at the intersection of quantum computing, artificial intelligence, and optimization, preparing the student for future studies and careers in quantum technologies, data science, computer science, or engineering. By the end of the placement, the student will have gained practical experience with quantum software tools, advanced computational methods, and the research process from problem formulation through analysis and presentation of results.
Skills required: The student should have taken a course in Quantum Mechanics or Quantum Computing at the third or fourth year level. A background in linear algebra, and some knowledge of computer programming or numerical software such as Matlab or Mathematica is preferred. Knowledge of quantum computation is not required, but would be advantageous. The student should have good written and oral skills, and be comfortable working with other students.
634. Quantum-Enhanced Green Computing Framework for Recommender Systems
Supervisor: Mahreen Nasir Butt
University: Algoma University (Sault Ste. Marie campus)
Large recommendation systems consume substantial computational resources and contribute to increasing energy demands in data centers. This project investigates strategies for building greener recommender systems by integrating quantum-inspired optimization into recommendation model selection and hyperparameter tuning.
Instead of focusing solely on recommendation accuracy, the framework will evaluate multiple objectives including accuracy, training time, memory consumption, and energy usage. Students will compare traditional optimization methods such as grid search with quantum-inspired optimization approaches to determine whether greener configurations can be identified more efficiently.
Research area, student roles & skills
Research area: Dr. Nasir leads the Research in Artificial Intelligence and Data Analytics for Real-World Solutions, conducting interdisciplinary research that integrates AI, Machine Learning, Generative AI, and Quantum Computing to advance intelligent recommendation systems and address complex real-world challenges.
Student roles: he student will be responsible for designing and evaluating a green computing framework for recommendation systems that incorporates quantum-inspired optimization techniques to improve computational efficiency and sustainability. The project will begin with a review of recommender system architectures, energy-efficient machine learning practices, sustainable AI principles, and emerging quantum-inspired optimization approaches. The student will investigate the computational and energy costs associated with training and deploying recommendation models and identify opportunities to improve efficiency without significantly compromising recommendation quality.
Using publicly available recommendation datasets, the student will implement baseline recommendation algorithms and evaluate their performance using metrics such as accuracy, training time, memory consumption, and estimated energy usage. The student will then explore quantum-inspired optimization techniques, including QUBO-based formulations and simulated quantum annealing methods, to optimize model selection, parameter tuning, or recommendation generation processes. Energy profiling tools will be used to quantify the environmental impact of different approaches. The student will conduct comparative experiments to analyze trade-offs between recommendation accuracy, computational cost, and energy consumption. Findings will be used to develop guidelines for designing environmentally sustainable recommendation systems. Throughout the project, the student will maintain research documentation, prepare technical reports and present results.
Skills required: Students participating in this project should have a foundational understanding of programming, machine learning, and data analytics. Experience with Python and common data science libraries such as Pandas, NumPy, and Scikit-learn is recommended. Familiarity with recommendation systems, model evaluation techniques, and basic optimization concepts will be beneficial. Knowledge of sustainable AI, green computing, or energy-efficient machine learning is an asset but not required. Students should possess strong analytical and problem-solving skills, an interest in emerging quantum technologies, and a willingness to learn quantum-inspired optimization methods. Good communication and research skills are also important for documenting findings.
635. Quantum-Enhanced Multimodal AI for Efficient Data Fusion and Scene Understanding
Ce projet propose un cadre théorique et computationnel novateur pour la fusion multimodale quantique, combinant l'informatique quantique variationnelle, les réseaux de neurones impulsionnels (SNN) et l'apprentissage automatique multimodal pour traiter simultanément des flux de données hétérogènes : vidéo, séries temporelles tactiles et nuages de points 3D.
L'architecture centrale repose sur des circuits quantiques variationnels (VQC) organisés en registres modaux distincts, reliés par des couches d'intrication structurées. Chaque modalité est d'abord convertie en représentation impulsionnelle via des encodeurs SNN bio-inspirés, le timing des impulsions étant mappé sur des rotations de phase de qubits (encodage angle et encodage temporel de phase). L'intrication quantique entre registres encode ensuite les corrélations inter-modales dans un espace de Hilbert de dimension 2^N, offrant une compacité représentationnelle exponentiellement supérieure à la concaténation classique pour les corrélations d'ordre élevé.
Sur le plan théorique, le projet formalise les bornes de variance du gradient pour les circuits d'intrication structurés en registres multiples, étendant le théorème des barren plateaus (McClean et al., 2018) au régime de fusion multimodale. Une règle d'apprentissage originale nommée Quantum-STDP est proposée, pondérant les mises à jour de paramètres par les corrélations temporelles inter-modales des spike trains, s'inspirant des mécanismes biologiques de plasticité synaptique.
Les nuages de points 3D, naturellement non-ordonnés, sont traités via une représentation événementielle originale où la densité locale détermine l'activation impulsionnelle, avant encodage par amplitude quantique. Cette approche étend les méthodes d'encodage quantique aux données géométriques non-structurées, un problème ouvert dans la littérature QML actuelle.
L'ensemble des développements repose exclusivement sur des simulateurs quantiques classiques (PennyLane, Qiskit Aer) sur infrastructures HPC standards, approche méthodologiquement appropriée et représentative de la recherche QML publiée en 2025-2026 pour des circuits de 8 à 24 qubits.
Research area, student roles & skills
Research area: La fusion multimodale quantique constitue le cœur de ce projet, où des circuits quantiques variationnels (VQC) servent de mécanisme central pour fusionner des flux de données hétérogènes, vidéo, séries temporelles tactiles et nuages de points 3D. L'intrication quantique entre registres modaux encode les corrélations intermodales de manière exponentiellement plus compacte que la concaténation classique. Des réseaux de neurones impulsionnels agissent comme pré-encodeurs bio-inspirés, mappant le timing des impulsions sur des rotations de phase de qubits. Les contributions théoriques ciblent les bornes de variance du gradient pour des circuits d'intrication structurés.
Student roles: L’étudiant contribuera au développement exploratoire et expérimental du projet, en collaboration avec une équipe comprenant deux étudiants de maîtrise. Son rôle sera d’appuyer les activités techniques, méthodologiques et scientifiques liées à la fusion multimodale quantique de données hétérogènes.
Dans un premier temps, l’étudiant réalisera une revue de littérature ciblée sur les réseaux de neurones impulsionnels, l’apprentissage multimodal, l’apprentissage automatique quantique et les méthodes de fusion de données vidéo, tactiles et 3D. Cette étape permettra d’identifier les approches pertinentes, les outils logiciels disponibles et les limites actuelles de la littérature.
L’étudiant participera ensuite à la préparation des données multimodales, incluant l’organisation, le prétraitement et la transformation des vidéos, séries temporelles tactiles et nuages de points 3D en représentations adaptées aux modèles proposés. Il contribuera notamment à l’exploration de représentations événementielles ou impulsionnelles compatibles avec les encodeurs SNN et les circuits quantiques simulés.
Sur le plan technique, l’étudiant aidera à développer des prototypes en Python, sous la supervision de l’équipe. Ces prototypes pourront inclure des encodeurs SNN, des modules simples de fusion multimodale, ainsi que des circuits quantiques variationnels simulés avec PennyLane ou Qiskit. L’objectif sera de tester la faisabilité de certaines composantes du cadre proposé plutôt que de construire un système complet finalisé.
L’étudiant participera également à l’exécution des expériences, à la comparaison des approches, à l’analyse des résultats et à la production de figures, tableaux et éléments de documentation. Il devra documenter son travail de manière claire afin de faciliter la continuité du projet par les étudiants de maîtrise.
Enfin, l’étudiant prendra part aux réunions de suivi, présentera ses progrès et contribuera à la préparation d’un rapport technique ou de résultats préliminaires pouvant soutenir une future publication scientifique.
Skills required: L’étudiant devra avoir une formation en informatique, intelligence artificielle ou domaine connexe, ainsi que de solides compétences en programmation Python. Des connaissances en apprentissage automatique, apprentissage profond et traitement de données multimodales sont souhaitables. Une expérience avec PyTorch, TensorFlow, Scikit-learn, PennyLane ou Qiskit serait un atout. L’étudiant devra être capable de manipuler des données variées, notamment vidéos, séries temporelles, signaux tactiles ou données 3D. Des notions en SNN, informatique quantique ou QML seraient appréciées, mais non obligatoires. Une bonne capacité d’apprentissage autonome, de lecture scientifique et d’expérimentation rigoureuse est essentielle.
636. Quantum-Inspired Energy-Efficient Recommender System for Sustainable E-Commerce
Supervisor: Mahreen Nasir Butt
University: Algoma University (Sault Ste. Marie campus)
Modern recommendation systems often require significant computational resources, especially when processing large-scale user-product interactions. This project explores the development of an energy-efficient recommender system using quantum-inspired optimization techniques. Rather than implementing a full quantum algorithm, interns will investigate quantum-inspired approaches such as Quadratic Unconstrained Binary Optimization (QUBO) formulations and simulated quantum annealing to improve recommendation quality while reducing computational overhead.
The system will generate personalized product recommendations while simultaneously minimizing energy consumption during model training and inference. Students will compare traditional collaborative filtering methods against quantum-inspired optimization approaches using metrics such as recommendation accuracy, execution time, and estimated energy usage. The project supports sustainable AI development by investigating greener recommendation technology.
Research area, student roles & skills
Research area: Dr. Nasir leads the Research in Artificial Intelligence and Data Analytics for Real-World Solutions, conducting interdisciplinary research that integrates AI, Machine Learning, Generative AI, and Quantum Computing to advance intelligent recommendation systems and address complex real-world challenges.
Student roles: The student will be responsible for designing, developing, and evaluating an energy-efficient recommendation system that leverages quantum-inspired optimization techniques for sustainable e-commerce applications. The project will begin with a literature review of traditional recommendation systems, sustainable AI practices, and emerging quantum-inspired optimization approaches. The student will collect and preprocess publicly available user-product interaction datasets and establish baseline recommendation models using conventional collaborative filtering methods. Subsequently, the student will implement and experiment with quantum-inspired optimization techniques, such as Quadratic Unconstrained Binary Optimization (QUBO) formulations and simulated quantum annealing frameworks, to improve recommendation efficiency. The student will compare the proposed approach against baseline models using metrics related to recommendation accuracy, computational performance, and estimated energy consumption. Energy profiling tools will be used to assess the environmental impact of different recommendation strategies. In addition to technical development, the student will analyze experimental results, identify trade-offs between accuracy and energy efficiency, and investigate how sustainable computing principles can be integrated into modern recommendation systems. Throughout the project, the student will document methodologies, prepare technical reports, present findings, and contribute to the development of a research manuscript suitable for submission to an undergraduate research symposium or IEEE-style conference.
Skills required: Students participating in this project should possess foundational knowledge of programming, data structures, and machine learning concepts. Familiarity with Python and common data science libraries such as Pandas, NumPy, and Scikit-learn is highly desirable. Understanding of recommendation systems, including collaborative filtering and matrix factorization techniques, will be beneficial. Students should also have an interest in artificial intelligence, optimization methods, and sustainable computing. Prior exposure to data analysis, model evaluation, and visualization tools is advantageous but not mandatory. Curiosity about emerging quantum technologies and willingness to learn quantum-inspired optimization techniques are essential for successfully completing the project.
637. Quantum-Inspired Graph Neural Networks for Spatiotemporal Multi-Omics Simulation
Supervisor: Abedalrhman Alkhateeb
University: Lakehead University (Thunder Bay campus)
Spatiotemporal multi-omics data capture how molecules interact across tissue locations and disease time points. However, classical machine learning models struggle with the exponential complexity of these interactions. This project will design quantum-inspired algorithms (e.g., tensor networks, quantum kernel methods, or parameterized quantum circuits simulated on classical hardware) to represent and simulate molecular dynamics more efficiently.
Two interns will work on:
-Quantum circuit encoding of spatial graphs from multi-omics data (cell-cell interactions, ligand-receptor pairs).
-Hybrid quantum-classical GNNs to extract spatiotemporal dependencies.
We will use publicly available spatial transcriptomics and proteomics datasets (STOmicsDB, cBioPortal). The goal is to show that quantum-inspired methods can capture interactions that classical GNNs miss, while remaining executable on current classical hardware.
Research area, student roles & skills
Research area: My research develops AI and quantum-inspired models for biomedical data integration. I focus on spatiotemporal multi-omics (genomics, transcriptomics, proteomics, metabolomics) to model molecular interactions across time points and tissue spaces. I apply graph neural networks for spatial dependencies, temporal convolutional networks for disease progression, large language models for clinical context, and simulation frameworks for validation. My goal is to advance personalized medicine by predicting drug response, disease prognosis, and treatment outcomes from complex molecular data. I also explore quantum kernel methods and tensor networks to handle the exponential complexity of multi-omics interactions.
Student roles: Intern 1: Implement quantum-inspired kernels for spatial graph similarity; benchmark against classical methods. Intern 2: Build a hybrid model using PennyLane or Qiskit for temporal progression forecasting. Both Interns: -Attend weekly lab meetings; contribute to open-source code repository. -Present findings at a group symposium.
Skills required: Python, PyTorch/TensorFlow, basics of linear algebra and quantum computing concepts (qubits, gates, circuits). Interest in biology/omics is a plus but not mandatory.
638. Quantum-Inspired Uncertainty Quantification for Reliable Breast DCE-MRI Segmentation and Treatment Response Prediction
Supervisor: Abedalrhman Alkhateeb
University: Lakehead University (Thunder Bay campus)
Breast DCE-MRI captures tumor hemodynamics across multiple time phases (pre-contrast, early post-contrast, late post-contrast, and subtraction images). However, classical deep learning models struggle to represent the high-order, non-linear interactions between these phases, particularly for small lesions where signal is weak.
This project will design quantum-inspired algorithms (e.g., quantum kernel methods, tensor networks, or parameterized quantum circuits simulated on classical hardware) to enhance segmentation reliability and uncertainty quantification. Two interns will work on complementary tracks:
Intern 1 – Quantum-inspired feature encoding for DCE-MRI:
1-Implement quantum kernel methods to map multi-phase DCE-MRI voxels into high-dimensional Hilbert spaces.
2-Compare classical vs quantum-inspired kernels for tumor/non-tumor classification.
3-Integrate best-performing kernels into a hybrid U-Net architecture.
Intern 2 – Quantum uncertainty quantification for treatment response:
1-Develop tensor-network-based uncertainty estimates from segmentation probability maps.
2-Use quantum-inspired amplitude encoding to represent spatial and temporal feature correlations.
3-Predict pathological complete response (pCR) from clinical + imaging features + quantum uncertainty metrics.
Both interns will:
1-Work with the I-SPY2 breast cancer DCE-MRI dataset (982 cases) and external MAMA-MIA cohorts (526 cases).
2-Evaluate segmentation using Dice, IoU, HD95, and average surface distance.
3-Test reliability via threshold robustness, calibration (ECE), and model disagreement.
4-Present findings at lab meetings and contribute to an open-source repository.
*The goal is to show that quantum-inspired methods improve small-lesion segmentation (target Dice > 0.78 for smallest 20% of tumors) and provide better-calibrated uncertainty for clinical decision support.
Research area, student roles & skills
Research area: My research develops AI and quantum-inspired methods for reliable breast MRI analysis. I focus on tumor segmentation, uncertainty quantification, and treatment response prediction using DCE-MRI. I combine convolutional and residual networks with quantum kernel methods and tensor networks to model complex spatiotemporal enhancement patterns. My goal is to create clinically interpretable, reliability-aware frameworks that quantify segmentation confidence and predict pathological complete response (pCR) to neoadjuvant therapy. I also explore how quantum-inspired representations can capture non-classical feature interactions that improve small-lesion segmentation and boundary accuracy in breast cancer imaging.
Student roles: Intern 1 (Quantum feature encoding focus): 1-Preprocess DCE-MRI data into multi-channel inputs (pre-contrast, early post, late post, early subtraction, late subtraction). 2-Implement quantum kernel methods using PennyLane or Qiskit (simulated on classical GPUs). 3-Compare radial basis function (RBF) kernels vs quantum-inspired kernels (e.g., ZZFeatureMap, IQP encoding). 4-Integrate the best kernel into a modified U-Net or ResEnc nnU-Net architecture. 5-Benchmark performance: Dice, IoU, HD95, ASSD, stratified by tumor size and HR/HER2 subtype. Intern 2 (Quantum uncertainty & response focus): 1-Implement tensor-network-based uncertainty quantification (matrix product states or tree tensor networks) from segmentation probability maps. 2-Extract DCE phase contribution features (absolute and relative Dice drop when each phase is removed). 3-Build a clinical + imaging + quantum-uncertainty model to predict pathological complete response (pCR) using logistic regression or XGBoost. 4-Validate on external non-I-SPY2 cohorts (I-SPY1, Duke, NACT from MAMA-MIA). 5-Quantify performance drop from internal cross-validation to external validation (a key reliability metric). Both interns: 1-Attend weekly lab meetings (1 hour/week). 2-Contribute to a shared GitHub repository with documented code and notebooks. 3-Present a 15-minute summary at a group symposium at the end of the internship. 4-Help prepare one figure or table for a manuscript targeting IEEE Transactions on Medical Imaging.
Skills required: 1-Python (PyTorch/TensorFlow) 2-Basics of linear algebra (eigenvalues, inner products, matrix operations) 3-Interest in or basic knowledge of quantum computing concepts (qubits, gates, circuits, or quantum kernels – prior experience not mandatory but willingness to learn) 4-Experience with medical imaging (DCE-MRI, NIfTI, MONAI, or nnU-Net) is a plus 5-Interest in uncertainty quantification (calibration, entropy, ensemble disagreement) is helpful but not required
639. Quantum-Resilient EEG Biometric Authentication with Optimized Segment Lengths
Supervisor: Garima Bajwa
University: Lakehead University (Thunder Bay campus)
Authentication is a growing concern for individuals and organizations worldwide as cybersecurity challenges continue to scale and grow in complexity. Fingerprint-based authentication has become commonplace but suffers from the problem of coercion and the need for device possession at the time of authentication. For an individual, the number of biometric features is limited. We have only 10 fingerprints, a single set of facial features, and two iris images. Though it is hard to replicate these biometrics, the critical problem is that there are not many alternatives once they are compromised. EEG measurements vary depending on a person’s state of mind, the activity being performed, and his/her mood. A set of measurements made during a particular activity represents a person’s brain state for that activity. The brain state varies not only from activity to activity but also from person to person. Thus, a person’s brain state is their unique signature, like their fingerprints or DNA.
The primary objective of this research project is to develop an EEG-based authentication framework that integrates unique brain-state biometrics with emerging quantum-resilient security mechanisms. Quantum machine learning will be used to improve EEG-based user recognition, while quantum-resilient template protection will secure non-revocable brain-signal biometrics against future quantum-era attacks. Additionally, the focus is on optimizing the length of EEG segments required for authentication. This optimization is crucial as it directly influences the usability of biometric-based authentication systems. We also aim to answer the question: Can quantum kernel methods improve subject separability or reduce the EEG segment length required for reliable authentication compared with classical machine learning baselines? To achieve these goals, research will develop classical and quantum machine learning benchmarks, post-quantum-secure template protection, and evaluate authentication accuracy, segment length, latency, and security trade-offs.
Research area, student roles & skills
Research area: My research work focuses on sensing and decoding human brain states using electroencephalography (EEG), a popular modality for building Brain-Machine Interfaces (BMIs) or Brain-Computer Interfaces (BCIs). Using a person’s brain state as their unique signature, I have built BMI applications such as biometric keys, driver distraction sensing, and a blood flow regulation index. However, BMI has also opened up a new frontier of cybersecurity and data analytics challenges. Thus, my academic interests focus on developing secure, resilient, context-aware BMI applications to enhance day-to-day human experiences.
Student roles: The intern would work with Dr. Bajwa to extend her existing research on applications of Brain-Machine Interfaces. The main tasks will include: - Identify relevant public EEG data sets that are collected with different types of stimuli and tasks. - Determine the relevance and usefulness of the EEG activity duration for user authentication. - Use basic digital processing techniques to obtain noise-free data sets to be used for the authentication study. - Perform feature extraction, selection, and localization based on the performance of different classical machine learning algorithms. - Develop a Quantum ML module with a Quantum kernel-based classifier or a variational quantum circuit. - Assess whether quantum algorithms can improve the separability of EEG signatures in low-dimensional feature spaces. - Explore post-quantum template protection. - Complete a final report and focus on reproducible code.
The intern will develop the ability to conduct data-mining research, gain hands-on experience in EEG signal processing, biometric authentication, and quantum machine learning, and produce publishable work for peer-reviewed journals. A final report and presentation will be required from each intern towards the end of the internship period.
Skills required: The student should have the following desired skill sets: - Be proficient in coding one or more programming languages, such as Python, Java, or C/C++, and emerging AI tools for coding. - Have a general familiarity with cybersecurity, data mining, and machine learning concepts. - Previous experience in working with EEG data or digital signal processing, quantum algorithms is a plus. - Ability to work as part of a multidisciplinary team. - Curious, creative, collaborative problem solver, and results-oriented.
640. Quantum-Secure Design for Financial Infrastructure
Supervisor: Chen Feng
University: University of British Columbia (Okanagan campus)
This 12-week project will investigate practical applications of SILMARILS, a new information-theoretic and quantum-secure designated-verifier signature framework, to real-world financial and fintech systems. Unlike ordinary digital signatures, designated-verifier signatures allow a chosen verifier to authenticate a message while preventing the transcript from becoming universally convincing public evidence. This property is highly relevant to finance, where institutions often need strong authenticity, auditability, and accountability without exposing sensitive trading, settlement, compliance, or client information.
The intern will study the SILMARILS construction and develop application scenarios in which transferable proof is undesirable or risky. Candidate use cases include confidential interbank attestations, private proof of trade intent, regulated digital asset transfers, confidential credit or risk attestations, selective disclosure for compliance, and blockchain-based financial workflows where compact keys and signatures are valuable.
The project will combine theory, systems thinking, and prototyping. The intern will map SILMARILS’ security properties to finance-sector requirements, compare it conceptually with conventional signatures and post-quantum signature schemes, design one or two concrete protocol blueprints, and implement a lightweight proof-of-concept demonstrating a selected use case. Possible deliverables include a technical report, prototype code, evaluation of communication/signature overhead, diagrams for industry audiences, and a short presentation suitable for academic or fintech stakeholders.
This project is ideal for a highly motivated undergraduate who wants to work at the frontier of cryptography, quantum-secure authentication, blockchain infrastructure, and financial applications. The intern will gain experience reading recent research papers, translating theory into applied protocols, and producing research outputs that could lead to a workshop paper, open-source artifact, or future graduate research direction.
Research area, student roles & skills
Research area: My research area is cryptography for decentralized and high-assurance digital systems, with a focus on post-quantum and information-theoretic authentication, blockchain security, privacy-preserving verification, and lightweight cryptographic protocols. I am particularly interested in primitives that enable selective trust: proving authenticity to an intended party without unnecessarily leaking transferable evidence to the public. This work connects theoretical cryptography with practical applications in financial technology, digital assets, compliance infrastructure, confidential settlement, and secure machine-to-machine transactions.
Student roles: The student will work as a junior researcher contributing to both the conceptual and prototype layers of the project. In the first phase, the student will read the SILMARILS paper and related background on designated-verifier signatures, Shamir secret sharing, finite-field algebra, and quantum-secure authentication. They will prepare a concise internal summary explaining the core construction, assumptions, security guarantees, and limitations.
In the second phase, the student will identify finance-sector workflows where designated-verifier authentication offers a meaningful advantage over ordinary digital signatures. They will analyze requirements such as confidentiality, non-transferability, regulatory auditability, settlement finality, key size, communication overhead, and compatibility with blockchain or financial messaging infrastructure.
In the third phase, the student will design one or two application protocols based on SILMARILS. Examples may include confidential transaction receipts, selective compliance attestations, private institutional approvals, or designated-verifier proofs for digital asset operations. The student will document the threat model, message flow, trust assumptions, and expected security properties.
In the final phase, the student will implement a small proof-of-concept or simulation, evaluate basic performance and communication costs, and prepare final deliverables: prototype code, a technical report, presentation slides, and a short application note aimed at fintech or digital asset audiences.
Throughout the internship, the student will participate in weekly research meetings, present progress, receive feedback from the supervisor and research group, and contribute original ideas. Strong performance may lead to co-authorship on a follow-up paper, continued remote collaboration, or future graduate research opportunities.
Skills required: The student should have strong foundations in discrete mathematics, algorithms, and probability, plus interest in cryptography, security, blockchain, or financial technology. Prior exposure to finite fields, secret sharing, digital signatures, zero-knowledge proofs, or post-quantum cryptography is helpful but not required. Programming ability in Python, Rust, Go, or TypeScript is desirable for prototyping and evaluation. The ideal student is mathematically mature, curious, independent, careful with definitions, and excited to read research papers and turn abstract cryptographic ideas into practical systems.
641. Quantum-Secured Full-Duplex Wireless Communication for Future 6G Networks
Supervisor: Malek Alsmadi
University: Lakehead University (Thunder Bay campus)
This project investigates quantum-secured full-duplex wireless communication for future 6G networks. Full-duplex communication allows wireless nodes to transmit and receive signals at the same time and on the same frequency, which can improve spectral efficiency. However, full-duplex systems face major challenges, including self-interference, noise, channel estimation errors, and security threats.
The student will study how quantum-secured or quantum-inspired communication concepts can enhance the security of full-duplex wireless systems. The project may consider quantum key distribution principles, quantum-inspired physical-layer security, and secure transmission methods under practical full-duplex impairments.
The student will develop a simplified simulation model using MATLAB, Python, or another suitable tool. The model may include full-duplex transmission, self-interference, channel noise, interference, eavesdropping scenarios, and selected security metrics. The student will compare conventional full-duplex secure communication with quantum-enhanced or quantum-inspired approaches.
The expected outcome is a focused simulation-based feasibility study, including technical results, plots, and a short final report. Depending on the quality of the results, the work may contribute to a future conference paper or research manuscript.
Research area, student roles & skills
Research area: My specialized research area includes full-duplex wireless communication, MIMO systems, signal processing, physical-layer security, millimetre-wave communication, visible light communication, and emerging 6G networks. My research focuses on designing reliable, high-capacity, and secure wireless systems under practical impairments such as self-interference, channel uncertainty, noise, and hardware limitations. This project extends these areas by exploring quantum-secured and quantum-inspired approaches for secure full-duplex communication.
Student roles: The students will work under my supervision to conduct a focused literature review, develop simulation models, analyze results, and prepare technical documentation. The students will begin by reviewing selected papers on full-duplex wireless communication, self-interference cancellation, physical-layer security, and quantum-secured or quantum-inspired communication.
The students will then implement a simplified full-duplex wireless communication model using MATLAB, Python, or another suitable tool. The simulation work may include modelling self-interference, wireless channel conditions, receiver noise, eavesdropping scenarios, and selected security-related performance metrics. The students will evaluate system performance under a limited number of scenarios and generate plots to compare conventional secure full-duplex communication with quantum-enhanced or quantum-inspired approaches.
The students will meet with me regularly to discuss progress, receive feedback, and refine the simulation assumptions. By the end of the 12-week internship, the students are expected to prepare a short technical report summarizing the literature review, simulation model, results, and main findings. The work may also contribute to a future conference paper or research manuscript.
Skills required: The student should have a background in engineering, computer science, physics, or a related field. Knowledge of communication systems, signal processing, probability, linear algebra, and programming is preferred. Experience with MATLAB, Python, wireless communication simulation, or quantum communication concepts would be an asset. Interest in full-duplex communication, cybersecurity, quantum communication, or 6G networks is highly desirable.
The increasing deployment of smart grids and smart meters enables utility providers to collect fine-grained energy consumption data in near real-time, improving load management, demand forecasting, and operational efficiency. However, the collection of detailed consumption data raises significant privacy concerns, as it can reveal sensitive information about users’ daily activities and behavior patterns.
Building on my previous research in privacy-preserving smart metering, this project aims to develop a quantum-resistant privacy-preserving smart metering framework that remains secure against both classical and future quantum computing threats. The research will investigate the integration of post-quantum cryptographic techniques (such as Lattice-Based Cryptography) with privacy-enhancing technologies (i.e, Homomorphic encryption and Anonymization ) to enable secure energy consumption aggregation, billing, and analytics without exposing individual user data.
The proposed framework will employ quantum-resistant cryptographic primitives and advanced encryption mechanisms to ensure confidentiality, integrity, and privacy throughout the data collection and processing lifecycle. Particular emphasis will be placed on designing protocols that are computationally efficient, scalable, and suitable for deployment in resource-constrained smart meter environments.
To evaluate the practicality of the proposed solution, prototype implementations will be developed and tested on embedded IoT platforms such as Raspberry Pi and microcontroller-based systems. The outcomes of this research will contribute to the development of secure and privacy-preserving smart energy infrastructures capable of withstanding emerging quantum computing threats.
Research area, student roles & skills
Research area: I hold Ph.D. in Computer Science and Engineering. My research focuses on cybersecurity, machine learning, and data science. During my Ph.D., I specifically concentrated on developing secure smart city systems using big data technologies and cryptographic methods. Over the past six years, as a postdoctoral fellow and assistant professor, my work has centered on designing novel information security and privacy protocols for resource-constrained IoT systems, as well as developing AI models for fraud detection in cryptocurrencies. In these areas, I have published over 70 peer-reviewed articles in journals and conferences, and 4 book chapters, with more than 6,000 citations.
Student roles: --Conduct a literature review on smart metering systems, data privacy, and post-quantum cryptography. --Assist in analyzing privacy and security challenges in smart grid environments. --Study existing privacy-preserving smart metering schemes and identify opportunities for enhancing their quantum resistance. --Support the implementation of selected cryptographic algorithms and protocols in a simulated or prototype environment. --Develop and test software modules using programming languages such as Python or C/C++. --Assist in implementing and evaluating the proposed framework on IoT platforms such as Raspberry Pi or Arduino. --Collect and analyze performance metrics, including computational overhead, memory usage, and communication costs. --Participate in regular research meetings and present progress updates. --Document experimental results and contribute to technical reports and research publications. --Gain hands-on experience in cybersecurity, privacy-preserving technologies, embedded systems, and applied research methodologies.
Skills required: -- Background in Computer Science, Computer Engineering, Electrical Engineering, Cybersecurity, or a related discipline. --Strong understanding of computer networks, information security, and cryptographic principles. --Knowledge of privacy-preserving technologies, homomorphic encryption, or post-quantum cryptography is desirable. --Proficiency in programming languages such as Python, C/C++, or Java. --Experience with algorithm design, data structures, and software development. --Familiarity with embedded systems and IoT platforms (e.g., Raspberry Pi, Arduino, ESP32) is an asset. --Interest in cybersecurity, smart grids, smart metering systems, and privacy-enhancing technologies. --Ability to conduct literature reviews, analyze research papers, and develop innovative solutions. --Strong analytical, problem-solving, and critical-thinking skills. --Good written
643. RISC-V Development
Supervisor: Felipe Gohring de Magalhaes
University: École Polytechnique de Montréal
Location: Montreal, Québec
Start date: 2027-06-01 (flexible)
Disciplines: Computer Science, Engg-Computer, Engg-Software, Engg-Systems and Technology
This internship is focused on developing hardware modules focused on avionics platforms, respecting different standards (ARINC 653, ARINC 429, AFDX, etc). The intern will have hands-on experience with avionic applications and work in collaboration with PhD students on the field.
Research area, student roles & skills
Research area: The project is under the scope of initiatives to improve the cyber resilience of avionic architectures, involving different scopes: operating systems modules, hardware monitors, dynamic behaviours.
Student roles: work with doctoral students supporting them in the creation of new modules, validation of existing modules, results collection and results analysis.
Skills required: Hardware description languages, test-benches, Embedded systems, understanding of operating systems and low level programming (C, assembly).
644. Rapid human digital twin modeling for human robot collaboration
Efficient and safe human-robot collaboration (HRC) is a typical application of the HDT and a critical component of smart manufacturing. This project explores HRC safety within NVIDIA Isaac Sim by simulating a human presence alongside a collaborative robotic arm. The core challenge is accurately representing the human body and the robot for realistic motion simulation while keeping the system computationally manageable for real-time physics.
The human body is modelled using a parametric geometric model, such as SMPL/SMPL-X or Momentum Human Rigs (MHR). The geometric model defines body shape and poses using a compact set of learned parameters, enabling the simulation to generate realistic body meshes and joint motion rather than relying on crude placeholder geometry. The parametric model is imported into Isaac Sim as a NVIDIA Universal Scene Description (USD) asset, with its skinned mesh and skeleton driving physically plausible motion sequences, such as walking, reaching, and bending, within a shared robot workspace. The robotic arm executes a repetitive pick-and-place task using Isaac's action graph or a ROS 2 controller. A proximity detection system then monitors distances between the human model's collision volumes and the robot's links each timestep, triggering a graduated safety response (slowdown → stop) as the human enters defined threshold zones.
Main Tasks:
(1) Basic Tools and Technical Concept: Get familiar with the basic concepts of the HDT, parametric human geometric model (e.g. SMPL/SMPL-X, MHR), and NVIDIA Issac Sim.
(2) Parametric Model Integration: Export SMPL-X or MHR mesh/skeleton into USD format and import into Isaac Sim.
(3) Motion Animation: Drive the human model through scripted motion sequences using pre-defined motion capture data.
(4) Collision Volume Setup: Attach bounding capsules/spheres to key body segments for efficient runtime queries.
(5) Proximity Detection: Develop a computational method to evaluate human-robot distances per simulation timestep
(6) Safety Response Logic: Implement zone-based
Research area, student roles & skills
Research area: My research focuses on developing modelling and simulation methods for human digital twins (HDT). The human digital twin is an executable digital copy of an individual. In this paradigm, a detailed digital replica of the human body can accurately simulate real-time biomechanical movements, organ system functions and changes, as well as biochemical and pharmacological conditions. My research mainly involves three objectives: first, to define a unified structure and components for HDT and to summarize related application fields and critical technologies; second, to develop HDT human rapid modelling approaches that integrate with AI-driven methods; third, to develop the HDT human simulation
Student roles: The student will work with a team of other undergraduate and graduate students in carrying out the required tasks. The working hours are 40 per week. The student will also need to conduct a literature survey and summarize the status quo of other methods through a thorough comparative analysis. Finally, the student will summarize their findings, analyze their results, and prepare tables, graphs, or other documentation to present their research. The research progress is expected to be presented during group meetings.
Skills required: The ideal candidate should have a background in mechatronic/control engineering, mechanical engineering, electronic engineering, computer science, or a related discipline. Preference will be given to students with experience in hardware setups, machine learning, robotic applications, and computer vision. Training will be provided in these areas depending on the candidate's qualifications. The student must be hardworking, motivated, and able to work effectively in teams. Preferred qualifications: Experience with C++, C #/Python, CAD, and ROS 1/2. Experience with parametric human geometric models (e.g., SMPL/SMPL-X, MHR) and NVIDIA Issac Sim is highly preferred!
645. Real-Time Backend System for Simultaneous Soil Sensor Data Streaming via HTTPS
Supervisor: Asim Biswas
University: University of Guelph
Location: Guelph, Ontario
Start date: 2027-05-03 (flexible)
Disciplines: Computer Science, Engg-Computer, Engg-Software, Engg-Systems and Technology
The research project focuses on developing a backend infrastructure capable of handling real-time data streams from numerous soil-based sensors transmitting simultaneously over HTTPS. As precision agriculture increasingly relies on dense sensor networks for monitoring soil moisture, nutrients, and other parameters, the ability to securely ingest, process, and store high-volume time-series data from thousands of concurrent sensor connections becomes critical.
We will design and implement a scalable microservices architecture specifically engineered for concurrent HTTPS data ingestion from soil sensors. The system will incorporate mutual TLS authentication for secure sensor-to-server communication, efficient data validation and normalization pipelines (including range checks and geolocation tagging), and optimized storage solutions for time-series soil sensor data using technologies like TimescaleDB or InfluxDB.
Key technical components include: HTTPS intake services with load distribution and horizontal scaling capabilities; RESTful APIs for data retrieval and aggregation; basic visualization dashboards (using Grafana or custom frontend); and comprehensive monitoring, fault tolerance mechanisms (circuit breakers, retry procedures), and security measures.
The project follows a phased approach over 10 weeks: requirements gathering and architecture design (weeks 1-2); core implementation of ingestion services, validation modules, and storage pipelines (weeks 3-6); load testing targeting 10K+ simultaneous HTTPS connections, performance optimization, and security audits (weeks 7-8); and finally cloud deployment with CI/CD pipelines, documentation production, and knowledge transfer (weeks 9-10).
Expected outcomes include a production-ready backend system capable of securely ingesting data from thousands of soil sensors with sub-second latency, scalable architecture accommodating growth in sensor count and data frequency, open-source codebase with permissive licensing, and detailed performance benchmarks. This system will enable precision irrigation and fertilization in farming, support early detection of soil contamination or erosion, provide a reusable platform for large-scale sensor network experiments, and deliver economic benefits through reduced water waste and increased crop yield via data-driven decisions.
Research area, student roles & skills
Research area: I am a professor and Canada Research Chair in Digital Agriculture at the University of Guelph, working in sustainable soil management, specifically data-driven management connecting technology, soil data, and traditional understanding of soil processes and crop production. To me, current-day multi-faceted problems need solutions with multiple dimensions. This requires an integrated and interdisciplinary approach to solving issues. With a strong background in soil science, mathematics, statistics, and sensor development, I collaborate with experts from other fields to develop multi-disciplinary projects.
Student roles: The student will act as a Backend Engineer contributing to core system development. Responsibilities include: designing and implementing HTTPS ingestion services with mutual TLS authentication; building data validation and enrichment modules (range checks, geolocation tagging); developing RESTful APIs for data retrieval; optimizing time-series database schemas and write pipelines; participating in load testing and performance optimization; assisting with security audits and fault tolerance mechanisms (circuit breakers, retry queues); and helping produce deployment documentation. They will work within the team structure, attending design discussions, writing clean, tested code, and contributing to knowledge transfer sessions. This hands-on role provides direct experience in scalable microservices, secure IoT data handling, and cloud-native deployment practices while advancing the project's technical milestones.
Skills required: The ideal student should have strong backend programming skills (Go or Node.js), experience with RESTful APIs and HTTPS/TLS security, and familiarity with microservices architecture. Knowledge of time-series databases (TimescaleDB/InfluxDB), containerization (Docker/Kubernetes), and Linux environments is essential. Understanding of data validation, normalization, and basic IoT concepts will be beneficial. Proficiency in Git, CI/CD pipelines, and performance testing tools is expected. Prior coursework or projects in distributed systems, cloud platforms (AWS/Azure/GCP), or sensor data processing is advantageous. Good communication and teamwork abilities are required for collaboration in this research setting.
646. Real-Time Image-Guided Surgical Scene Segmentation via Spatio-Temporal State Space Models (Mamba)
This research project focuses on developing a real-time, spatio-temporal computer vision system for image-guided surgery (IGS) using the cutting-edge Mamba architecture (State Space Models). In minimally invasive and robotic clinical procedures, tracking fine-grained instrument positions and structural organ boundaries simultaneously is a critical safety requirement. However, processing high-resolution video streams in real-time creates significant computational challenges. Traditional deep learning architectures, such as Transformers, suffer from quadratic computational complexity over time, making them too resource-intensive to deploy on standard hospital computing hardware.To address this operational constraint, the selected intern will build a next-generation surgical scene segmentation network based on 2D/3D bidirectional State Space Models (such as U-Mamba or SegMamba). Unlike attention mechanisms, Mamba achieves linear time complexity while maintaining long-range dependency tracking across sequential frames. The student will use public benchmark video datasets (such as Cholec80 or EndoVis) to train the model to map spatial intra-frame features and temporal inter-frame dependencies simultaneously.The project emphasizes structural efficiency. The intern will couple the Mamba token sequence layers with a lightweight convolutional decoder to generate pixel-level segmentation masks. The core objective is to deliver state-of-the-art segmentation accuracy (measured via Dice score and Mean Intersection over Union) while consistently maintaining inference speeds above 30 frames per second. This research aims to prove that next-generation sequence modeling can enable lightweight, instantaneous clinical AI navigation systems directly at the surgical bedside.
Research area, student roles & skills
Research area: My research focuses on developing Machine Learning algorithms and Advanced Computer Vision systems to solve complex data and automation challenges. This includes core work in matrix-based dimensionality reduction, incremental visual learning, and secure intelligent systems. My current focus centers on deep learning and generative AI architectures applied across text and visual streams for secure large language model (LLM) refinement, data-centric cybersecurity, and pixel-level medical image segmentation. Ultimately, this research aims to design robust, resource-efficient machine learning models optimized for real-time deployment and digital health security.
Student roles: The selected intern will act as a primary machine learning researcher and will be responsible for the following sequential tasks: - Literature Review & Setup: conduct a targeted literature review on State Space Models (SSMs) and Mamba architectures in computer vision, and configure the PyTorch development pipeline. - Data Management: preprocess visual data streams and structure sequential surgical video benchmarks (Cholec80/EndoVis). - Model Development: code, train, and specialize bidirectional Mamba sequence layers for spatio-temporal tissue and tool segmentation. - Benchmarking & Testing: design validation loops to measure segmentation accuracy (mIoU/Dice scores) against operational latency constraints (Frames Per Second). - Academic Contribution: writing a peer-reviewed scientific paper, structuring experimental results, and preparing the manuscript for publication. - Documentation: maintain a clean GitHub codebase and compile findings into a structured final research report.
Skills required: - Level & Field: Undergraduate or Master's student in Computer Science, Data Science, Software Engineering, Biomedical Engineering, or a related field. - Programming: High proficiency in Python and deep learning frameworks like PyTorch or TensorFlow is required. - Deep Learning: Solid understanding of core sequence modeling concepts (CNNs, RNNs, or Transformers) and semantic image segmentation metrics. - Tools: Practical experience with Git/GitHub and using open-source machine learning repositories is highly desired.
647. Real-Time Multimodal Signal Capture for AI-Assisted Data Analysis
This project investigates how real-time multimodal signals, including eye-tracking, speech, and interaction events, can be captured from users during data analysis tasks and used to ground AI agent assistance in what users are actually seeing, saying, and doing. Rather than relying on explicit user queries, the we aim to develop a system that proactively streams contextual signals to a multimodal AI model, enabling more natural and responsive human-AI collaboration during exploratory data analysis.
The project extends the open-source reVISit platform (https://revisit.dev/) with new signal capture capabilities, developed and deployed as a reusable open-source library. A central technical challenge is investigating which combinations of signals are most informative, for example, eye-tracking measures reveal where attention is directed, speech captures the analyst's in-the-moment reasoning, and interaction events reflect deliberate analytical decisions. Understanding how these signals complement one another, and how reliably they can be captured and synchronized in a browser-based environment, is a core research question.
Working closely with a doctoral student and the principal investigator, the intern will design and integrate signal-capture modules, develop logging and debugging tools, and evaluate the resulting system across a range of data analysis scenarios. A particular focus will be assessing data quality, including identifying failure modes, signal dropout, synchronization issues, and other sources of noise that may affect downstream AI performance.
The expected outcomes of this research include a better understanding of which multimodal signals most effectively support AI-assisted data analysis, design guidelines for building proactive AI agents that respond to implicit user context, and an open-source infrastructure that the broader visualization and HCI research community can build upon. The intern will also contribute to a research paper reporting these findings, and will participate actively in lab life throughout the internship.
Research area, student roles & skills
Research area: Dr. Crisan is a computer scientist whose research in human-centered artificial intelligence lies at the intersection of human-computer interaction and artificial intelligence. Her research focuses on designing AI systems that people can understand, trust, and effectively collaborate with, emphasizing human-AI partnership rather than replacing human judgment. Drawing on her background in data visualization, she investigates how interactive visual interfaces can help people inspect, question, and guide AI systems, enabling more effective decision-making. A central theme of her work is responsible, transparent, and trustworthy AI, including approaches that make AI systems more interpretable, auditable, and aligned with human goals.
Student roles: The student will work collaboratively with a doctoral student and the PI to advance the project across four areas: 1) Platform Development. The core responsibility is extending the reVISit platform with new multimodal signal capture capabilities. This includes investigating the utility of different input signals, integrating eye-tracking measures alongside existing speech and interaction capabilities, and building and testing the supporting infrastructure. React proficiency is essential for this work. The developed components should be packaged and shipped as a reusable open-source library.
2) Case Studies and Data Quality. The student will help configure and test a range of analysis scenarios and case studies using the platform, working closely with the doctoral student to examine signal quality, identify failure modes, and assess the reliability of captured multimodal data across different conditions.
3)User Study Support. The student will contribute to the design of a formal user study evaluating the platform, and will support the doctoral student in recruiting participants, running sessions, and ensuring smooth data collection. Prior experience with HCI study protocols is an asset but is not required.
4)Research Dissemination. The student will contribute to the preparation of a research paper, including producing figures, supporting quantitative and qualitative data analysis, and writing or editing sections of the manuscript as appropriate.
In addition to the above, the student is expected to be an active member of the research group: - Attending regular lab meetings and scheduled meetings with the PI - Presenting progress and interim findings at lab meetings throughout the internship - Collaborating respectfully and effectively with other lab members and collaborators across the wider team
This is a hands-on, collaborative research internship with real deliverables; the student will leave with tangible contributions to an active open-source project and an in-progress research paper.
Skills required: The ideal candidate has a background or strong interest in human-computer interaction research and experience working with modern AI/LLMs via open-source models or APIs. Prior web development experience is required; while AI coding tools are permitted, a solid programming foundation remains important. The ability to analyze both quantitative and qualitative user data (from experiments, interaction logs, and think-aloud protocols) is required. Experience conducting HCI user studies is not required but would be a meaningful advantage. Strong written English is essential, as the project involves a technical report and conference paper draft.
648. Real-Time Signal Quality Enhancement for Optical Brain Monitoring Using Adaptive Filtering and Motion Artifact Correction
This project focuses on developing algorithms to improve the quality and reliability of optical brain monitoring signals. Near-infrared spectroscopy (NIRS) and time-domain optical signals are often contaminated by motion artifacts, physiological interference, and changes in probe-tissue coupling, which limit their clinical and research usability.
The intern will develop and evaluate signal processing methods for artifact detection, suppression, and correction in optical time-series data. Approaches may include adaptive filtering, frequency-domain analysis, wavelet-based denoising, and machine learning methods for signal classification and reconstruction.
The student will work with real experimental datasets from optical brain monitoring systems and evaluate the performance of different techniques for preserving physiological information, such as cerebral oxygenation dynamics, while removing noise components.
The goal is to produce a robust preprocessing pipeline that can be integrated into next-generation deepNIRS and time-domain optical imaging systems developed in the laboratory. This work is essential for improving downstream AI analysis and imaging reconstruction accuracy.
The student will gain experience in biomedical signal processing, computational modeling, and optical neurotechnology applications.
Research area, student roles & skills
Research area: We develop advanced signal-processing methods for biomedical optical systems used in noninvasive brain monitoring. Our research focuses on improving the quality of near-infrared spectroscopy and time-domain optical signals by removing motion artifacts and physiological noise using adaptive filtering and computational methods.
Student roles: The student will design and implement algorithms to improve the quality of optical brain-monitoring signals. Responsibilities include preprocessing NIRS/DTOF datasets, identifying and characterizing motion artifacts, and developing filtering and correction techniques.
The student will evaluate different signal enhancement strategies using quantitative metrics such as signal-to-noise ratio improvement and preservation of physiological features. They will also compare classical signal processing methods with modern data-driven approaches.
The intern will work closely with graduate students and faculty members, participate in lab meetings, and contribute to documentation and visualization of results. Depending on progress, the student may integrate their signal enhancement pipeline into existing optical imaging systems in the lab.
This project provides interdisciplinary training in biomedical signal processing, optical neurotechnology, and computational methods for healthcare applications.
Skills required: Background in engineering, physics, computer science, or biomedical engineering. Strong programming skills (Python or MATLAB) required. Familiarity with signal processing, time-series analysis, or machine learning is highly desirable. Interest in biomedical data and applied computation is essential.
649. Real-Time Surgical Tool and Tissue Segmentation via Vision Foundation Models
This research project focuses on adapting large, pre-trained Vision Foundation Models (such as Meta's Segment Anything Model - SAM) to achieve real-time semantic computer vision within image-guided surgery (IGS). Minimally invasive clinical interventions and robotic surgeries demand extreme geometric and structural precision. However, live surgical video feeds are heavily corrupted by dynamic artifacts, including bleeding, camera smoke, lens fluid, specular reflections, and unexpected tissue deformations. Traditional deep learning approaches fail here due to a lack of massive, pixel-perfect annotated medical datasets.To overcome this bottleneck, the selected intern will develop a domain-adaptation pipeline utilizing Parameter-Efficient Fine-Tuning (PEFT) techniques, with a specific emphasis on Low-Rank Adaptation (LoRA). Instead of updating billions of parameters, the model will learn specialized low-rank matrices tailored to identify metallic tools, fine-grained instrument contours, and boundary tissue interfaces. The student will leverage open-source benchmarks (such as Cholec80 and EndoVis datasets) to design custom, domain-specific visual loss functions that penalize tracking errors near critical structures.A major pillar of this project is computational optimization. The intern will apply model acceleration frameworks to streamline the network's spatial encoders. The ultimate objective is to maintain state-of-the-art segmentation accuracy (measured via Mean Intersection over Union) while pushing inference speeds beyond 30 frames per second. This ensures the entire computer vision architecture can seamlessly execute on standard, low-resource operating room computing hardware, transforming heavy foundation networks into responsive, life-saving surgical navigation assistants.
Research area, student roles & skills
Research area: My research focuses on developing Machine Learning algorithms and Advanced Computer Vision systems to solve complex data and automation challenges. This includes core work in matrix-based dimensionality reduction, incremental visual learning, and secure intelligent systems. My current focus centers on deep learning and generative AI architectures applied across text and visual streams for secure large language model (LLM) refinement, data-centric cybersecurity, and pixel-level medical image segmentation. Ultimately, this research aims to design robust, resource-efficient machine learning models optimized for real-time deployment and digital health security.
Student roles: The selected intern will act as a primary machine learning researcher and will be responsible for the following sequential tasks: - Literature Review & Setup: Conduct a targeted literature review on parameter-efficient fine-tuning (PEFT) and adapt software environments using Python and PyTorch. - Data Management: Preprocess and organize open-source surgical video datasets (such as Cholec80 or EndoVis). - Model Development: Implement domain-adaptation layers (like Low-Rank Adaptation - LoRA) to optimize pre-trained Vision Foundation Models (SAM) for surgical environments. - Benchmarking & Testing: Design validation tests to evaluate pixel-level segmentation accuracy (mIoU) against processing speed (Frames Per Second). - Academic Contribution: Collaborate on writing a peer-reviewed scientific paper, structuring experimental results, and preparing the manuscript for publication. - Documentation: Maintain a clean GitHub codebase and compile findings into a structured final research report.
Skills required: - Academic Level & Field: Undergraduate/Master's student in Computer Science, Data Science, Software Engineering, Biomedical Engineering, or a related quantitative field. - Programming & Frameworks: Strong proficiency in Python is mandatory. Hands-on experience developing and training machine learning models using deep learning frameworks like PyTorch or TensorFlow is required. - Computer Vision Foundations: A good to solid conceptual understanding of standard computer vision workflows, including convolutional neural networks (CNNs), semantic image segmentation, and object detection. - Tools & Repositories: Practical experience with Git/GitHub and open-source model repositories like Hugging Face is highly desirable.
650. Realistic pedestrian models for driving simulation
One of the biggest challenges in autonomous and assistive driving is safe interaction with pedestrians. These interactions in a real world can be quite complex and varied. For instance, road layout, visibility, background traffic, etc., can change behaviors of all road users. To develop systems for assisting drivers, we need data that represents the diversity of real-world traffic. However, collecting such data is time-consuming and costly; it requires many vehicles driving around in hopes to record an interesting scenario, which then has to be extracted and annotated by hand.
In contrast, simulations can be used to generate virtually any type of scenario along with accurate annotations on demand. But the main limitation of the simulation is lack of realism. Specifically, simulated pedestrians rarely jaywalk, are able to everything around them, and tend to walk with constant speed.
This project will build on the past work in my group to develop a more realistic pedestrian model within an open-source driving simulation CARLA. Specifically, we aim to implement one or more of the following:
1) realistic perceptual limitations, e.g. agents that can perceive only a portion of the environment;
2) realistic movement, e.g. agents that change their speed and trajectory depending on their internal goals and characteristics (e.g. age, gender, risk-taking) and external environment (e.g. other cars and pedestrians nearby);
3) complex interaction scenarios, e.g. how to design the environment and position multiple agents within it to result in non-trivial interactions;
4) a procedure for validating the properties of the new agents. This can be in the form of testing autonomous driving models on the new data or by comparing the behavior of agents to recordings of pedestrians in the wild.
Research area, student roles & skills
Research area: My research area is building computer vision systems inspired by human vision and cognition for various applications, with focus on intelligent transportation, assistive driving systems, and autonomous driving. My main goal is to advance human-centered traffic safety by developing human-compliant models and identifying open problems through application of domain knowledge, extensive data analysis, and systematic benchmarking.
I am interested in understanding decision-making of pedestrians (e.g. how and when they decide to cross or yield) and visual attention of the drivers (e.g. what do drivers look at in the busy traffic scene).
Student roles: The student will work on the project under my supervision. I will provide them with a set of academic papers that will help them learn about past relevant work, a source code repo containing our ongoing work on CARLA, and access to workstation to run the simulation.
The student will: - design enhanced pedestrian agents with one or more features described above; - implement enhanced pedestrian agents in CARLA in Python and/or C++; - design evaluation procedure and criteria for evaluation; - conduct evaluation; - write a report in a conference paper format describing the implementation.
Skills required: The minimum required skills/background are: - strong programming skills (especially in Python); - interest in research, desire to learn and explore new problems.
Any of the following will make your application stand out: - research experience (e.g. working in research labs, personal projects, publications); - experience with computer vision (e.g. courses, personal projects); - experience with PyTorch (or other deep learning frameworks) - experience with data analysis (e.g. pandas) and visualization - exposure to cognitive science/psychology (e.g. through courses or personal interest) - familiarity with simulation engines (e.g. Unity) and ROS
651. Recent Advances in Operations Management and Business Analytics
Supervisor: Tim Huh
University: University of British Columbia (Vancouver campus)
Supply chains are a vital part of the Canadian economy, and they consist of an array of industries such as production, distribution, transportation, international trade and retail. Matching supply and demand is challenging due to the scale, complexity and constantly changing nature of a supply chain. Supply chain decisions often require information visibility across functional areas and affect multiple stakeholders. However, supply-side and demand-side decisions are frequently made in silos. The research goal is to improve the overall performance of supply chains. While we are interested in reducing cost on the supply side and increasing revenue on the demand side, we will focus on maximizing profit through the joint control of supply and demand and assessing the benefit of joint control. Details of this project will be finalized through consultation through the student and professor.
Research area, student roles & skills
Research area: Operations management; supply chain management
Pricing and revenue management
Capacity planning
Inventory control
Student roles: Undergraduate students will be exposed, as much as possible, to the spectrum of research activities ranging from research problem identification to problem formulation, modeling, analysis, simulation and the communication of research outcomes.
Skills required: Strong mathematics, computer science or statistics background and maturity. Ability to work independently. Serious interest in pursuing doctoral level graduate research.
652. Reconstruction of Text and Illustrations from Sealed Historical Books
Supervisor: George Knopf
University: Western University (London campus)
Location: London, Ontario
Start date: 2027-05-03 (flexible)
Disciplines: Computer Science, Engg-Software, Engg-Electrical, Engg-Mechanical, Cultural Studies, Archaeology
Historical books, scrolls, and codices are often very fragile and extensively damaged due to material decomposition and the effects of the storage environment (e.g., humidity, mold). Opening the physical document to reveal the text and images will often result in irreparable damage and the loss of critical information. However, these fragile sealed documents can be safely digitized using micro-CT scanning and other non-destructive imaging techniques. The CT scan of the artifact provides a 3D point cloud of individual voxels where each voxel reflects the material properties at that spatial location. The process of digitizing and reconstructing the closed book involves several stages: acquiring a series of planar micro-CT images, segmenting the coordinate data associated with the cover and individual pages, determining the freeform geometry of the individual pages in the book, and mapping the corresponding material information (i.e., ink) onto each page through a process called texturing. However, the process of separating pages and mapping the text and images onto the digital page is complex because each handmade page can have a different thickness, be partially decomposed, and is likely to be warped (i.e., non-flat). In addition, the pixel intensities of CT images, or slices, represents the percentage of photons absorbed in the material. These mass absorption coefficients can vary within the document due to paper fabrication, ink formulation, and material decomposition. The Mitacs project involves developing, implementing, and validating robust algorithms for creating 3D representations of handwritten, or early movable type, text and illustrations (including hand drawn doodles and marginalia). The sliced images of the closed historical manuscripts and books are acquired using a micro-CT scanner available at Western University. The collaborative research also involves cultural experts from Western’s Department of Anthropology and Western Libraries (Special Collections).
Research area, student roles & skills
Research area: Prof. Knopf’s interdisciplinary research activities involve 3D geometric modeling, new materials and fabrication techniques for creating the next generation of mechanically flexible optical and electronic devices, light-driven micromachines, and wearable sensor systems. His research activities include: chemical synthesis of environmentally benign electrically conductive graphene-derivative inks; novel fabrication processes for printing optically transparent electrodes and circuitry on mechanically flexible substrates; design of microstructures for large area optical sheets; advanced laser material processing; and biologically-based light activated transducers. He has recently authored the books “Light Driven Micromachines” (CRC Press) and “Elastomeric Optics: Theory, Design and Fabrication” (De Gruyter GmbH).
Student roles: The student will undertake a preliminary study on digitizing historical manuscripts and books, explore volumetric segmentation and 3D surface reconstruction techniques, develop Python code to recreate text and illustrations in 3D space, and develop tests to verify algorithm performance. The research is challenging because many of the programs need to be developed from fundamental principles and involve novel algorithms. The work will be carried out with the faculty member and research collaborators who are experts at digitizing cultural artifacts.
Skills required: The applicant should have a strong programming background in Python and research interest in shape recovery from 3D coordinate data. A basic understanding of image processing, geometric modeling, and surface fitting algorithms would be an asset. The specific tasks to be completed by the student will be developed based on the applicant’s academic background and demonstrated technical strengths.
653. Registration of Magnetic Resonance Images (MRI)
Supervisor: Mehran Ebrahimi
University: Ontario Tech University (Oshawa campus)
Analysis of medical images is essential in modern medicine. With the ever increasing amount of patient data, new challenges and opportunities arise for different phases of the clinical routine, such as diagnosis, treatment, and monitoring. This research project will focus on the registration (or alignment) of patients' medical scans, more specifically magnetic resonance images (MRI).
A number of registration algorithms have been shown to be useful for various imaging modalities including MRI. Our group has access to repositories of MRI data including breast, cardiac, and brain, obtained from partner Ontario hospitals.
The project will involve developing image registration algorithms aimed at computer-aided diagnosis (CAD) and computer-assisted surgery (CAS) will be the main focus of this project.
The potential candidate will be exposed to a wide range of medical image processing algorithms and is expected to perform literature review, implement and test algorithms on medical imaging data, and prepare scientific reports.
Research area, student roles & skills
Research area: Broadly speaking, our research interests lie in the diverse area of mathematical imaging and inverse problems. Our long-term research objective is directed towards developing and validating efficient numerical methodologies for solving real-world, ill-posed inverse problems in the field of medical image processing.
Student roles: The potential candidate will be responsible for utilizing and extending our image registration tools and algorithms aimed at computer-aided diagnosis (CAD) and computer-assisted surgery (CAS).
Skills required: The potential student is required to have excellent programming skills in Matlab. The candidate may have background in different disciplines including computer science, applied mathematics, physics, electrical and computer engineering, biomedical engineering, or a related field. Excellent knowledge of calculus and linear algebra is also required. Experience working with medical imaging data and numerical optimization schemes is desirable but not required.
654. Reinforcement Learning and Alignment of Large Language Models
Supervisor: Igor Gilitschenski
University: University of Toronto (Mississauga campus)
Reasoning models have significantly improved the capabilities of modern LLMs. We are interested in making training the reasoning models more robust and accessible.
Towards this end, several potential projects are available. These include:
- Distillation of reasoning traces from larger models. Reasoning requires generation of vast amount of text, which is expensive using the largest language models. Distillation helps to transfer knowledge from larger models to smaller models, which enables cheaper inference while maintaining high quality of outputs.
- Mitigation of reward overoptimization and reward hacking. When doing reinforcement learning, models discover unintended ways to solve problems that are erroneously assigned high reward. This may happen either because of imperfect reward model (which was trained to give reward from human feedback), or imperfect verifier that doesn't account for certain unintended solutions (e.g. changing test cases instead of writing correct code). The project focuses on ways to measure and mitigate such issues.
- Measuring and improving alignment of language models. Given a dataset of human preferences or a set of rules, how do we teach the model to follow these rules and adhere to the preferences.
Research area, student roles & skills
Research area: Our research focuses on improving reinforcement learning for LLMs. We work on distilling reasoning from larger models as well as improving the robustness of the training process via mitigating issues like reward overoptimization (reward hacking). We also work on improving alignment of language models with RLHF and Constitutional AI.
Student roles: The student will conduct research using Reinforcement Learning to train Language Models. This includes defining the research question and solution ides together with the graduate student supervisors, implementing the proposed solutions, implementing tests and experiments, and presenting their findings to the lab.
Skills required: Required: - Strong programming skills in Python - Close familiarity with software engineering tools such as IDEs, remote ssh, and Git - Familiarity with machine learning such as can be obtained from introductory courses in the field
Bonus/nice to have - Familiarity with LLM finetuning - Familiarity with reinforcement and imitation learning
655. Reinforcement Learning for Adaptive Disaster Mitigation and Climate Resilience
Supervisor: Reda Snaiki
University: École de Technologie Supérieure (Montréal campus)
Location: Montreal, Québec
Start date: 2027-05-10 (flexible)
Disciplines: Computer Science, Engg-Computer, Engg-Software, Engg-Systems and Technology
Natural hazards such as extreme wind events, extreme heat, flooding, and other climate-related hazards can cause major disruptions to buildings, infrastructure systems, public services, and community wellbeing. Effective disaster mitigation requires adaptive decision-making under uncertainty, particularly when emergency managers and stakeholders must allocate limited resources, prioritize actions, and respond to evolving conditions in real time.
This project aims to develop an advanced reinforcement learning (RL) framework for disaster mitigation and resilient intervention under multi-hazard conditions. The objective is to investigate how RL-based agents can learn efficient strategies for reducing risk, limiting cascading impacts, and improving the resilience of exposed systems before, during, and after hazardous events. The framework will be designed to support sequential decision-making problems such as resource allocation, infrastructure protection, evacuation support, intervention prioritization, and recovery planning.
The project will consider applications to different hazard types, with representative case studies such as extreme wind events affecting infrastructure and extreme heat events affecting urban systems and vulnerable populations. Depending on the research scope, the study may explore deep reinforcement learning, multi-agent reinforcement learning, or constrained reinforcement learning approaches. The expected outcome is a computational decision-support framework capable of identifying adaptive mitigation strategies that enhance resilience, reduce losses, and support more effective disaster management across a range of hazard scenarios.
Research area, student roles & skills
Research area: Dr. Reda Snaiki is an Assistant Professor in the Department of Construction Engineering at ÉTS Montréal, Université du Québec. His research interests span wind engineering, structural engineering, coastal engineering, climate change, and artificial intelligence. He aims to develop advanced tools for assessing risks associated with extreme winds, in order to design more resilient infrastructure for future climate events.
Student roles: The student will contribute to the development and testing of a reinforcement learning framework for disaster mitigation and resilient intervention. Their tasks will include conducting a literature review on reinforcement learning and hazard-informed decision support, assisting with the formulation of the mitigation problem as a sequential decision-making framework, preparing and organizing relevant datasets, developing simulation environments for selected hazard scenarios, implementing and evaluating suitable RL algorithms, and analyzing the performance of the resulting strategies. The student will also assist with interpreting results, preparing visualizations, documenting findings, and contributing to research dissemination through reports and, where appropriate, conference or journal publications.
Skills required: The ideal student should have a background in one or more of the following areas: computer science, artificial intelligence, data science, civil engineering, environmental engineering, applied mathematics, operations research, or a related discipline. The student should have good programming skills, preferably in Python, and a solid interest in machine learning and computational modeling. Familiarity with reinforcement learning, optimization, stochastic simulation, or decision-making under uncertainty would be an asset. Knowledge of natural hazards, climate risks, infrastructure resilience, or urban systems would also be beneficial.
656. Reinforcement Learning for Adaptive Grasping in Soft Robotic Manipulators
Supervisor: Farrokh Janabi-Sharifi
University: Toronto Metropolitan University
Location: Toronto, Ontario
Start date: 2027-05-31 (flexible)
Disciplines: Computer Science, Engg-Computer, Engg-Electrical, Engg-Mechanical, Engg-Systems and Technology
Soft robotic manipulators offer significant advantages over conventional rigid robots due to their flexibility, compliance, and ability to safely interact with fragile or irregular objects. However, controlling soft robots is challenging because of their highly nonlinear and deformable behavior. This project aims to develop a robust reinforcement learning–based grasping framework for soft robotic manipulators using the SOFA open-source simulation platform and the Cosserat rod modeling extension. The student will help create realistic simulation environments that capture the continuum behavior of soft robots during grasping and interaction tasks. A key component of the project is the use of domain randomization techniques, where object properties, contact conditions, and simulation parameters are varied during training to improve the robustness and generalization of the learned control policies. The developed framework will be evaluated on grasping tasks involving objects with different geometries and physical properties.
The student will gain hands-on experience in soft robotics, physics-based simulation, reinforcement learning, and robot control. The project provides an opportunity to work on cutting-edge robotic technologies with applications in automation, medical robotics, and human-robot interaction. The student will work closely with graduate researchers and contribute to simulations, algorithm development, experimental analysis, and research dissemination.
Research area, student roles & skills
Research area: Soft robotics investigates flexible and compliant robotic systems capable of safely interacting with humans and delicate objects. This research focuses on developing intelligent soft robotic manipulators for adaptive grasping tasks using physics-based simulation and machine learning. By combining soft robot modeling, reinforcement learning, and advanced simulation tools, the project aims to create robotic systems that can robustly grasp objects with different shapes, sizes, and material properties in uncertain environments.
Student roles: • Work as part of a research team including graduate students and postdoctoral fellows • Develop and test reinforcement learning algorithms for soft robotic grasping • Build and run simulations in the SOFA framework • Analyze simulation and experimental results • Assist with theoretical investigations and validation studies • Prepare technical reports and research documentation • Contribute to conference and journal articles • Present research progress during team meetings and seminars
Skills required: The ideal candidate should have: • Background in robotics or mechatronics, and/or mechanical engineering, computer engineering, or related fields • Programming experience in Python and/or C/C++ • Interest in robotics, machine learning, and simulation • Familiarity with reinforcement learning or control systems is an asset • Experience with ROS, SOFA, or simulation tools is beneficial but not required • Strong problem-solving and communication skills • Ability to work independently and collaboratively in a research environment
657. Reinforcement learning for platform fulfillment services
Supervisor: Audrey Bazerghi
University: Université du Québec à Montréal
Location: Montreal, Québec
Start date: 2027-05-03 (flexible)
Disciplines: Computer Science, Engg-Computer, Engg-Industrial, Engg-Systems and Technology, Mathematics
Emerging business models in the platform economy, like those of Fulfill-by-Amazon, Shopify, and Carvana, rely on scalable decision-making algorithms to offer automated inventory management, order fulfillment, and delivery services between third-party sellers and end customers. In this project at the intersection between artificial intelligence and operations, you will start by identifying and modeling the key common characteristics of these platform fulfillment services. Then, we will aim to develop and test an agent that can control these digital platforms through reinforcement learning.
Research area, student roles & skills
Research area: I develop decision-making tools for modern supply chain management. As the decisions faced by supply chain managers grow in scale, frequency, and complexity, my research addresses these challenges by leveraging advances in automated and data-driven decision-making in the digital age. I use a combination of reinforcement learning algorithms, interviews, and game-theoretic models to provide practical recommendations.
Student roles: The student will conduct a literature review on platform fulfillment services and reinforcement learning, assist with the development of mathematical and simulation models, and implement computational experiments in Python or Julia. The student will analyze results, contribute to the evaluation of different decision-making approaches, and participate in regular research meetings. The project will provide hands-on experience at the intersection of artificial intelligence and supply chain management. Success will require strong analytical, problem-solving, and communication skills, as well as curiosity and initiative.
Skills required: A student applicant should have a background in engineering, operations research, computer science, applied mathematics, or a related field. Familiarity with programming (preferably Python or Julia) and optimization is required. The student should be comfortable working with mathematical models and computer simulations. Prior exposure to machine learning is desirable, while knowledge of reinforcement learning or supply chain management is an asset but not mandatory.
658. Reinforcement learning for solving large scale shortest path problems with resource constraints
Shortest path problems with resource constraints arise as subproblems in many large-scale applications, particularly in vehicle and crew scheduling problems across air, urban, and rail transportation systems. In these contexts, resource constraints are used to model complex business rules, which may involve a large number of dimensions and operational restrictions. When such problems are solved using standard dynamic programming approaches, they typically suffer from the well-known curse of dimensionality, leading to a dramatic increase in computational complexity. As a result, solving large-scale instances remains a major bottleneck both in research and industrial applications. To address this challenge, we propose a reinforcement learning–based approximate dynamic programming approach to efficiently tackle large and high-dimensional problems. This perspective may open new opportunities for the development of scalable optimization methods with significant impact for both academia and industry.
Research area, student roles & skills
Research area: Operations research, large scale optimization, supply chains
Student roles: - Algorithm design - Coding - Testing - writing a paper
Skills required: RL, approximate dynamic programming, operations research, coding languages
659. Reliability-Aware Workload Generation for LLM Serving Systems
Supervisor: Yasaman Amannejad
University: Mount Royal University (Calgary campus)
Large Language Model applications, such as chatbots, coding assistants, and AI-powered APIs, are increasingly used in real-world settings where user demand can change quickly. Unlike traditional software systems, LLM serving systems must handle requests with highly variable input lengths, output lengths, conversation patterns, and arrival rates. Some requests are short and simple, while others require long responses or extended multi-turn conversations. At the same time, user requests may arrive in bursts, creating sudden pressure on computing resources such as GPUs and memory.
This project investigates how realistic workload generation can help evaluate the reliability of LLM serving systems. Instead of only measuring average latency or throughput, the project focuses on understanding when and why systems become slow, unstable, or unable to serve requests successfully.
Students working on this project may analyze workload traces, design workload-generation strategies, and run experiments using open-source LLM serving frameworks. The work may involve measuring metrics such as response time, time to first token, throughput, memory usage, request failures, and service-level objective violations. The project can include both data analysis and systems experimentation, depending on the student’s background and interests.
This project is suitable for students interested in machine learning systems, cloud computing, performance engineering, software reliability, or experimental research. Strong applicants should have programming experience and be comfortable learning new tools. Experience with Python, Linux, data analysis, or machine learning frameworks is helpful, but not required. The project provides an opportunity to work on a timely research problem at the intersection of AI infrastructure, workload modeling, and dependable software systems.
Research area, student roles & skills
Research area: My specialized research area is software engineering for dependable and data-driven systems, with a focus on software reliability, performance engineering, and AI-enabled systems. My research investigates how software systems behave under realistic and changing conditions, and how data analysis, testing, monitoring, and workload characterization can be used to improve their reliability and performance. I am particularly interested in experimental and applied research that connects software engineering methods with emerging computing domains, including machine learning systems, cloud-based services, and AI infrastructure. My work often involves analyzing real-world data, designing evaluation methods, and developing practical tools to support dependable
Student roles: The student will contribute to the project by reviewing related work, analyzing workload traces, designing small experiments, and evaluating how different workload patterns affect LLM serving reliability. They will help generate realistic test workloads, run experiments using open-source tools, collect performance metrics, and summarize findings.
The student is expected to take an active role in problem solving, documenting results, and discussing research directions with the supervisor. Depending on their background, they may focus more on data analysis, system experimentation, or tool development.
Skills required: Students should have basic programming experience, preferably in Python, and be comfortable learning new tools. Familiarity with Linux, Git, command-line tools, and data analysis libraries such as pandas or NumPy is helpful.
A background in software engineering, machine learning, cloud computing, operating systems, or distributed systems would be an asset, but prior experience with LLM serving is not required. Strong candidates should be curious, self-motivated, and interested in experimental research, performance measurement, and understanding how AI systems behave under realistic workloads.
660. Remotely Piloted Aircraft System Perception and Control
Supervisor: Geoff Fink
University: Thompson Rivers University (Kamloops campus)
This research project aims to improve the autonomy of Remotely Piloted Aircraft Systems (RPAS) by developing novel perception and control algorithms. More specifically, we will enhance the aircraft's position and velocity estimates within dynamic airspace environments. We will do this with two main approaches. The first approach is to improve mapping and visual-inertial sensor fusion by including aircraft and atmospheric dynamics in simultaneous localization and mapping (SLAM). The second approach uses a team consisting of a ground-based robot (quadruped) and an aerial robot (quadrotor) to perform coordinated state estimation.
Research area, student roles & skills
Research area: Remotely Piloted Aircraft Systems (RPAS) are increasingly used for a wide range of both indoor and outdoor applications. They have remarkable flight capabilities; however, most are still remotely piloted because they lack a high level of autonomous decision-making. RPAS must be robust while operating in a wide variety of environmental conditions. The conditions may vary in meteorology and physical properties, such as wind patterns, turbulence, or atmospheric density. Reliable state estimation is a significant aspect for the deployment of RPAS because most flight path planners and control strategies rely on an accurate estimate of the position, altitude, and velocity of
Student roles: The student will work as part of an interdisciplinary team on perception and control algorithms for Remotely Piloted Aircraft Systems (RPAS). This includes analysing and implementing computer vision algorithms, implementing control algorithms, improving the RPAS software framework, simulating RPAS, running experiments with RPAS, and writing reports.
Skills required: This is a large project where students can focus on any one or more of the following areas: control, software engineering, robotics, simulation, computer vision, or computer/electronic engineering.
661. ResOps: Designing an Automated Platform for Research Management
Academic research laboratories frequently suffer from severe operational inefficiencies due to a reliance on fragmented, manual administrative processes. Faculty and student researchers routinely divert substantial cognitive energy away from actual engineering tasks to manage project milestones, track equipment allocations, compile compliance audits, and navigate unstructured onboarding workflows. This administrative friction introduces operational bottlenecks that delay technical velocity and degrade research quality. This innovative project addresses this overhead by developing an automated, data-driven Research Operations (ResOps) Management Engine specifically designed to optimize laboratory productivity and streamline project lifecycles.
The intern will audit our laboratory’s existing operational workflows, resource allocations, and administrative touchpoints to map out systemic friction areas. Moving beyond generic project management tools, the student will engineer a centralized, localized platform that integrates data tracking, milestone verification, and resource dependency mapping into a single cohesive interface. The core research focus centers on process automation and deterministic decision support: building intelligent pipelines that automatically track project health metrics, signal resource bottlenecks, and deliver clear, structured data to support laboratory leadership decisions.
From a research entrepreneurship perspective, this project serves as a live internal incubation run. The intern will design the system using modern web technologies and modular architectures, ensuring that the resulting engine is completely independent of external proprietary software. The platform will include structured, metric-driven onboarding workflows for new lab members, systematically reducing their time-to-competence and accelerating their integration into active projects.
The final deliverable is a production-grade, self-sustaining software application complete with installation blueprints and administrator documentation. By automating routine governance, this project leaves behind a permanent operational asset that transforms how our lab manages its research portfolio, maximizing deep cognitive work and setting a new reference standard for academic efficiency.
Research area, student roles & skills
Research area: Our laboratory bridges the gap between academic innovation and real-world adoption across trustworthy technology, security operations, and human readiness. A critical dimension of our program is Research Operations (ResOps) optimization. We study the systemic bottlenecks and administrative friction that pull researchers away from active scientific discovery. By analyzing process workflows, automation pipelines, and data-driven management frameworks, we develop reproducible tools to optimize laboratory productivity. We build deterministic decision-support systems that streamline project lifecycle management and researcher onboarding, ensuring our team maximizes focus on high-fidelity technical contributions while eliminating operational inefficiencies.
Student roles: The selected intern will function as a Product Engineer and Workflow Analyst, taking complete ownership of designing, building, and launching our internal ResOps management platform. Over the 12-week internship, the student will drive a rapid, end-to-end development lifecycle:
Initially, the student will interview laboratory researchers and analyze current operational workflows to catalog administrative bottlenecks and data dependencies. They will translate these findings into a structured schema database design, ensuring the system can accurately model complex lab dynamics like multi-project resource allocation and timeline milestones. The core engineering task involves writing the backend logic and building an intuitive front-end interface that unifies these disparate data streams into a centralized cockpit for laboratory management.
A critical milestone involves building the automated metric and reporting modules. The student will write scripts that aggregate project progress data, generate automated alert flags for slipping deadlines, and construct interactive dashboards that visualize laboratory capacity and output in real time. They will also build the automated onboarding pipeline designed to fast-track training tracking for future cohorts.
In alignment with our lab’s emphasis on sustainable adoption, the student will test the system with live laboratory members, gathering feedback to refine usability and minimize interface friction. They will package the code to professional enterprise standards, including Docker container configurations, API schemas, and explicit user deployment guides.
Skills required: Applicants should be pursuing a degree in Computer Science, Software Engineering, Information Systems, or Management of Technology. Strong proficiency in full-stack web development (Python/Django/FastAPI or Node.js/TypeScript/React) and experience with relational databases (PostgreSQL/SQLite) are required. Familiarity with process mapping, API integration, or data visualization frameworks is highly desirable. We seek organized, system-oriented students who are passionate about productivity engineering, workflow optimization, and clean software architecture, and who want to build a polished, production-ready operational platform that directly scales the capacity and output of an active research organization.
662. Resolution-agnostic Machine Learning Approaches for Medical Image Analysis
Deep learning has transformed how medical images such as X-rays, CT scans, and MRIs are analyzed. However, trained models have a major limitation: they usually only work with images of a specific size and resolution, which is a challenge in real-world setups where data comes in many different sizes depending on the scanner/acquisition system used. Within this project, we will develop new approaches to solve this problem by creating models that work with any image size or resolution without needing to change the original data first. We aim to achieve this by using and advancing state-of-the-art image representation techniques like implicit neural representations. Our research in this area currently focuses on applications in neuroimaging and retinal imaging.
Research area, student roles & skills
Research area: I am a computer scientist and an expert in machine learning-based medical image analysis. My research centers around the development of machine learning solutions for health data science and precision medicine problems. I am specifically interested in developing and advancing machine learning methods that accurately model the complex dynamics and variations of normal or pathological processes in the human body by integrating and combining diverse, large-scale medical data (e.g., images, clinical data, text reports).
Student roles: The student will work on implementing and testing deep learning models for medical image analysis tasks in the context of implicit neural representation. The student will work in close collaboration with graduate students and other trainees and will be exposed to the newest techniques in deep learning-based medical image analysis. The student will utilize state-of-the-art Python-based deep learning frameworks such as PyTorch and MONAI to implement, train, and test the developed models on powerful hardware (i.e. NVIDIA RTX4090 GPUs).
Skills required: The student for this project should have: - A background in Biomedical Engineering, Computer Science, Electrical Engineering, or a closely related field - Very good math skills - First experience in machine learning-based medical image analysis or a related field (e.g., computer vision, machine learning) - Some hands-on experience with modern software frameworks for deep learning like PyTorch - Programming skills in Python
663. Resource Allocation in Split and Federated Models
Supervisor: Yasaman Amannejad
University: Mount Royal University (Calgary campus)
In this project, we will build and study split and federated learning models under various client resource configurations. We have already developed the codebase for the models to be studied. The ideal intern will use the existing codebase, extend it if needed, run experiments, collect experimental data, and analyze different configuration settings to identify optimal configurations for the clients involved in training.
Research area, student roles & skills
Research area: My research is in the intersection of software engineering and performance engineering. In this project, we will build and study the split and federated learning models under various client resource configurations.
Student roles: The student will read and understand the existing codebase, run experiments, collect data, analyze the results, and write a report. The student may also need to modify or extend the code as needed.
Skills required: Attention to detail Programming experience with Python Comfort with terminal and command-line environments in Linux Strong analytical and communication skills
664. Resource Management in Hybrid Classical and Quantum Wireless Networks
Supervisor: Ekram Hossain
University: University of Manitoba (Winnipeg campus)
Future wireless systems are expected to evolve towards hybrid classical-quantum networks, in which BSs integrate both conventional communication capabilities and quantum functionalities. Many previous studies treated the communication medium as a shared resource that is orthogonally partitioned between classical and quantum users, and optimize system performance using multi-objective formulations. An important research problem in this context is designing joint resource management strategies for tightly integrated hybrid networks, where classical and quantum components coexist and interact more directly rather than being treated as isolated subsystems. The project will attempt to answer the following questions:
1) Non-orthogonal channel sharing. If classical and quantum signals simultaneously occupy the same transmission medium, how should cross-domain noise and interference be modeled? In particular, how can the power and spectral characteristics of classical signals be controlled to protect fragile quantum states while ensuring that both domains satisfy their QoS requirements?
2) Base station (BS) resource allocation. How should the shared resources such as power, bandwidth, scheduling time, memory, and processing units in a classical-quantum BS be allocated between classical and quantum tasks to maximize overall network performance?
3) Unified performance metrics. What new performance metrics can capture the efficiency, reliability, and latency of hybrid classical-quantum networks, beyond evaluating classical and quantum systems as independent subsystems? Since performance of one domain may directly constrain or degrade the other, separate optimization of classical and quantum metrics can be misleading. Therefore, unified metrics that jointly reflect classical QoS and quantum feasibility will need to be developed for joint resource management and effective system design.
Research area, student roles & skills
Research area: My research is in the area of wireless communications and sensing systems, qunatum information systems, and machine learning. More information about our research can be found at: https://home.cc.umanitoba.ca/~hossaina
Student roles: The student will work with a Post-Doctoral Fellow who will assist the student in this project. The student will need to do programming to solve some optimization problems and also develop some simulation models.
Skills required: The student needs to have background on cellular wireless networks, basics of quantum communication, and optimization techniques. Experience on MATLAB and/or Python programming will be required. Background on optimization and machine learning will be a plus.
665. RoboGames: Educational Game Development for Computer Science
Supervisor: Jeremy Bradbury
University: Ontario Tech University (Oshawa campus)
This research project focuses on enhancing the learning of students within first year computer programming courses. These courses often have high enrollments and have students with a wide range of programming experience. To engage these diverse cohorts, we advocate the use of alternative interactive learning tools like educational games to assist in learning new programming concepts. The use of game-based learning has been demonstrated beneficial in Computer Science and over the past ten years we have developed games to enhance the learning of different programming concepts including program understanding (RobotON) debugging (RoboBUG), multi-threaded programming (ThreadedPaws), and testing (TestMiner). The goal of this project will be to continue development of these existing games as well as to develop new educational games that can enhance student learning and engagement in computer science.
Research area, student roles & skills
Research area: My research interests span software development and computer science education. In particular I focus on applying AI to software development, as well as using AI to support personalised learning in educational games.
Student roles: Successful applicants will work on developing open source educational games for use in first-year Computer Science courses. Activities will include design and development of games, testing & debugging of games as well as the deployment of games on web servers.
The research project can focus in one of these two areas, we are currently researching:
(1) Vision-Language-Action models for robotics. We study how VLA-style policies can be adapted for real robotic control, with focus areas including (i) robustness to visual and environment variations, (ii) generalization across tasks, objects, and embodiments, and (iii) reinforcement-learning-based post-tuning (e.g., policy improvement in simulation or offline RL from robotic experience) to improve closed-loop performance beyond imitation learning.
(2) Simulation and world models for robotics. We explore how video-generation and video-prediction models can be used as world models for planning, evaluation, and training. Topics may include using generative video models to augment simulation data or generate counterfactual rollouts, extract policies from video generation models, and building scalable evaluation protocols that stress-test policies under diverse simulated conditions.
Research area, student roles & skills
Research area: The Toronto Intelligent Systems Lab conducts research in robot learning, with an emphasis on learning robust and generalizable visuomotor policies that can operate under distribution shift. We are particularly interested in leveraging large pretrained models and modern reinforcement learning to build embodied agents that can learn from diverse data and improve through post-training and interaction.
Student roles: The student will contribute to the full research loop: reading key papers, setting up and extending existing codebases, implementing and training models, designing and running experiments (primarily in simulation), and analyzing results. The student will work closely with graduate mentors, participate in regular research meetings, and help document findings; strong projects may lead to a workshop/conference submission.
Skills required: 1. Strong Python programming and experience with PyTorch 2. Solid fundamentals in machine learning and deep learning (e.g., Transformers, representation learning) 3. Familiarity with reinforcement learning and/or imitation learning is highly desirable (but not strictly required) 4. Experience with robotics simulators (e.g., MuJoCo/Isaac/Gym-style environments) or video models is a plus
Enabling robots to acquire new skills efficiently remains a significant challenge. Traditional approaches that rely on collecting large amounts of robot-specific data are resource-intensive and impractical for solving large amounts of real-world problems. Leveraging human or animal video datasets for robot learning has been explored, such as imitating the hand pose trajectories of the human demonstrators or tracking the skeleton of a quadruped animal.
However, a few key obstacles remain: (1) The fundamental mismatch between human and robot embodiments. Consequently, directly imitating online videos is often infeasible, as robots cannot reproduce the same states or transitions within their own configuration space. (2) While prior work requires that the demonstrator and the robot have identical viewpoints, the public videos are often unstructured, and demonstrations rarely capture full-body movements clearly.
To address the embodiment gap, we propose learning a latent alignment mapping between the demonstrator and the robot’s own latent representation of feasible actions. We will utilize publicly available datasets for manipulation in the kitchen for learning manipulation tasks as our first experiments. Subsequent analysis will scale this to the locomotion task from the animals walking in the wild dataset.
Research area, student roles & skills
Research area: Robot learning for manipulators and quadruped robots
Student roles: The student will be responsible for testing the proposed method with publicly available datasets and eventually deploying the learned models on real robotic systems.
Skills required: Required skills: Knowledge/experience in robot manipulation, forward/inverse kinematics and dynamics. Modern approach to supervised learning. Good to have: Experience with robotic hardware
668. Robotic Software Framework
Supervisor: Geoff Fink
University: Thompson Rivers University (Kamloops campus)
Location: Kamloops, British Columbia
Start date: 2027-05-03 (flexible)
Disciplines: Computer Science, Engg-Computer, Engg-Software, Engg-Systems and Technology, Engineering
This project advances an existing robotic software framework specifically designed to accelerate the deployment of autonomous RPAS and ground-based robotic systems. The framework integrates custom-built middleware solutions with standardized communication interfaces, enabling researchers to rapidly prototype and test perception algorithms, control strategies, and multi-robot coordination systems. The project focuses on enhancing the unified development environment to support real-time data processing, distributed computing architectures, and seamless hardware abstraction layers for both aerial and terrestrial robotic platforms.
Research area, student roles & skills
Research area: Our research in custom robotic software frameworks specializes in creating comprehensive, modular architectures that standardize robotic system development. This research area encompasses middleware design, real-time communication protocols, and cross-platform integration solutions that enable seamless interoperability between diverse hardware components and control algorithms. The field emphasizes developing reusable software components, fault-tolerant architectures, and standardized APIs that facilitate rapid prototyping while maintaining system reliability and performance optimization for autonomous operations.
Student roles: The student will work as part of an interdisciplinary team on developing and enhancing the robotic software framework for autonomous systems. This includes analyzing and implementing middleware components, designing and testing API interfaces, integrating hardware abstraction layers, developing real-time communication protocols, conducting software performance analysis, testing framework compatibility across different robotic platforms, and documenting software architecture and user guides. The student will collaborate closely with researchers working on perception and control algorithms to ensure seamless integration between the framework and robotic applications.
Skills required: The ideal student should have a strong foundation in software engineering with proficiency in C++ and Python. Experience with robotics middleware frameworks (ROS/ROS2), real-time systems programming, and distributed computing architectures is highly desirable. Knowledge of software design patterns, API development, and version control systems is essential. Familiarity with robotic hardware interfaces, sensor integration, and multi-threaded programming will be advantageous. The student should demonstrate problem-solving abilities, attention to software quality and testing practices, and the ability to work independently on complex software architecture challenges.
669. Robust Aggregation for Multi-Node LLM Inference
Supervisor: Sara Rouhani
University: University of Calgary
Location: Calgary, Alberta
Start date: 2027-06-10 (flexible)
Disciplines: Computer Science, Engg-Computer, Management Information Systems
As large language models are increasingly deployed in distributed and multi-party settings, ensuring the reliability and integrity of their outputs becomes a critical challenge. In such settings, a single model instance may not be sufficient to guarantee trustworthy responses which could be due to resource constraints, fault tolerance requirements, or the need for redundancy. Distributed LLM inference, where multiple model instances collaborate to produce a final output, offers a promising direction, but introduces new challenges around consistency, coordination, and adversarial robustness.
This project investigates robust aggregation mechanisms for distributed LLM inference in the presence of adversarial or unreliable nodes. In a multi-node inference setting, responses from multiple LLM instances must be aggregated into a single reliable output. This process that becomes challenging when some nodes behave maliciously, are compromised, or return inconsistent results, degrading overall system trustworthiness.
The goal is to design and evaluate an aggregation layer that improves the reliability and robustness of distributed inference. This involves exploring strategies to detect inconsistent or outlier responses and select or synthesize outputs that reflect consensus across nodes. The work involves understanding how different aggregation approaches perform under varying levels of adversarial pressure, and what trade-offs emerge between robustness, accuracy, and computational overhead. The project will also examine whether aggregation strategies can be made adaptive, adjusting their behavior based on observed node reliability over time.
The system will be evaluated under a range of adversarial scenarios, including random incorrect outputs, coordinated malicious responses, and subtle misinformation attacks. Beyond adversarial robustness, the project will also consider practical deployment constraints such as latency and scalability, ensuring that the proposed mechanisms remain viable in real-world distributed settings. Key evaluation metrics will include response accuracy, attack resilience, latency, and system efficiency.
Research area, student roles & skills
Research area: My research lies at the intersection of distributed systems and cybersecurity, focusing on advancing security and trust in decentralized environments. I study how decentralized paradigms enable secure, reliable, and scalable coordination in multi-party systems, particularly for collaborative and decentralized machine learning (e.g., federated and split learning), addressing challenges such as robustness, privacy, and verifiability. I also explore agentic AI systems and protocols such as MCP for automated vulnerability detection and threat analysis. In parallel, I investigate blockchain infrastructures, including Layer 2 solutions, interoperability, and decentralized identity, as foundations for secure and trustworthy decentralized applications.
Student roles: Intern 1 — Distributed Inference System & Environment Setup: Design and implement a distributed LLM inference framework with multiple coordinated inference nodes Develop infrastructure for orchestrating requests across nodes and collecting their outputs efficiently Implement logging, data collection, and preprocessing pipelines for model responses and system metrics Support the generation of datasets under normal and adversarial conditions for evaluation purposes
Intern 2 — Aggregation & Robustness Framework: Design and implement aggregation mechanisms to combine outputs from multiple LLM instances Develop techniques to detect inconsistent, outlier, or adversarial responses across nodes Implement and compare aggregation strategies (e.g., voting, ranking, confidence-based, adaptive methods) Define and compute evaluation metrics (e.g., accuracy, robustness, latency, efficiency, attack resilience) Analyze system behavior under different adversarial scenarios and characterize trade-offs between reliability and performance
Skills required: Candidates should have a solid foundation in computer science or a related field, with strong programming skills in Python. Familiarity with machine learning concepts and experience working with large language models is also essential. A basic understanding of distributed systems concepts, including node coordination, fault tolerance, and consistency, is expected. Candidates should also have familiarity with cybersecurity concepts, particularly adversarial settings and threat modeling.
670. Robust Multimodal Fusion under Missing Modalities
Supervisor: Abedalrhman Alkhateeb
University: Lakehead University (Thunder Bay campus)
In real-world hospitals, data is never complete—an MRI might be missing, a blood test result could be corrupted, or a patient's record may have gaps. Current multimodal AI models often fail catastrophically under these conditions, limiting clinical trustworthiness.
This project will develop and evaluate a robust multimodal fusion model that maintains predictive performance even when some data modalities are missing. The student will:
1-Implement a baseline multimodal fusion model (e.g., late fusion with modality-specific encoders) on a public biomedical dataset such as BraTS (brain tumor imaging + clinical data) or a multi-omics cancer dataset.
2-Implement two robustness strategies and compare them against the baseline:
* Modality dropout training: Randomly dropping entire modalities during training to simulate missing data at test time
* Disentangled latent space inference: A lightweight VAE-based module that infers missing modality representations from the available ones
3- Systematically evaluate performance under controlled missing data scenarios: random missing, structured missing (e.g., always missing a specific modality), and varying missing rates (10%, 25%, 50%).
The outcome will be a clear benchmark and practical recommendations for building clinically reliable multimodal systems that degrade gracefully, not catastrophically.
Research area, student roles & skills
Research area: My lab builds clinically trustworthy AI systems that perform reliably in messy real-world conditions. We focus on robustness to missing data, noise, and distribution shifts. Our research develops novel architectures for multimodal fusion that maintain performance even when some data streams are unavailable. We aim to bridge the gap between academic benchmarks and hospital deployment.
Student roles: The intern will take ownership of the entire robustness benchmarking pipeline, from data preprocessing to final reporting, with regular guidance from the PI and a senior lab member.
Specific duties by phase:
*Weeks 1-2 (Setup & Data):*
- Download and explore a public multimodal dataset (e.g., BraTS 2024 or a TCGA multi-omics subset)
- Preprocess the data and create consistent train/validation/test splits
- Implement missing data injection functions for various patterns and rates
*Weeks 3-6 (Baseline & First Strategy):*
- Implement a standard multimodal late-fusion baseline model in PyTorch
- Train and evaluate the baseline under all missing scenarios
- Implement modality dropout training and repeat the evaluation
*Weeks 7-10 (Second Strategy & Analysis):*
- Implement a lightweight disentangled VAE module for missing modality inference
- Train and evaluate this model under all missing scenarios
- Compare all three approaches (baseline, dropout, VAE-inference) using metrics: accuracy, AUROC, and performance degradation slopes
- Write a 4-5 page technical report summarizing methods, results, and recommendations
- Prepare a 10-minute final presentation for the lab
- Submit clean, documented code to the lab's GitHub repository
The student will work independently but will have weekly one-on-one meetings with the PI and access to a senior PhD student for technical troubleshooting.
Skills required: - Strong Python programming with PyTorch. - Experience building and training deep learning models (CNNs, MLPs, or VAEs). - Familiarity with data preprocessing (Pandas, NumPy). - Understanding of train/val/test splits and basic evaluation metrics. - Interest in robustness, missing data, or representation learning. - Biomedical data experience is a plus.
671. Robust computer vision models for traffic scenarios
Foundation models trained on Internet-scale datasets are being used in many applications, including assistive and autonomous driving. While these models perform well across a wide range of scenarios, they are often not robust to even relatively minor change in the environment. For example, change of lighting conditions, configuration of objects in the scene, or instructions, can have large effects on the output of the model. For obvious reasons, such unexpected reactions are undesirable in traffic environments.
In our past work, we found many cases of robustness issues for foundation models. For example, we experimented with adding, removing, and modifying objects in the traffic scene as well as phrasing of the command, all of which had measurable effects on model performance. However, what causes the problems and whether they can be solved remains an open question.
The goal of this project is to find what is causing these variations in model behavior. Is it data, architecture, training, or combination of these factors? Answering these questions would require one or more of the following:
- analysis of the training data to identify potential biases;
- examining model parameters and how they change with different inputs;
- modifying the models (e.g. by fine-tuning) to improve their robustness.
Research area, student roles & skills
Research area: My research area is building computer vision systems inspired by human vision and cognition for various applications, with focus on intelligent transportation, assistive driving systems, and autonomous driving. My main goal is to advance human-centered traffic safety by developing human-compliant models and identifying open problems through application of domain knowledge, extensive data analysis, and systematic benchmarking.
I am interested in understanding decision-making of pedestrians (e.g. how and when they decide to cross or yield) and visual attention of the drivers (e.g. what do drivers look at in the busy traffic scene).
Student roles: The student will work on the project under my supervision. I will provide them with a set of academic papers that will help them learn about past relevant work, a source code repo containing our ongoing work related to the project, and access to the GPU server.
The student will; - analyze available experiment data and literature to identify potential sources of robustness issues; - design and run additional experiments to confirm that the identified sources cause robustness problems; - based on the findings, propose and implement a solution; - describe their work in a conference paper format.
Skills required: The minimum required skills/background are: - strong programming skills (especially in Python); - interest in research, desire to learn and explore new problems.
Any of the following will make your application stand out: - research experience (e.g. working in research labs, personal projects, publications); - experience with computer vision (e.g. courses, personal projects); - experience with PyTorch (or other deep learning frameworks); - experience with data analysis (e.g. pandas) and visualization; - exposure to cognitive science/psychology (e.g. through courses or personal interest)
672. SE-INTEL: Agentic Code Intelligence for Scalable Software Engineering via Graph-RAG, Quality Analysis, and Probabilistic Estimation
This project develops SE-INTEL, a holistic agentic software engineering intelligence framework integrating Graph-RAG-based code assistance, graph-augmented defect prediction, combinatorial test optimization, and probabilistic effort estimation. Conventional AI code assistants rely on static context windows and shallow retrieval, limiting correctness and architectural coherence. SE-INTEL addresses this through four tightly integrated research pillars. The first pillar, Enhanced RAILS, transforms code generation into a secure agentic intelligence system by integrating Graph-based RAG, dynamic retrieval, and verification-driven generation. Code assistance is framed as probabilistic reasoning over a hybrid knowledge base combining vector embeddings, dependency graphs, real-time documentation, and versioned metadata. Graph-RAG captures structural API relationships while dynamic retrieval adapts context to code edits. The architecture integrates supervised fine-tuning, reward modeling, PPO-based RLHF, LoRA adaptation, and active learning, supported by multi-layer security and compilation checks. The second pillar, SYNAPSE, embeds code quality analysis within a graph-theoretic probabilistic intelligence engine that simultaneously predicts defects, ranks refactoring priorities, and exposes architectural anti-patterns across the class-and-package hierarchy. Object-oriented metrics including WMC, CBO, LCOM, and LOC are integrated with graph convolutional networks to propagate defect risk across dependency structures. The third pillar, OptiTest, addresses regression test suite optimization through integer programming formulations that select minimal test subsets satisfying coverage and budget constraints, coupling model-driven optimality with data-driven prioritization from historical execution records. The fourth pillar, ProbaSAC, advances software effort estimation using lognormal Bayesian modeling, confusion-matrix-aware cost calibration, and misclassification-adjusted baselines that explicitly account for label noise in defect datasets. All pillars are validated through cross-project benchmarking, controlled ablations, and statistical significance testing. Results will be disseminated through top-tier venues.
Research area, student roles & skills
Research area: This research area advances equitable, explainable, and agentic software engineering intelligence by unifying retrieval-augmented generation, graph-augmented learning, combinatorial optimization, and probabilistic modeling across the full software development lifecycle. The program spans four interconnected pillars: agentic code intelligence via secure Graph-RAG and dynamic retrieval (Enhanced RAILS); graph-augmented software quality analysis and structural risk assessment (SYNAPSE); combinatorial and constraint-driven regression test optimization (OptiTest); and probabilistic effort estimation with misclassification-aware baselines (ProbaSAC). Together, these pillars constitute a lifecycle-spanning intelligence framework for modern software engineering applicable in enterprise and resource-constrained development environments.
Student roles: The student will contribute substantially to the implementation and evaluation of the SE-INTEL framework across selected pillars based on their background and project priorities. In the Enhanced RAILS pillar, primary responsibilities will include implementing the hybrid Graph-RAG retrieval engine combining vector similarity search and graph traversal over code dependency structures, integrating this with a fine-tuned code LLM, and validating the system on code generation and repair benchmarks. The student will implement and tune the RLHF training pipeline including supervised fine-tuning, reward modeling, and PPO optimization with LoRA adaptation, and evaluate output correctness and architectural coherence against static RAG and LLM-only baselines. In the SYNAPSE pillar, the student will assist in constructing the code quality graph, implementing graph convolutional network layers for defect risk propagation, integrating object-oriented software metrics (WMC, CBO, LCOM, LOC), and evaluating defect prediction accuracy and refactoring recommendation quality on open-source Java project benchmarks drawn from repositories such as the PROMISE defect dataset. In the OptiTest pillar, the student will implement integer programming formulations for test suite minimization and prioritization, evaluate coverage-preserving subset selection against greedy and random baselines, and analyze budget-accuracy trade-offs across multiple test histories. In the ProbaSAC pillar, the student will implement lognormal Bayesian effort estimation models, develop confusion-matrix-aware cost calibration procedures, and benchmark against standard COCOMO and use-case-based estimation baselines on cross-project effort datasets. Beyond technical contributions, the student will engage in knowledge mobilization through drafting manuscript sections, preparing conference presentations, and maintaining reproducible documentation of all experimental pipelines and results, developing both academic writing and professional communication skills. This project will equip the student with advanced expertise in agentic AI, graph learning, RAG systems, and software analytics, alongside critical soft skills in research design, collaboration, and technical communication, preparing them for careers in AI-augmented software engineering or applied machine learning research.
Skills required: The student should have foundational knowledge in software engineering and machine learning, with solid Python programming skills. Familiarity with object-oriented software design, code metrics, and version control systems is important. Basic understanding of graph algorithms, natural language processing, or retrieval-augmented generation is an asset. Prior experience with testing frameworks, defect datasets, or software project data is beneficial. Knowledge of deep learning libraries such as PyTorch and experience with graph neural networks or transformer-based models will be advantageous. The student should be analytically rigorous, comfortable working with large codebases and software repositories.
673. Scalable Federated Learning on Large-Scale Data: Performance, Privacy, and Communication Trade-offs
This project investigates how federated learning (FL) systems perform when scaled to large datasets distributed across many clients. Using frameworks such as Flower or TensorFlow Federated, the student(s) will design and implement a distributed training pipeline where data remains decentralized and only model updates are shared. The study focuses on three core aspects: (1) performance under realistic conditions such as non-IID data distributions, (2) the impact of lightweight privacy mechanisms (e.g., noise injection) on model accuracy, and (3) communication efficiency as the number of clients and data volume increase. Through systematic experiments and analysis, the project aims to quantify the trade-offs between scalability, accuracy, privacy, and communication cost, providing practical insights into deploying federated learning in real-world big data settings.
Research area, student roles & skills
Research area: Big Data Analytics
Federated Learning
Privacy-Preserving Data Mining and Machine Learning
Student roles: The student is expected to act as a junior researcher and system developer, responsible for the end-to-end design, implementation, and evaluation of a scalable federated learning system. 1. System Development • Design and implement a federated learning pipeline using tools such as Flower or TensorFlow Federated • Develop modular, well-documented code for: • Client-side training • Server-side aggregation (e.g., FedAvg) • Simulate multiple clients and distributed datasets 2. Data Engineering • Select and preprocess a suitable large-scale dataset • Partition data across clients (IID and non-IID scenarios) • Ensure reproducibility of data splits and experiments 3. Experimental Design & Execution • Design controlled experiments to evaluate: • Model performance • Scalability (number of clients, data size) • Communication cost • Run experiments systematically and log results 4. Analysis & Interpretation • Analyze results using appropriate metrics (accuracy, convergence, communication rounds, etc.) • Generate visualizations (plots, tables) • Interpret trade-offs between performance, privacy, and efficiency 5. Extension Implementation • Implement at least one enhancement, such as: • Communication optimization (e.g., reduced rounds, compression), or • Basic privacy mechanism (e.g., noise addition) 6. Documentation & Reporting • Maintain clean, well-structured code with documentation • Prepare a technical report describing: • Problem formulation • Methodology • Experimental results • Key insights 7. Professional Practices • Use version control (e.g., Git) • Follow reproducible research practices • Manage time effectively to meet project milestones
Skills required: Programming in Python Basic machine learning knowledge Familiarity with tools such as PyTorch or similar frameworks Analytical thinking and problem-solving
674. Scalable RL Solutions for Future Wireless Networks
This research project focuses on developing scalable reinforcement learning (RL) solutions for future 6G wireless networks. While RL has demonstrated strong potential for optimizing resource allocation, scheduling, and network management, its scalability remains a major challenge in realistic large-scale deployments. As the number of user equipments (UEs) increases from a few tens to hundreds or even thousands, conventional RL approaches often suffer from exponentially growing state and action spaces, leading to prohibitive training complexity and poor generalization.
The objective of this project is to design RL algorithms that can scale efficiently with network size while maintaining high performance. In particular, the research investigates methods that can generalize across different network configurations and user populations, reducing or eliminating the need to retrain the model whenever the number of users, traffic patterns, or network conditions change. By developing scalable, adaptive, and generalizable learning frameworks, the project aims to enable practical AI-driven optimization for next-generation 6G wireless systems.
Research area, student roles & skills
Research area: I am working on future wireless networks. I develop and propose innovative solutions to solve future challenges related to wireless communications. I work on scheduling algorithms, mathematical optimization, algorithm design, ML/RL algorithms.
Student roles: The student will play an active role in a research project focused on developing scalable reinforcement learning (RL) solutions for next-generation 6G wireless networks. The primary objective is to investigate how AI-driven resource management and decision-making algorithms can be designed to operate efficiently in large-scale and highly dynamic wireless environments.
The student will begin by conducting a comprehensive review of the scientific literature related to wireless communications, reinforcement learning, network optimization, and scalable machine learning techniques. Based on this review, the student will help identify key challenges and research opportunities associated with applying RL to large wireless networks, where the number of users, devices, and network elements can vary significantly over time.
The student will contribute to the development of mathematical models and algorithmic frameworks aimed at improving the scalability and generalization capabilities of RL-based solutions. Responsibilities will include implementing algorithms using programming languages such as Python and Julia, designing simulation environments, conducting experiments, and evaluating the performance of proposed methods under various network scenarios. Particular attention will be given to approaches that can adapt to changes in network size and conditions without requiring costly retraining procedures.
The student will also be responsible for analyzing experimental results, comparing different approaches, and identifying strengths, limitations, and potential improvements. Research findings will be documented through technical reports, presentations, and scientific publications. The student will participate in regular meetings with the research team, present progress updates, and collaborate with other researchers and graduate students.
This role provides an excellent opportunity to gain experience at the intersection of wireless communications, artificial intelligence, optimization, and scientific computing. Through this project, the student will develop strong research, programming, analytical, and problem-solving skills while contributing to innovative solutions for the future of intelligent and autonomous 6G networks.
Skills required: The candidate should have a background in Electrical Engineering, Computer Engineering, Computer Science, or a related field. A good understanding of wireless communications, networking, probability, optimization, and artificial intelligence is desirable. Knowledge of machine learning and reinforcement learning will be considered an asset. The candidate should also possess strong programming skills, particularly in Python and/or Julia. Experience with scientific computing tools and algorithm development is advantageous. The ability to work independently, strong analytical skills, scientific curiosity, and a keen interest in research are essential for success in this project.
675. Scientific visualization: web apps for sport rating
What if you could open an app and instantly see live ratings, match predictions, and player rankings for your favourite sport presented in a clear, interactive, visually compelling way? That's what this project is building. Our lab develops sophisticated sport rating and prediction algorithms (see brbalab.com for a prototype). The challenge addressed in this project is on the presentation side: how do you take complex probabilistic outputs and make them intuitive, beautiful, and useful for a general audience?
You will work on designing and building the web and/or mobile frontend that displays our rating results across sports (football, ice hockey, and others). The work includes creating visually appealing and interactive displays of rating and prediction results, where users can access rankings and forecasts conveniently.
This project is ideal for students who love building things people actually use and who want their code to have a visible, real-world impact.
Research area, student roles & skills
Research area: - Web programming
- Mobile app programming
- Data visualisation
Student roles: The student will design and implement the application layer that presents our sport rating and prediction algorithms to end users. The focus is on software engineering and UI/UX — making complex results accessible, interactive, and visually engaging across web and mobile platforms.
The goal of the project is to study rendering methods in augmented reality that allow for the seamless integration of real and virtual image elements. This study will specifically address two simple visualization challenge cases: 1) How to make a real cavity disappear in a real wall and 2) How to create a virtual cavity in a real wall without a cavity. An experimental wall will be built in the laboratory to carry out this experiment. The project will aim to identify 3D rendering and real-time image processing algorithms that maximize the realism of the visualization for both use cases.
Research area, student roles & skills
Research area: The laboratory aims to explore the paradigms of visualization and human-machine interaction that are made possible by new virtual reality and augmented reality (VR/AR) devices. The laboratory's work is applied mainly to two areas: 1) Medical interventions and 2) Immersive multimedia experiences. In the medical field, VR and AR tools aim to improve techniques for simulating, planning and guiding surgical interventions. In the context of immersive multimedia experiences, the laboratory's work focuses on the new paradigms of interaction and visualization opened up by AR and VR.
Student roles: The intern's role is to develop an augmented reality application for virtual reality headsets such as Meta's Quest 3. This application will render the two visual illusions mentioned in the project description. In particular, the intern will have to study different real-time image processing algorithms that are likely to make these illusions realistic and the distinction between real and virtual seamless in the application.
Skills required: The candidate should have a background in computer science or computer engineering and/or strong programming and software engineering skills. They should be motivated to deepen their knowledge of 3D rendering, computer vision and human-computer interaction techniques and be willing to study relevant literature. The candidate should have knowledge of the following tools: * Game engines (Unity, Unreal Engine ou Godot) and their associated programming language (C#, C++ or Python) * Graphics programming SKDs (OpenGL, Vulkan, DirectX, Metal, WebGL)
677. Searching for Needles in the Big Data Haystack: an Automated Pipeline for a Robotic Astronomical Observatory
The Colibri Telescope Array (https://www.colibri-telescope-array.ca) is a robotic astronomical observatory that acquires large volumes of imaging and time-series data every night. Over several years of operation, numerous software tools have been developed to support telescope control, image calibration, photometric extraction, quality assessment, data archiving, and scientific analysis. As is common in research environments, many of these components were developed independently by different students and researchers, resulting in a code base that has become increasingly difficult to maintain, extend, and deploy.
The objective of this project is to modernize and refactor the Colibri software ecosystem using contemporary software-engineering practices. The student will analyze the existing pipeline architecture, identify inefficiencies and redundancies, and redesign key components to improve maintainability, performance, reliability, and reproducibility. Particular emphasis will be placed on modular software design, automated testing, documentation, version control, and workflow automation. The student will also be able to design novel modes of observation and data reduction to expand the science capabilities of the telescope array.
The student will work closely with astronomers to understand the scientific requirements of the system and translate them into robust software solutions. Depending on the student’s interests and background, the project may involve redesigning database interfaces, optimizing computational performance, containerizing software environments, implementing continuous integration workflows, developing web-based monitoring tools, or incorporating machine-learning techniques for data quality assessment.
The project provides an opportunity to apply modern computer-science techniques to a real-world scientific infrastructure used in active astronomical research. The resulting software improvements will directly support ongoing exoplanet and time-domain astronomy programs and will help ensure the long-term sustainability of the Colibri observatory.
At the conclusion of the internship, the student will deliver documented software modules, updated developer documentation, a modernization roadmap for future development, and a presentation summarizing the technical improvements achieved during the project.
Research area, student roles & skills
Research area: Modern astronomy increasingly relies on automated software systems to process and analyze massive datasets. Robotic observatories routinely generate terabytes of imaging and time-series data that must be calibrated, archived, analyzed, and monitored with minimal human intervention. Our research group operates the Colibri Telescope Array, a network of robotic telescopes dedicated to time-domain astronomy, including the detection of small Solar System bodies, stellar variability, and exoplanet transits, stellar variability. The scientific success of such facilities depends not only on the telescopes themselves but also on reliable, efficient, and scalable software pipelines that transform raw observations into scientifically useful data products.
Student roles: The student will work as a software developer within Prof. Metchev's Colibri Telescope Array team and will be responsible for improving the architecture, reliability, and maintainability of the observatory’s data-processing infrastructure. The internship will begin with an assessment of the existing code base. Under supervision, the student will review software modules, map dependencies, identify technical debt, and document the current pipeline architecture. Based on this assessment, the student will help develop a modernization plan prioritizing components with the greatest impact on reliability and scientific productivity.
The student will refactor selected portions of the pipeline using modern software-engineering practices. Activities may include restructuring legacy Python code into reusable modules, implementing automated unit and integration tests, improving error handling and logging, optimizing computational performance, simplifying deployment procedures, and developing automated workflow management tools. The student may also create containerized environments using Docker to improve software portability and reproducibility across computing platforms.
A significant fraction of the internship will involve programming, code review, debugging, and documentation. The student will use Git-based collaborative development workflows and will participate in regular meetings with the research team to discuss requirements, review progress, and evaluate proposed software solutions. They will gain experience working within a multidisciplinary environment where software engineering directly supports scientific discovery. Depending on project progress, the student may also contribute to the development of monitoring dashboards, automated quality-control systems, database interfaces, or machine-learning tools for anomaly detection in astronomical datasets.
Throughout the internship, the student will maintain technical documentation and present progress updates to the team. By the end of the project, the student will deliver production-quality software improvements that will become part of the operational Colibri infrastructure. The internship will provide valuable experience in software architecture, scientific computing, collaborative software development, and the management of complex data-processing systems used in modern research environments.
Skills required: The ideal applicant will be pursuing a degree in Computer Science, Software Engineering, Data Science, Computer Engineering, or a related discipline. Strong programming experience in Python and/or C, C++ is required, including familiarity with object-oriented programming, software design principles, testing frameworks, and version-control systems such as Git. Experience with Linux, scientific computing, databases, web technologies, cloud computing, or continuous-integration tools is desirable. Familiarity with machine learning, distributed computing, or scientific data processing would be advantageous but is not required. No prior astronomy background is necessary; training in the scientific context will be provided.
678. Secure LLM-Based Code Generation
Supervisor: Jinqiu Yang
University: Concordia University (Montréal campus)
Large Language Models (LLMs), trained on billions of
parameters across extensive corpora, have significantly ad-
vanced various software engineering tasks, such as code
generation and program repair. Prominent
models such as ChatGPT and GitHub Copi-
lot [8] have recently gained widespread popularity for their
effectiveness in these domains.
Despite the success of LLMs in generating functionally
correct code, researchers have identified quality concerns in
LLM-generated code. One of the most critical issues is the
presence of security vulnerabilities. Pearce et al. [9] revealed
that approximately 40% of code snippets generated by
GitHub Copilot contain vulnerabilities. Recent studies [10,
11] also found similar security issues in code generated by
other LLMs. These findings suggest that, despite advances,
vulnerabilities continue to pose a significant challenge in
LLM-based code generation.
In this research project, we will research how to improve the LLM-based code generaiton for better secure code generation.
Research area, student roles & skills
Research area: large language models for software engineering (LLM4SE)
large language models for code generation
secure code generation
Student roles: The students will conduct experiments, write papers under the supervision of the PI.
Skills required: Java Python
679. Secure Retrieval-Augmented Generation Against Poisoning Attacks
Retrieval-Augmented Generation (RAG) systems enhance large language models (LLMs) by integrating external knowledge retrieval into the generation process. While RAG improves factual accuracy and domain adaptability, its dependence on external knowledge sources introduces serious security vulnerabilities, particularly poisoning attacks. In these attacks, adversaries inject malicious or misleading documents into the knowledge base so that the retriever selects poisoned content during inference, causing the LLM to generate manipulated, biased, or harmful responses. Such vulnerabilities pose significant risks for real-world applications in healthcare, finance, education, and cybersecurity.
This research project aims to develop a secure and robust RAG framework capable of detecting and mitigating retrieval-layer poisoning attacks. The project will investigate how poisoned documents affect retrieval rankings, embedding representations, and downstream text generation in modern RAG architectures. Various attack strategies, including semantic poisoning, adversarial document injection, stealthy trigger-based attacks, and retrieval manipulation techniques, will be explored to understand their impact on system reliability and trustworthiness.
To improve RAG security, the research will examine defense mechanisms such as anomaly detection, secure reranking, retrieval filtering, embedding consistency verification, and trust-aware retrieval strategies. Privacy-preserving techniques and robust retrieval optimization methods will also be investigated to reduce the influence of malicious content while maintaining retrieval quality and generation accuracy.
The expected outcome of this research is a scalable and attack-resilient RAG framework that enhances the security, reliability, and trustworthiness of LLM-based systems operating in adversarial environments. The proposed framework will contribute toward the safe deployment of AI-driven knowledge systems in critical real-world applications.
Research area, student roles & skills
Research area: My research focuses on developing secure and privacy-preserving AI systems with applications in cybersecurity, healthcare, smart and connected environments, and trustworthy intelligent systems. I have worked on behavioral biometric authentication, federated learning frameworks, adversarial attack and defense mechanisms, and the secure deployment of LLMs and Retrieval-Augmented Generation (RAG) systems. I have published in journals such as IEEE Access and MDPI, and in conferences including FPS, IFIP SEC, SECRYPT, IEEE BigData, and IEEE COMPSAC. I also have experience supervising students under the Mitacs GRI program, supporting their research training and project development.
Student roles: The student will play an important role in supporting the development of secure and robust Retrieval-Augmented Generation (RAG) systems against poisoning attacks. Their work will begin with an in-depth literature review on RAG architectures, retrieval-layer vulnerabilities, adversarial attacks, and poisoning techniques targeting large language model (LLM)-based systems. The student will assist in designing and implementing experimental frameworks to evaluate how poisoned or malicious documents influence retrieval quality and downstream text generation in modern RAG pipelines.
A major responsibility will involve developing scripts and pipelines to simulate poisoning attacks, including semantic poisoning, adversarial document injection, and trigger-based retrieval manipulation. The student will help evaluate system robustness using retrieval accuracy, ranking consistency, attack success rate, and generation reliability metrics. They will also investigate defense mechanisms such as anomaly detection, secure reranking, retrieval filtering, embedding verification, and trust-aware retrieval strategies to improve the security and robustness of RAG systems.
Additional responsibilities include maintaining reproducible codebases, analyzing experimental results, preparing visualizations and technical reports, and contributing to research publications and presentations. The student will participate in regular research meetings and collaborative discussions throughout the project.
Through this project, the student will gain hands-on experience in RAG systems, LLM security, adversarial machine learning, information retrieval, and trustworthy AI, contributing toward the development of secure and reliable AI-driven knowledge systems for real-world applications.
Skills required: The student should have strong programming skills in Python and a solid understanding of machine learning, deep learning, and Retrieval-Augmented Generation (RAG) systems. Proficiency with PyTorch or TensorFlow is essential, along with familiarity with large language models (LLMs) such as GPT, BERT, or LLaMA. Experience with natural language processing, vector databases, embedding models, and information retrieval techniques is highly desirable. The student should be comfortable working with open-source LLMs, RAG frameworks, and benchmarking or evaluation tools. A strong foundation in linear algebra, probability, and statistics is required.
We will explore new approaches to face poisoning threats, such as backdoor attacks, and generalize the anomaly-learning module from label-noise signatures to attack-agnostic client-update representations. The method builds multi-signal descriptors of each client update (e.g., direction similarity, magnitude/dispersion statistics, and temporal consistency), applies calibrated anomaly scoring to detect malicious contributions under non-IID data, and triggers a mitigation policy (down-weighting, quarantine, or robust-aggregation fallback). We evaluate robustness–utility trade-offs across attacker fractions, heterogeneous client populations, and adaptive poisoning strengths, demonstrating improved resilience without materially degrading convergence or clean accuracy.
Research area, student roles & skills
Research area: Secure AI, Safe AI, Cybersecurity
Student roles: State-of-the-art analysis. Experiment evaluation. New research proposal implementation.
Skills required: Cybersecurity, machine learning
681. Securing AI Agent Tool Access: A Security Benchmark for Model Context Protocol Deployments
AI agents increasingly act on enterprise and critical systems through the Model Context Protocol (MCP), which connects language models to tools, APIs, and data sources. Recent security disclosures show these integrations are vulnerable to tool poisoning, prompt injection via tool descriptions, privilege escalation, and confused-deputy attacks, yet no standardized, reproducible benchmark exists to measure agent susceptibility or evaluate defenses under the constraints of regulated environments. This project will develop such a benchmark. The intern will build a sandboxed MCP testbed with deliberately vulnerable and hardened server configurations, implement a defensive evaluation suite covering known attack classes, and measure how mitigation tools, such as allowlisting, human-in-the-loop consent gates, scoped credentials, and audit logging, reduce attack success across open-weight language models. The research is strictly defensive in orientation. Expected outcomes include an open-source benchmark, empirical mitigation guidance, and a co-authored paper targeting AI security venues.
Research area, student roles & skills
Research area: My research focuses on safe and trustworthy AI for regulated and critical-infrastructure environments. I work at the intersection of AI governance, AI security, and autonomous software systems: how agentic AI can be deployed with verifiable controls, how machine learning can be secured and audited in high-stakes settings, and how privacy-preserving techniques such as federated learning enable collaboration across organizations that cannot share data. My work combines applied research with enterprise practice, including the Model Context Protocol (MCP), policy-as-code controls, audit evidence generation, and risk-based AI lifecycle governance aligned with frameworks such as NIST AI RMF and ISO/IEC 42001.
Student roles: The intern will be the primary developer and experimentalist on the project. Weeks 1–4: review the MCP specification and recent attack literature, define the threat model with the supervisor, and build the sandboxed testbed. Weeks 5–8: implement the attack evaluation suite and produce baseline susceptibility results across model configurations. Weeks 9–12: evaluate mitigation layers, analyze results, document the benchmark for open-source release, and contribute to a draft workshop paper as co-author.
The student will participate in weekly supervision meetings, maintain a reproducible codebase, and present findings at the end of the internship.
Skills required: Strong Python programming skills and comfort working with APIs, containers, and Linux command-line environments. Foundational knowledge of cybersecurity concepts (authentication, authorization, least privilege, common attack patterns) from coursework or projects. Familiarity with large language models or agent frameworks (e.g., LangChain, OpenAI/Anthropic APIs) is an asset but can be learned. Curiosity about AI security and the responsible deployment of autonomous systems is essential. Experience with Docker, Git, and experiment scripting is helpful. No prior research experience required; strong software engineering fundamentals matter most.
682. Security Auditing of LLM Based Code Generation for Detecting and Mitigating Vulnerable Code Synthesis
Supervisor: Yasir Malik
University: Bishop's University (Sherbrooke campus)
Large language models (LLM), including GitHub Copilot, CodeLlama, and GPT-4o, are increasingly integrated into professional software development workflows, generating substantial proportions of production code with limited human review. Empirical studies have reported measurable rates of critical vulnerabilities in LLM generated code, including buffer overflows, SQL injection, insecure deserialization, and weak cryptographic implementation, particularly under ambiguous or underspecified prompting conditions. As reliance on such tools increases, the security characteristics of a codebase are determined in part by properties of the generating model rather than solely by the practices of the individual developer.
This project conducts a systematic empirical investigation of the vulnerability profile exhibited by open source code generating large language models under varying prompting conditions, and examines whether prompt engineering and fine tuning strategies reduce insecure output. The student will construct a benchmark of security relevant coding prompts spanning weakness categories defined by the Common Weakness Enumeration taxonomy, generate corresponding code samples across multiple open source models under baseline, adversarial, and structured prompting conditions, and analyze the resulting code using established static analysis methods. The investigation will determine whether system level security directives, exemplar based secure coding prompts, and structured output constraints produce statistically measurable reductions in vulnerability incidence.
The project will produce a comparative empirical assessment of code generating large language model security profiles, a reproducible evaluation methodology, and evidence based prompting recommendations. It addresses a current and insufficiently examined risk in software engineering practice, situated at the intersection of software security research and the empirical evaluation of generative artificial intelligence systems.
Research area, student roles & skills
Research area: My research interests are in the disciplines of software engineering and applied artificial intelligence, with applications in cybersecurity. I investigate AI-driven approaches to detect software vulnerabilities, secure the growing class of systems built on large language models, including code-generating assistants and autonomous AI agents, and identify AI-generated and human-driven social engineering attacks. My work addresses the security of AI systems themselves, including vulnerabilities introduced by LLM code generation and the emerging risks associated with agentic AI, alongside the use of AI to detect and counter malicious content and manipulation online.
Student roles: 1. Build a benchmark of security-relevant coding prompts spanning common vulnerability categories. 2. Generate code samples from multiple open-source LLMs under varying prompting conditions. 3. Run static analysis tools to detect and classify vulnerabilities in the generated code. 4. Compare vulnerability rates across models and prompting strategies. 5. Document findings and propose prompting guidelines to reduce insecure code generation.
Skills required: Strong Programming Competence – Python, with experience interfacing with large language model APIs Foundational Knowledge of Software Security – Familiarity with the OWASP Top Ten and the Common Weakness Enumeration taxonomy Empirical and Analytical Skills – Capacity to design and execute systematic empirical experiments Other Relevant Skills – Prior exposure to static analysis or linting tools is an asset; strong academic writing ability in English for technical reporting
683. Segmentation of Magnetic Resonance Images (MRI)
Supervisor: Mehran Ebrahimi
University: Ontario Tech University (Oshawa campus)
Analysis of medical images is essential in modern medicine. With the ever increasing amount of patient data, new challenges and opportunities arise for different phases of the clinical routine, such as diagnosis, treatment, and monitoring. This research project will focus on the automatic segmentation of patients' medical scans, more specifically magnetic resonance images (MRI).
A number of segmentation algorithms including adaptive thesholding, active contours, fast marching, level-sets, and graph partitioning have been shown to be useful for various imaging modalities including MRI. Our group has access to repositories of MRI data including breast, cardiac, and brain, obtained from partner Ontario hospitals.
The project will involve automatic and semi-automatic delineation and measurement of healthy anatomy and anomalies. Separating boundary of an organ from the surrounding tissue for various applications including computer-aided diagnosis (CAD) and computer-assisted surgery (CAS) will be the main focus of this project.
The potential candidate will be exposed to a wide range of medical image processing algorithms and is expected to perform literature review, implement and test algorithms on medical imaging data, and prepare scientific reports.
Research area, student roles & skills
Research area: Broadly speaking, our research interests lie in the diverse area of mathematical imaging and inverse problems. Our long-term research objective is directed towards developing and validating efficient numerical methodologies for solving real-world, ill-posed inverse problems in the field of medical image processing.
Student roles: The potential candidate will be responsible for utilizing and extending our image segmentation tools and algorithms aimed at computer-aided diagnosis (CAD) and computer-assisted surgery (CAS)
Skills required: The potential student is required to have excellent programming skills in Matlab. The candidate may have background in different disciplines including computer science, applied mathematics, physics, electrical and computer engineering, biomedical engineering, or a related field. Excellent knowledge of calculus and linear algebra is also required. Experience working with medical imaging data and numerical optimization schemes is desirable but not required.
684. Semantic Segmentation of Urban LiDAR Data for Infrastructure Analysis
Supervisor: Karim El-Basyouny
University: University of Alberta (Edmonton campus)
This project focuses on the semantic segmentation of high-resolution LiDAR point clouds collected from urban environments to enhance city infrastructure evaluation, planning, and digital twin creation. One of the core goals is to generate semantically enriched 3D models that can be used as components of a digital twin—a virtual representation of the city’s infrastructure. These models will support real-time monitoring, simulation, and planning applications used by municipal stakeholders.
Research area, student roles & skills
Research area: Our research focuses on the development of AI-enabled methodologies for semantic understanding of large-scale 3D LiDAR data collected in urban environments. A major component of our work involves the annotation, curation, and management of high-quality point cloud datasets and the development of deep learning models for semantic segmentation and object classification. Using advanced machine learning techniques, we extract and classify transportation infrastructure elements such as roads, sidewalks, curb ramps, signs, vegetation, and street furniture to support digital twin development, infrastructure assessment, and smart city applications.
Student roles: The research intern will support efforts to develop, train, and validate machine learning models that classify urban features such as roads, sidewalks, vegetation, and street furniture. A significant part of the work will involve implementing and optimizing deep learning algorithms for 3D point cloud segmentation, leveraging frameworks like PyTorch or TensorFlow. Interns will also assist in integrating these models into processing pipelines that automate both spatial analysis and visualization within a GIS environment. An average day will involve working at a computer in our campus lab, conducting data pre-processing, training models, evaluating outputs, and contributing to documentation. The intern will actively participate in weekly meetings with graduate students and the lead researcher to discuss progress, challenges, and refinements.The research intern will contribute to the development of large annotated LiDAR datasets used for training and validating semantic segmentation models. Responsibilities will include point cloud labeling, annotation quality assessment, label verification, and the preparation of benchmark datasets for machine learning applications. The student will assist in developing, training, and evaluating deep learning models for semantic segmentation of urban point clouds using frameworks such as PyTorch and TensorFlow. They will investigate state-of-the-art architectures, perform performance benchmarking, analyze classification accuracy, and support model refinement. Additional responsibilities include data preprocessing, feature engineering, dataset management, visualization of segmentation results, and integration of developed models into automated geospatial processing pipelines. Students will work closely with graduate researchers and participate in weekly meetings to discuss progress, challenges, and methodological improvements.
Skills required: Basic proficiency in Python. Interest in machine learning frameworks. Understanding of core principles of deep learning, particularly as applied to 3D data. Exposure to point cloud data or willingness to engage with 3D spatial data processing. Effective written communication skills, particularly for documenting code and summarizing findings. Interest in digital twin technologies, semantic segmentation, and applications in urban infrastructure.
Highly Desirable: Interest in machine learning modelling and AI applications for geospatial data analysis.
Considered an Asset: Experience with GIS tools or geospatial analysis techniques. Experience with data annotation, image labeling, or point cloud labeling tools.
685. SensaComfort-AI: A Hierarchical Transformer and Bayesian Deep Learning Framework for Local/Global Thermal Sensation and Comfort Prediction from Continuous Skin Temperature Fields
Supervisor: Abdelatif Merabtine
University: École de Technologie Supérieure (Montréal campus)
Predicting how a human feels in a given thermal environment is a fundamentally more complex problem than measuring physical temperatures. Thermal sensation and comfort are psychophysiological constructs shaped by:
• The absolute temperature of individual body regions
• The rate of change of temperature (dynamic sensation)
• The spatial asymmetry across the body (e.g., cold feet with a warm torso)
• Highly subjective, individual-specific physiological and psychological factors
Classical predictive models such as Fanger's PMV rely on steady-state whole-body energy balance and fail under non-uniform and transient conditions. Even more advanced models like Zhang's sensation/comfort model rely on simplified zone-level skin temperatures and do not fully leverage the rich spatial and temporal texture of a continuous thermal field.
The Core Problem: Given a high-resolution, time-varying 3D skin temperature field (as produced by Proposal 1), how can one accurately, and interpretably, predict:
1. Local Thermal Sensation per body segment
2. Global Body Thermal Sensation
3. Overall Body Thermal Comfort
Research objectives are :
- Develop a local thermal sensation prediction model that leverages region-specific spatio-temporal features extracted from the continuous temperature field
- Design a global thermal sensation model that learns the non-linear, asymmetric aggregation of local sensations using attention mechanisms
- Build a Bayesian Ordinal Regression model to predict overall body thermal comfort, including uncertainty quantification
- Validate all three prediction layers against human subject subjective response data collected in controlled climatic chamber experiments
- Benchmark the proposed hierarchical framework against state-of-the-art models (Fanger PMV, Zhang's model, standard ANNs)
Research area, student roles & skills
Research area: The supervisor's work sits at the intersection of building physics, environmental engineering, and human physiological science. Their primary scientific focus is understanding and quantifying how the built thermal environment affects human occupants ranging from the physical measurement of heat exchange between the human body and its surroundings to the psychological and physiological responses that these exchanges generate in terms of thermal sensation, comfort, and health.
His recent research focuses on the development and application of advanced machine learning architectures that integrate physical knowledge, spatial structure, and temporal dynamics applied to complex engineering and scientific systems where data is sparse,
Student roles: The student occupies a central and autonomous role in this project. They are not a support technician, they are the main researcher responsible for the scientific output working under the guidance of two supervisors (one thermal engineering expert, one ML expert). Below the main requirements and the proposed timeline.
Requirements/deliverables : 1. validated HT-ViT model for local thermal sensation prediction with per-zone interpretability 2. A Cross-Attention Global Sensation model with attention-based visualization of body zone contributions 3. A Bayesian Ordinal Comfort model providing probabilistic comfort predictions with uncertainty estimates 4. A benchmarking report comparing against Fanger's PMV, Zhang's model, and standard ANNs 5. At least two peer-reviewed publications (one on sensation modeling, one on comfort prediction) 6. An open-source inference API that accepts sparse sensor readings and outputs a comfort prediction in real time
Timeline (12 weeks) : weeks 1–2: Literature review weeks 3–4: Preprocessing existing experimental data weeks 5–6: HT-ViT development, training, and evaluation weeks 7–8: Cross-Attention aggregation model development and validation weeks 9–10: Bayesian Ordinal Regression model development weeks 11–12: End-to-end pipeline testing and benchmarking weeks 13–14: Writing / publication
Skills required: The candidate must have practical hands-on experience building and training deep learning models, theoretical knowledge alone is insufficient. The candidate must have a solid proficiency in : Self-attention mechanism: Query, Key, Value matrices, scaled dot-product attention Multi-head attention: parallel attention heads, concatenation and projection Positional encoding: sinusoidal and learnable positional embeddings Feed-forward sublayers, layer normalization, residual connections The complete Transformer encoder block and its computational properties Image patch tokenization: splitting a 2D image into fixed-size patches and projecting to embedding space Application of ViT to non-standard 2D representations
686. SimSnap: A cross-device interactive platform for collaborative environments
Supervisor: Ali Mazalek
University: Toronto Metropolitan University
Location: Toronto, Ontario
Start date: 2027-05-03 (flexible)
Disciplines: Computer Science, Engg-Computer, Design, Engg-Software, Engg-Systems and Technology, Engineering, Information Studies, Science and Technology, Studies Science and Technology
Mobile devices offer a great opportunity for the design of collaborative cross-device systems due to their accessibility, affordances, and easy adaptation. While much research has been conducted on cross-device systems and different interaction methods, few works effectively combine dynamic spatial device rearrangement and interactive cross-device gestures. Most existing applications focus on one of the two, either generating intricate device arrangements through simple interactions or facilitating complex cross-device interactions in a position-agnostic system.
Our project aims to design technologies that combine dynamic cross-device interaction and dynamic spatial device re-arrangement to provide improved engagement and collaborative user experiences for multi-user applications. We intend to design and implement several cross-device interactions e.g., connecting and disconnecting multiple devices with gestures, moving interactive objects (i.e., GUI components) across devices, and facilitating multiple subgroup interaction. We also intend to develop collaborative applications that apply these cross-device interactions, as well as incorporate dynamic spatial device reconfiguration and embodied interactions to ensure uninterrupted interactive activities. We will apply an iterative user-centred design approach to understand the user experience and level of engagement of our designed cross-device interactions and applications. In that process, interaction and technology implementation will be followed by a usability analysis study. We will further refine our cross-device interactions and technology, incorporating the outcomes of the user study. The overall outcome of our project will contribute to design interactive cross-device technologies for collaborative environments that will also support social engagement.
Research area, student roles & skills
Research area: Our work seeks to design novel interactive cross-device technologies that enable users to engage with each other in a collaborative environment. Our goal is to develop a new generation of cross-device technologies that can facilitate social engagement for a wide-range of application domains such as gaming, learning, path-finding, and music creation among multiple users. We aim to design these technologies to leverage and support collaborative and social relationships. We apply a variety of human-computer interaction (HCI) techniques and approaches to study the relationships between users and technologies to identify ways to improve the user experience (UX) in collaborative tasks.
Student roles: The student’s role in this project would be working with either or both of our design and development teams, depending on the student’s interest and background. These teams consist of post-doctoral researchers, graduate students, and undergraduate students. The student’s ultimate goal will be to develop cross-device interactions and applications for collaborative platforms. This may include conducting background research on cross-device interaction and dynamic spatial re-arrangement, analyzing potential interaction scenarios, designing cross-device gestures, developing under the existing frameworks, and implementing cross-device interactions. The cross-device interactions will be developed for multiple Chromebooks and other personal touchscreen devices. Other duties may include helping to develop test interfaces for experimental studies and conducting user studies.
The milestones for this project have been designed to be modular, such that if one component proves more challenging than anticipated, the student will still be able to achieve a demonstrable product by the end of the internship. The student will gain design and development experience with a number of domains (e.g., cross-device interaction, web-based mobile apps, multi-device communication), and with human-computer interaction methods and techniques.
The internship would take place in Synaesthetic Media Lab (Synlab) at Toronto Metropolitan University (TMU). Synlab is an interdisciplinary research lab, where members come from diverse educational backgrounds, including computer science, digital media, human computer interaction, and biomedical engineering. Thus the student would benefit from exposure to a wide-range of academic interests and possible research-related career paths.
Skills required: This project is seeking students for both cross-device interaction design and cross-device interaction implementation. Students should have a background in at least one of the following areas: computer science/software engineering, user interface/experience design, and human-computer interaction. Experience with collaborative design tools like Figma (for design) and mobile/web development frameworks like React.js, Express.js, and Node.js, and WebSocket (for implementation) would be an asset. Students should be confident with their assigned tasks and will be collaborating with our design and development team. They should be open to critical thinking, design thinking, and coding best-practices.
687. Simulation of Epidemic Spreading and Control in Temporal Multilayer Networks
Supervisor: Aresh Dadlani
University: Mount Royal University (Calgary campus)
This project investigates epidemic spreading and control in multilayer networks. In many real systems, spreading does not occur through a single type of interaction. For example, individuals may interact through physical contact, workplace or school connections, family networks, and online social platforms. Similarly, in cyber-physical systems, malware or misinformation may spread through communication links, device proximity, and social behaviour. A multilayer network model can capture these different interaction channels more realistically than a single-layer graph.
The student will develop a Python-based simulation framework for epidemic spreading on multilayer networks. Each layer will represent a different type of interaction, such as household contacts, workplace/school contacts, online contacts, or mobility-based contacts. The student will implement simple epidemic models such as SI, SIS, or SIR and compare spreading patterns on single-layer and multilayer networks.
The project will also examine basic control strategies, such as random immunization, degree-based immunization, layer-specific intervention, and targeted protection of highly connected nodes. The student will evaluate the impact of these strategies using metrics such as infection peak, final outbreak size, time to extinction, number of protected nodes, and effectiveness of intervention across layers.
The emphasis is on simulation, implementation, visualization, and interpretation rather than advanced mathematical derivation. The expected outcomes include documented Python code, network visualizations, epidemic curves, comparison plots, and a short technical report. This project is suitable for students interested in network science, epidemic modeling, data science, simulation, and applied computing.
Research area, student roles & skills
Research area: My research focuses on network science, epidemic modeling, complex systems, IoT networks, Age of Information, and data-driven analysis of spreading processes. I study how information, infections, malware, or behaviours propagate through complex networks and how control strategies can reduce spreading, improve reliability, and support timely decision-making. My work combines simulation, graph-based modeling, stochastic processes, and optimization-inspired algorithms for applications in public-health networks, cyber-physical systems, IoT security, and next-generation communication networks.
Student roles: The student will be responsible for implementing and evaluating a Python-based simulation framework for epidemic spreading in multilayer networks. The student will begin by reviewing basic concepts related to graph models, multilayer networks, and simple epidemic models such as SI, SIS, and SIR.
The student will then construct synthetic multilayer networks in which each layer represents a different type of interaction. For example, one layer may represent close physical contacts, another may represent workplace or school contacts, and another may represent online or long-range interactions. The student will implement epidemic-spreading rules and compare how the outbreak evolves when spreading occurs on a single layer versus multiple interacting layers.
The student will also implement simple intervention strategies, such as random immunization, degree-based immunization, layer-specific protection, and targeted protection of influential nodes. They will run simulation experiments under different network structures, transmission probabilities, recovery rates, and intervention levels.
The student will analyze the results using plots and summary statistics, including infection curves, outbreak size, infection peak, time to extinction, and intervention effectiveness. They will maintain well-documented code, participate in regular research meetings, and prepare a final report and presentation. The role is implementation-focused; the student is not expected to derive new mathematical models or conduct advanced theoretical analysis.
Skills required: The student should have good programming skills, preferably in Python. Familiarity with NumPy, Pandas, Matplotlib, NetworkX, or basic data analysis would be helpful. Basic knowledge of graph theory, probability, algorithms, or epidemic models is an asset but not required. The student is not expected to know advanced stochastic processes, differential equations, stability analysis, or mathematical epidemiology. A motivated student with experience in coding, simulation, and visualization should be able to complete the project successfully.
688. Simulation of arrhythmia in a computer model of the heart
Atrial fibrillation is the most frequent rhythm disorder in humans (nearly 250,000 patients in Canada). It often leads to severe complications such as heart failure and stroke. Diagnosis of this arrhythmia is mainly performed through the inspection of electrical signal recordings (electrograms and electrocardiograms). To develop and validate new diagnostic tools, it is necessary to understand the link between what the cardiologist observes (these electrical signals) and what is going wrong in the heart (the underlying cardiac pathology).
In parallel with the dramatic increase in computer power over the last few decades, computer models of cardiac electrical activity have evolved from small strings of cells to a detailed description of the whole heart. Integrating information from the molecular scale to the whole organ, our models can not only simulate arrhythmias and investigate mechanisms but also can evaluate diagnostic and therapeutic approaches. Used in combination with experimental and clinical research, computer modeling is expected to play an increasing role in the interpretation of biomedical measurements.
We create three-dimensional virtual models of the human atria based on anatomical, histological and electrophysiological data. In these models, conditions are set up that trigger and maintain an arrhythmia, as inspired by clinical observations and physiological hypotheses. A variety of conditions are simulated to reproduce different diseased states of increasing severity. The evolution of the electrical activity generated by the heart during an arrhythmia is simulated. Then, electrical signals obtained from computer simulations, animal experiments and patients can be analyzed and compared.
Research area, student roles & skills
Research area: My research topic is computational bioelectricity with special emphasis on modeling and signal processing. I combine integrative biophysical modeling and signal processing techniques to help improve diagnostic interpretation of cardiac bioelectric signals. I'm notably interested in atrial arrhythmias. In these studies, I collaborate with biomedical engineers, experimentalists, electrophysiologists and cardiologists.
Student roles: We have developed software to simulate the electrical activity of the atria and to compute electrical signals.
The role of students involves: - learning about bioelectricity and electrophysiology - collaborating with graduate students - writing python scripts to run simulations and analyze the results - setting up conditions in the model to initiate arrhythmias (complex dynamics characterized by the propagation of multiples "chaotic" electric waves in the cardiac tissue)
The project can be adjusted to fit the student's career objectives. Ultimately, the goal is to contribute to a research article.
Skills required: 1- Programming skills (preferably Python) 2- Some notions about Linux operating system 3- Interest in biomedical research 4- Mathematically-oriented problem solving skills 5- Background: B.Sc. student in mathematics/physics or electrical/computer/biomedical engineering or equivalent 6- Basic knowledge in signal processing
689. Sketching in Augmented Reality using Novel Input Devices
This research project will focus on developing novel input devices for sketching in AR. The goal is to support more accurate and intuitive sketching, improve precision, usability, and the overall experience of creating 3D content in AR. Specifically, the student will use the DataFeel dots, commercial devices that provide haptic and thermal feedback, to create a novel input device for sketching in AR.
First, we will design and develop the input device. We will use thermal and haptic feedback to help users identify the positions of surfaces in mid-air. As part of this step, we will develop an AR sketching system that enables novice designers to modify existing objects using AR. The goal is to help them increase their creativity. Second, we will compare the novel input device in a laboratory study with existing commercial controllers/pens used for AR interaction to determine whether our device improves sketching accuracy and user experience.
This project will help future applications for sketching in AR that use novel feedback types. In addition, the findings will contribute to the broader field of AR for human–computer interaction by informing how we can create precise input devices for AR.
Research area, student roles & skills
Research area: Augmented Reality (AR) allows users to place and interact with 3D digital content in the real world. One use of AR is to allow people to sketch new features onto existing furniture or buildings. However, mid-air sketching is challenging because users must estimate depth, distance, and orientation without a physical surface, reducing accuracy. Controlling devices in six degrees of freedom can also be difficult. With the new wave of AR glasses, it is important to develop novel input devices and interaction techniques for accurate sketching in AR.
Student roles: The student will take a leading role in the project from conception to completion, under the supervision of the Vertex lab PI. Through this role, the student will develop expertise in immersive systems development, 3D user interface design, experimental methods, and human-computer interaction research. Responsibilities include designing and developing the new input method, integrating it in an AR sketching app developed in Unity and compatible with the Meta Quest 3, and conducting a user study to evaluate system performance and user experience. For more details, see below:
Input Device Design: The student will identify related work for sketching in AR and an input device via a literature review. Topics explored will include sketching interfaces, input devices, and design. Using this past work, the student will design a novel input device.
Prototype implementation for the Quest 3: The student will design an AR sketching system that uses the novel input method. The goal will be for novice designers to add new features to existing physical products.
Laboratory study: The student will evaluate the prototype in a user study. The study will examine sketching performance, user experience, and creativity among the participants using either the proposed input device or existing controllers/pens. The student will assist with study design, participant recruitment, and data collection and analysis.
By leading this project, the student will gain the skills needed to conduct research, develop AR applications, and design novel input devices. They will also help disseminate findings through presentations or manuscript preparation.
Skills required: This project requires students with experience using Unity and C# to develop a 3D application. Students need a background in computer science or its equivalent and feel comfortable with object-oriented programming. They should also have experience working with multiple SDKs, as this project requires them to use them simultaneously.
A background in design or fabrication is preferred. Previous VR/AR development is not required, but a plus.
690. SnapCode: Empowering Underserved Students to Learn Web Development on Mobile Devices in India
Supervisor: Jeremy Bradbury
University: Ontario Tech University (Oshawa campus)
Developed as a tool to support the nonprofit organization Go Girl, which teaches coding in regional languages to marginalized communities in India, SnapCode enables users to capture images of handwritten HTML text from paper-based learning resources and seamlessly translate it into interactive code on their mobile devices.
By leveraging the ubiquity of smartphones, SnapCode aims to democratize access to coding literacy leading to positive learning outcomes among populations of low socioeconomic status. This project delves into the continued development, implementation, and potential impact of SnapCode, highlighting its significance as a beneficial tool in advancing the need for equitable education.
Research area, student roles & skills
Research area: My research interests span software development and computer science education. In particular I focus on applying AI to software development, as well as using AI to support personalised learning in educational games.
Student roles: Successful applicants will work in a team developing SnapCode, an open source educational tool for use in web development workshops. Activities will include design and development of mobile web applications, optical character recognition (OCR) libraries, testing & debugging of mobile web applications as well as the deployment of web applications on mobile platforms. There is also an opportunity for interested students to explore utilizing machine learning (ML) or large language models (LLMs) to support student learning through the mobile application.
Skills required: - General web development (required) - Debugging (preferred) - React, Node.js (preferred) - GitHub experience (preferred)
691. Social Media and Communication Analysis in Diaspora Communities
Supervisor: Rashid Khokhar
University: Algoma University (Sault Ste. Marie campus)
Location: Sault Ste. Marie, Ontario
Start date: 2027-05-04 (flexible)
Disciplines: Computer Science, Engg-Computer, Engg-Software, Media Studies, Sociology
This research project delves into how diaspora communities in Canada use social media to maintain their cultural identity, communicate, and engage with both their host and origin countries. By applying advanced deep learning techniques, the study analyzes large-scale, multilingual social media content generated by these groups. Transformer-based models like BERT, RoBERTa, and LLaMA are utilized for sentiment analysis, topic modeling, and misinformation detection to uncover trends in online discourse.
The research involves collecting and preprocessing data from platforms like Twitter/X, Reddit, and Facebook, targeting diaspora groups using location, language, or hashtag-based filters. Multilingual data is processed with language detection and translation tools, ensuring comprehensive inclusion of diverse communities. The models are fine-tuned for specific tasks such as sentiment classification and topic modeling, and evaluated based on accuracy, F1-score, and perplexity.
The project also emphasizes transparency and interpretability by employing explainable AI tools such as SHAP and LIME to understand model decisions, especially for policy-relevant insights. The results will be presented through interactive visualizations and dashboards, providing accessible and actionable insights.
This interdisciplinary research sits at the intersection of natural language processing (NLP), computational social science, and diaspora studies, offering valuable contributions to both academic research and real-world applications in Canada. The findings will support evidence-based policy-making, enhance community engagement, and provide a deeper understanding of the evolving communication patterns, cultural identity expressions, and social challenges faced by diasporic populations across the country.
Research area, student roles & skills
Research area: My research spans data privacy, cloud computing, cybersecurity, social network analysis, and AI/machine learning. I have addressed challenges in safeguarding personal data in cloud environments and developed secure, reliable systems. I proposed a cloud solution incorporating an entropy-based trust computation approach for data-as-a-service environments. Additionally, I have worked on heterogeneous information networks, focusing on healthcare data privacy, where I developed a differentially private algorithm for releasing de-identified health data with reduced information loss and enhanced scalability. My work also includes contributions to user credibility systems in social networks, blockchain-based solutions, as well as supply chain cybersecurity risk analysis.
Student roles: The student will play a key role in supporting various aspects of the research project focused on diaspora communities in Canada and their use of social media. The primary responsibilities will include collecting and preprocessing large-scale multilingual social media data from platforms like Twitter/X, Reddit, and Facebook. The student will assist in designing data collection strategies by applying location, language, or hashtag-based filters relevant to diasporic groups. This task will also involve cleaning, tokenizing, and labeling data for model training, ensuring the data is properly structured for analysis.
The student will contribute to model development by fine-tuning transformer-based models (e.g., BERT, RoBERTa) for sentiment analysis, topic modeling, and misinformation detection. They will collaborate on evaluating model performance using key metrics like accuracy, F1-score, and perplexity, and apply explainable AI tools such as SHAP or LIME to interpret model outcomes. The student will also assist in analyzing trends in sentiment, cultural identity expressions, and online discourse, offering insights into the evolving communication patterns of diaspora communities.
In addition, the student will engage in academic writing, assisting with literature reviews, drafting sections of research papers, and preparing reports for submission to academic conferences or journals. They will collaborate with team members to summarize findings, create visualizations, and develop interactive dashboards for both academic and community use. Throughout the project, the student will be expected to work independently on specific tasks, while also participating in collaborative team discussions to ensure the project progresses smoothly. Effective communication and presentation of research findings will be key responsibilities, contributing to the broader academic and policy-making community.
Skills required: - Strong foundation in programming (preferably Python) and basic knowledge of machine learning. - Familiarity with NLP tasks and basic deep learning (TensorFlow/PyTorch), plus willingness to learn transformers, is an asset. - Willingness to learn and work with transformer-based models. - Interest in data preprocessing and working with multilingual text. - Ability to conduct literature reviews, critically analyze research, and communicate findings through strong academic writing. - Ability to work independently and in a team.
692. Social network analysis and mining (2027)
Supervisor: Carson Leung
University: University of Manitoba (Winnipeg campus)
This Globalink 2027 research project focuses on social network computing and mining. Over the past few years, the rapid growth and the exponential use of social digital media has led to an increase in popularity of social networks and social computing. In general, social networks are structures made of social entities (e.g., individual users) that are linked by some specific types of inter-dependency relationship such as friendship. The emergence of Web-based communities and hosted services such as social networking sites (e.g., Facebook, Instagram, LinkedIn, Weibo, X) has facilitated collaboration and knowledge sharing between users. Most users of social media have many linkages in terms of friends, connections, and/or followers. Among all these linkages, some of them are more important than others. For instance, some friends of a user may be acquaintances, whereas some others may be close friends that care about him in such a way that they frequently post on his wall, view his updated profile, send him messages, invite him for events, and/or follow his tweets. In this project, students will apply the knowledge and skills they acquired in their undergraduate database courses to build a database for capturing relevant data from a social network. They will then develop a data science, data analytics, socio-economic analysis, business analytics, business intelligence (BI), and/or human-centered data-driven artificial intelligence (AI) solution that applies data mining or artificial intelligence (AI) techniques to social networks so as to help users of the social digital media to discover implicit, previously unknown, and useful knowledge from the social networks.
Research area, student roles & skills
Research area: My specialized research area in computer science relates to data management and data mining. In particular, my research team focuses on (a) data science, (b) data mining and analytics, (c) visualization and visual analytics, as well as (d) social network analysis and mining in recent years. Similar to the Mitacs Globalink Research Interns (GRIs) in summers 2013-2026, the student participates in this Globalink 2027 project will learn knowledge discovery and data mining techniques to provide a data science or business intelligence solution for social network computing, mining and analytics. Recent research outcomes can be found in https://sites.google.com/site/dblabuofm/publications
Student roles: Research within the Database and Data Mining Laboratory in Department of Computer Science at University of Manitoba mainly focuses on databases and data mining, which includes efficient and effective management of, knowledge discovery from, as well as analysis of, large amounts of data (such as transactional, uncertain, social media, Web, biomedical & healthcare data, and/or streams of data, as well as Big data). My current research programs focus on data mining as well as applications of data mining technologies to social computing and/or social network analysis. Our lab is a multi-cultural environment. Current and previous lab members include students from Asia (e.g., Bangladesh, China, India, Japan, Malaysia, South Korea), Central and South Americas (e.g., Brazil, Mexico, Paraguay), and Africa (e.g., Tunisia). Current and previous collaborators include researchers from Australia and France. Following our success in the Globalink 2013-2026 projects with GRIs (who have published their research findings from their Globalink research with me in refereed international conferences/journals & who are now graduate students in Canadian/US universities such as UBC, Waterloo, Columbia, CMU & UIUC or working in companies such as Google), the student participating in this Mitacs Globalink 2027 project will work under the same supervision model. In other words, the student will academic supervised by me (a Full Professor at the University of Manitoba) for an approximate 12-week research internship in the summer (with a flexible start date in May, June, or July). In addition, the student will also work closely with a senior lab member (who will serve as a mentor) and will contribute a small concrete data mining (social network analysis/mining) component of the project suitable at the level for the undergraduate students.
Skills required: Similar to skill requirements for the Globalink 2013-2026 GRIs (now in top Canadian/US universities like UBC, Waterloo, Columbia, CMU & UIUC or companies like Google), students participating in this 2027 project are expected to have taken some third-year computer science courses (e.g., database courses) when they started the internship. Specific details of the project will be finalized after discussion and mutual agreement between the professor and student. In general, the student will be asked to design/implement in C/C++, Java, Python, R (or other programming languages) a data mining component that is suitable and manageable for an undergraduate student.
The research field of “soft robotics” is very active. Like people and animals, soft robots conform to their environment and can interact much more gently with objects than traditional robots. (For more examples see: http://www.a3automate.org/many-uses-soft-flexible-robots/). The objective of this project is to design and test software for controlling a pneumatically-powered soft robot arm. The software will build upon existing code. It will acquire real-time data from several sensors, and implement a suitable controller (e.g. model predictive control). This project will provide an opportunity to learn and apply advanced control systems to a real world problem. The software development will be in C++.
Research area, student roles & skills
Research area: Creating control software, sensors and actuators for robotic systems with a focus on collaborative and autonomous robots. Mentoring students is a very important part of my job.
Student roles: The student is expected to design, program and experimentally test software for controlling the soft robot arm. They will be working as a member of the robotics research team at McMaster University.
Skills required: Programming in C++ and/or Python. Some knowledge of robotics.
694. Soft Wearable Haptic Devices using Magnetic Actuation
Supervisor: Aditya Nittala
University: University of Calgary
Location: Calgary, Alberta
Start date: 2027-05-14 (flexible)
Disciplines: Computer Science, Engg-Computer, Engg-Electrical, Engg-Software, Engg-Systems and Technology
Advances in materials science, chemistry, mechanical engineering, and robotics engineering have brought soft actuators to interactive devices. A wide range of working principles have been used for developing soft actuators: pneumatics, hydraulics, electroactive polymers, acoustofluidics, electrorheological and magnetorheological fluids. However, current soft actuators possess several limitations which include: slow response, low/no force output, difficulty in precisely controlling them, or require high operating voltages with bulky control setups.
Pneumatic actuators are commonly used in soft robotics because of their ability to generate large forces but they need high pressure, have low efficiency, are slow in response, and need noisy actuation valves. On the other hand, electrostatic actuators (e.g. dielectric elastomers) provide a high-frequency response but require thousands of volts with bulky control setups. This prohibits the use of these actuation principles in soft wearable settings.
Recently electromagnetic actuation has been receiving a lot of attention because of its fast response, lower actuation voltage, and highly programmable and well-controlled movements. However, their deployment in interactive applications is not fully explored. Current state-of-the-art soft magnetic actuators in HCI are limited to low-resolution and enable only subtle vibration feedback.
This project aims to develop high-resolution soft magnetic actuators that can deliver high force output and a variety of haptic sensations. By studying, simulating, and investigating novel device architectures, we aim to develop soft actuators that can be precisely controlled and programmed to open up new possibilities in AR/VR, Accessible computing, sports, and rehabilitation.
Research area, student roles & skills
Research area: My research vision is to enable the seamless integration of interactive computing devices into our daily environment. One theme of this research vision is to bring computing and interactivity closer to the user’s body. For this, I have designed and developed epidermal devices, i.e., interactive devices that reside on the human epidermis. These soft devices are ultra-slim, adapt themselves to the complex geometries of the body, and enable a wide range of expressive interactions. My research is interdisciplinary, and I blend diverse disciplines (e.g., Materials Science, Electrical Engineering, Digital Fabrication, Printed Electronics, and Human-Computer Interaction) to solve fundamental research problems.
Student roles: The project requires up to 2 students with skills in one or more of the following disciplines: Computer Science/Engineering, Electrical/Electronics Engineering, Mechanical/Mechatronics, and Materials Engineering.
Discipline-specific tasks for the students are as follows:
For students with CS background: Typical tasks will include designing and developing computational tools to create the actuator design. This involves programming tasks in Python/Unity/Web technologies depending on the student's expertise. Additionally, students will employ state-of-the-art optimization techniques to create the device design.
Students with Electrical/ Electronics Background: Typical tasks will include designing the control circuitry (using Arduino or other microcontroller platforms) to test and evaluate physiological sensing principles and device architectures. Experience developing miniaturized PCBs is desirable but is not a mandatory requirement. If students have experience using simulation tools (e.g. Comsol/Ansys), then tasks will include designing simulation experiments. Additionally, if students have experience in signal processing techniques, then tasks will include employing state-of-the-art filtering mechanisms to get a high throughput from the actuators.
Students with Mechanical/Mechatronics: If students have experience in FEM (Finite Element Methods) or have prior experience using simulation tools (e.g. Comsol/Ansys), then tasks will include designing simulation experiments. Students will work with functional materials to fabricate device designs.
Some of the other common tasks irrespective of the discipline include: reading relevant literature, designing technical experiments to measure the performance of the sensor, and working closely with other team members to ensure the success of the project. This is a highly multi-disciplinary project and hence students will be working with peers from other disciplines.
Skills required: This project requires 1-2 students with a background in any one of the disciplines: Computer Science, Electronics and Electrical Engineering, Chemistry, Materials Engineering, or Mechatronics. The following skills are desirable for students from non-CS disciplines: designing PCBs, designing/working with printed electronics, signal processing techniques, knowledge/experience of simulation engines such as COMSOL/Ansys/FEMM, etc. For Students with a CS background, the following skills are desired: strong implementation skills in one or more of the following: Python, Unity, JAVA/C#.Net, Android, and ML tools (Pytorch/TensorFlow). Highly motivated students with a desire and aptitude to learn new technologies quickly are also encouraged to apply.
695. Software/AI Pipeline for Automated High-Throughput Recording and Analysis of Brain Activity and Behaviour
Supervisor: Majid Mohajerani
University: McGill University (Montréal campus)
Location: Montreal, Québec
Start date: 2027-06-01 (flexible)
Disciplines: Computer Science, Engg-Software, Engg-Systems and Technology
This project aims to develop an integrated software and AI pipeline for the automated, high-throughput recording and analysis of brain activity and behavior. By combining large-scale neural imaging techniques, such as two-photon tomography, with continuous behavioral monitoring, the project will generate rich, multimodal datasets that capture brain function across spatial and temporal scales.
The core objective is to design scalable data acquisition, processing, and analysis workflows capable of handling petabyte-scale datasets. This includes automated image processing, signal extraction, and behavioral tracking using advanced machine learning and computer vision methods. The pipeline will enable unbiased detection of neural activity patterns and behavioral phenotypes, reducing reliance on manual annotation and improving reproducibility.
A key component of the project is integrating edge computing with centralized high-performance computing infrastructure to support real-time or near-real-time analysis. This will allow rapid data interpretation and iterative experimental design.
By bridging neuroscience, artificial intelligence, and data engineering, this project will accelerate the study of brain function in health and disease. The resulting platform will facilitate large-scale, collaborative research efforts and provide a foundation for developing predictive models of neural dynamics and behavior.
Research area, student roles & skills
Research area: This project focuses on developing an integrated software and AI pipeline for automated, high-throughput acquisition, processing, and analysis of brain activity and behavior. The work combines large-scale neural imaging (e.g., two-photon tomography) with behavioral monitoring to generate multimodal datasets at unprecedented scale. Advanced machine learning and computer vision methods are used for signal extraction, feature detection, and behavioral classification, enabling unbiased and reproducible analyses. The pipeline emphasizes scalability, automation, and real-time or near-real-time processing, supporting petabyte-scale data. This research lies at the intersection of neuroscience, artificial intelligence, and data engineering, with the goal of accelerating discovery in brain function and
Student roles: The student will join a small team and assist with the mechanical and electrical design, manufacture, and testing of instrumentation for use in several research labs across North America and Europe. The candidate will work closely with the software teams to develop new products and improve existing ones. Building on the success of previous and existing systems, a new class of intelligent systems will replace current methods for assessing brain activity.
Skills required: The software developer will join our team to contribute to the progress of an exciting and interesting project that includes several areas in which the applicant could work (e.g. Cloud computation, database programming, UI/UX, computer vision, machine learning, and 3D game development. Position requirements: The students with the following skillsets are encouraged to apply: • Experience in Linux • Strong background in coding (Python, MATLAB, etc.) • Experience with AI, deep learning, and machine learning tools (especially for behavioural analysis) • Previous research experience in computer vision Effective communication skills, both written and verbal
Soil moisture is a foundational pillar of agricultural productivity, directly impacting crop growth, water resource management, and overall irrigation efficiency. Accurately forecasting moisture levels empowers farmers and water managers to optimize irrigation schedules, minimize water waste, and proactively mitigate the severe economic impacts of agricultural drought.
However, achieving high accuracy in soil moisture predictions remains a critical and complex challenge. The moisture content of the soil is constantly shaped by highly nonlinear, interconnected environmental variables. Atmospheric conditions (such as temperature, precipitation, and relative humidity), soil characteristics (including texture and porosity), and vegetation health continuously influence how water infiltrates, evaporates, and is absorbed by plant roots. Because traditional physics-based models struggle to capture these intricate, rapidly fluctuating relationships across varying spatial and temporal scales, researchers are increasingly turning to advanced data-driven approaches.
This research aims to delve into the growing body of literature surrounding data-driven soil moisture prediction, specifically investigating how deep neural networks can extract hidden patterns from large-scale meteorological and hydrological data. By utilizing extensive, publicly available environmental datasets—such as those hosted by the International Soil Moisture Network — the study will develop and train a deep learning model to forecast future soil moisture states with high precision.
Furthermore, a comparative analysis will be conducted. By testing the newly developed model against state-of-the-art benchmarks (such as Long Short-Term Memory networks or Transformer architectures), this research will evaluate the proposed model's strengths, limitations, and overall predictive accuracy. This systematic evaluation will highlight opportunities for further improvement, ultimately providing a more robust, reliable, and scalable tool for sustainable agricultural water management.
Research area, student roles & skills
Research area: My research area is focused on the data analytics using Artificial Neural Networks (ANNs) on the sensor data coming from wearable sensors, smartphone, smartwatch, and so on. For agricultural sensor data, my team has already implemented Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), hybrid CNN-LSTM for soil pH level prediction and soil moisture level prediction.
Student roles: The primary research objective for this project is to predict soil moisture from sensor data. To achieve this, the student is required to create an ANN-based model and validate the developed model by performing a comparison study with the existing state-of-the-art models.
1. A literature review of the soil moisture prediction using ANN-based models: This work involves conducting an exhaustive literature review of current studies related to soil moisture prediction from sensor data using ANN-based models. This will include the implementing the state-of-the-art models and comparing them using a common benchmark dataset.
2. Development of an ANN-based model for soil moisture prediction: This phase involves the design and implementation of a deep neural network model. A preliminary review has identified the performance metrics and available libraries to be used for the development. Key criteria for performance include R2 and Root Mean Square Error (RMSE) among others. Concurrently, data will be investigated from the International Soil Moisture Network (ISMN), targeting the RISMA network stations located in Saskatchewan. These datasets provide in situ soil moisture and temperature measurements at various depths. This will be supplemented with corresponding historical meteorological data, such as precipitation and air temperature, from Canadian weather data providers. The collected time-series data will be preprocessed, which includes cleaning, normalization, and handling of any missing values. The model will be trained to predict soil moisture for multi-depth at various time steps
3. Validation: The trained model will be evaluated on a held-out test dataset using common performance metrics, including RMSE, Mean Absolute Error (MAE), and R2. The developed deep neural network model will also be compared with existing state-of-the-art models in the domain of soil moisture prediction from sensor data.
Skills required: The student must have advanced coding skills using Python. The student should be able to work with Keras backend. Preliminary knowledge in developing Artificial Neural Network (ANN) models is an asset. Skill to adapt and learn new concepts and technologies is a must.
697. Solving a vehicle routing problem by swarm intelligence algorithm
Supervisor: Mohamed Tawhid
University: Thompson Rivers University (Kamloops campus)
In the last decade, new insights and algorithms have been obtained for the
classical deterministic VRP as well as for natural stochastic and dynamic variations of it. These new developments are based on theoretical analysis, combined probabilistic and combinatorial modeling, and lead to new and effective algorithms and a deeper understanding of uncertainty issues in vehicle routing problems. The VRP is an NP-hard problem with many extensions such as the Vehicle Routing Problem with Time Windows (VRPTW).
The goal of this project is to use and develop swarm intelligence algorithm for a provider of software planning tools for distribution logistics companies. The algorithm computes a daily plan for a heterogeneous fleet of vehicles, that can depart from different depots and must visit a set of customers for delivery operations. Besides multiple capacities and time windows associated with depots and customers, the problem considers incompatibility constraints between goods, depots, vehicles and customers, maximum route length and durations, upper limits on the number of consecutive driving hours and compulsory drivers’ rest periods, the possibility to skip some customers and to use express courier services instead of the given fleet to fulfill some orders, the option of splitting up the orders, and the possibility of “open” routes that do not terminate at depots. Moreover, the cost of each vehicle route is computed through a system of fares, depending on the locations visited by the vehicle, the distance traveled, the vehicle load
and the number of stops along the route. We will develop hybrid swarm intelligence algorithms (combine two
algorithms) and show the perofrmance of the proposed algorithm by comparing other algorithms in the literature and applying on-parametric statistical tests to show the superority of the proposed algorithm.
Research area, student roles & skills
Research area: The vehicle routing problem (VRP), first introduced by Dantzig and Ramser, is a well-known combinatorial optimization problem in the field of service operations management and logistics (Transportation).
The Vehicle Routing Problem (VRP) (problem formulation) is a generic name given to a whole class of problems in which a set of routes for a set of vehicles based at one or several depots must be determined for a number of geographically dispersed cities or customers. The objective of the VRP is to deliver a set of customers with known demands on minimum-cost vehicle routes originating and terminating at a depot.
Student roles: The student is expected to know some skills of programming like Python, or R or MatLab and swarm intelligence algorithms (is able to learn it quickly if s/he does not know). Student will write many computer programs by using Matlab software or Java and he will apply swarm intelligence algorithms such as Genetic algorithm, Particle Swarm Optimization Algorithm, differential evolution algorithm, spider, bat, and other algorithms. student will compare these algorithms with other algorithms. Also, student will help in writing a report and graph as a research assistant.
Skills required: Writing extensive computer programs in Python, or R or Matlab, Knowing swarm intelligence algorithms and optimization will be assets.
698. Speira
Supervisor: Luc Lavoie
University: Université de Sherbrooke
Location: Sherbrooke, Québec
Start date: 2027-06-07 (flexible)
Disciplines: Computer Science, Management Information Systems, Engg-Software
Projet Speira
Le développement logiciel fait appel à plusieurs procédés, dont l’organisation générale
est souvent présentée de façon séquentielle, mais qui procède plutôt, concrètement,
de façon itérative de sorte que l’enchainement des processus qui en découlent
forme en réalité une spirale.
L’intégration de l’activité de la documentation aux différents procédés est souvent
négligée, de même que l’intégration des systèmes logiques aux procédés de
modélisation et de conception. IAR est un prototype d’atelier de développement
logiciel proposant une approche intégrée visant à faciliter l’accomplissement
de ces deux tâches, la production des artéfacts associés, leur vérification
et leur validation.
Deux sous-projets ont été définis afin de refactoriser IAR et
de permettre la tenue d’expérimentation dans un contexte réaliste.
Sous-projet IAR_Logo (2 postes à pourvoir)
Au sein d’une équipe de développement, il est rare qu’un seul formalisme soit utilisé
pour exprimer les modèles et, pour un même formalisme, plusieurs outils sont
susceptibles d’être utilisés, outils ne partageant pas forcément les mêmes notations.
Pour permettre le partage de composantes entre les formalismes et le développement
d’outils communs (gabarits et vérificateurs),
un méta-modèle élémentaire de formalisme a donc été développé dans le cadre
de la preuve de concept IAR.
Ce méta-modèle doit être étendu dans le cadre du sous-projet.
Sous-projet IAR_Docu (2 postes à pourvoir)
Au sein d’une équipe de développement, il est rare qu’un seul outil soit utilisé
pour la documentation.
Les outils sont choisis en fonction de contraintes associées
* aux procédés;
* aux lectorats;
* aux supports;
* aux interactions;
* aux plateformes;
* etc.
Pour permettre le partage de composantes entre artéfacts et le développement
d’outils communs (gabarits et vérificateurs),
un méta-modèle élémentaire de document a donc été développé dans le cadre
de la preuve de concept IAR.
Ce méta-modèle doit être étendu dans le cadre du sous-projet.
Research area, student roles & skills
Research area: Le professeur Luc Lavoie a participé tout au long de sa carrière à plusieurs projets
liés aux systèmes d’information pour les petites et moyennes organisations.
La théorie relationnelle, la temporalité et la modélisation de données sont
au centre de ses intérêts tant en recherche qu’en enseignement.
Il s’intéresse également aux domaines de la spécification des exigences,
de la systématisation des processus de développement, d’évolution, de vérification
et de validation du logiciel.
Il s’intéresse présentement plus particulièrement aux outils de modélisation
de données temporalisées, historicisées ou relatives, aux langages multi-paradigmes
et à leur intégration aux procédés de développement.
Student roles: Le laboratoire Μῆτις dispose de postes de travail regroupés dans un espace ouvert favorisant l’échange et la collaboration. Cet espace, situé à la Faculté des sciences de l’Université de Sherbrooke, est partagé avec les groupes de recherche GRIIS et Hora.
Les projets sont pilotés selon le procédé en spirale (Boehm et Turner) afin de favoriser la créativité tout en permettant la maitrise progressive des exigences, des ressources utilisées, des jalons à respecter et de la qualité des produits.
Sur la base d’un document de spécification, la personne stagiaire devra proposer une conception, la mettre en oeuvre, la tester et participer à sa validation.
Skills required: Bonnes bases * en logique, * en théorie des ensembles, * en programmation, * en structures de données, * en algorithmique
Notions élémentaires * en langages formels (plus particulièrement les langages rationnels et algébriques) * en procédés de développement.
Dans le cadre de ses tâches, la personne stagiaire est appelée à consolider ces bases et développer ainsi ses compétences en regard des méthodes, des technologies et des outils en usage au laboratoire Μῆτις.
699. Stance Detection and Political Bias Mitigation in Generative AI
This project will address the challenge of building generative AI systems capable of detecting stance in politically sensitive contexts while remaining fair, explainable, and trustworthy. A central focus will be Open-Target Stance Detection (OTSD), where the stance target is not explicitly given and must be inferred from the text—reflecting real-world situations where opinions are expressed indirectly. We will develop a structured reasoning approach that combines Graph of Thoughts (GoT) with Retrieval-Augmented Generation (RAG) to infer implicit targets and generate transparent explanation traces that justify stance predictions.
To evaluate the quality of these predictions, we will implement a novel LLM-as-Judge evaluation framework. This system will assess whether the inferred target is specific and stance-relevant, and whether the predicted stance is logically consistent with both the target and the text. Special attention will be given to fairness: we will design mechanisms to detect if evaluations or explanations systematically favor dominant ideologies or overlook minority perspectives.
A key research contribution will be the creation of datasets with automatically inferred demographic and ideological annotations, enabling fairness auditing and robustness testing across sensitive groups.
Finally, we will extend this work to political bias mitigation through a debate-style inference framework in which multiple LLM agents simulate diverse ideological perspectives. A fairness-constrained moderator model will generate balanced final responses.
This project will deliver models, datasets, and evaluation frameworks that promote fairness and transparency in politically sensitive generative AI tasks.
Research area, student roles & skills
Research area: My specialized research area is Natural Language Processing (NLP). NLP is one of the most important and useful fields of artificial intelligence, dealing with automatic text processing. Its widespread application is due to the ubiquity of language and text as a main human communication channel, e.g., emails, blogs, social media, web search, chatbots, medical reports, etc. The objective of my research is to devise new methods for machine language understanding, which include language representation, summarization, and classification from unstructured and unconventional text streams, mainly in an unsupervised fashion.
Student roles: Expertise in any of these fields: Computer Science, Computer Engineering, Information Science and Technologies. Prior programming experience in Python. Experience with Large Language Models prompting and Fine-tunning. Experience with PyTorch. Knowledge of basic multivariate calculus, linear algebra, probability, and statistics. Familiarity with the essentials of data mining, machine learning, or artificial intelligence. Being previously introduced or worked on a project involving NLP.
Skills required: • Participate in collecting and preprocessing social media and political discourse datasets. Help automatically annotate these datasets with demographic and ideological attributes to support fairness auditing and robustness testing. • Explore and synthesize research literature related to stance detection, generative AI, explanation fairness, and political bias in large language models (LLMs). Gain a deep understanding of concepts such as Open-Target Stance Detection (OTSD), explanation specificity, and fairness evaluation strategies. • Contribute to the design and implementation of the LLM-as-Judge evaluation framework. This includes helping define evaluation criteria (e.g., target specificity, stance consistency), running experiments, and comparing LLM-based assessments
• Help
700. State of Practice of Procedural Generation in Commercial Games
Supervisor: Cristiano Politowski
University: Ontario Tech University (Oshawa campus)
Procedural content generation (PCG) — the use of algorithms to generate game content such as textures, terrain, levels, quests, and behaviour — has been studied academically for over three decades. The survey by Hendrikx et al. introduced a six-layer taxonomy spanning game bits, game space, game systems, game scenarios, game design, and derived content. Since then, the field has matured rapidly with the rise of machine-learning-based PCG (PCGML), large language models, and constraint-based methods. However, the state of practice in commercial games remains poorly understood. Academic surveys describe methods, but few systematically map which techniques studios actually use, in which genres, at what scale, and with what success.
This project will produce a contemporary empirical study of PCG adoption in commercial games. The intern will design and execute a mixed-methods study combining: (1) systematic literature review of PCG research from 2013 onwards, building on the Hendrikx taxonomy; (2) large-scale analysis of commercial games using a dataset of nearly 100,000 Steam titles maintained by the lab, identifying PCG use through metadata, store descriptions, reviews, and developer-authored patch notes; and (3) structured case studies of representative commercial titles spanning indie roguelikes, AAA open worlds, and procedural simulators.
The intern will codify findings into an updated taxonomy that captures modern PCG categories (PCGML, LLM-based generators, mixed-initiative tools), measure prevalence by genre and studio size, and identify gaps between academic methods and industry practice. Deliverables include an open dataset linking games to PCG techniques, a technical report mapping the state of practice, and a draft research paper suitable for IEEE Transactions on Games or FDG.
The work will lay groundwork for future research on benchmarking PCG systems and informing tool development for resource-constrained studios.
Research area, student roles & skills
Research area: My research focuses on empirical software engineering for video games, including the systematic study of game development practices, content generation methods, and quality assurance. I conduct large-scale empirical studies of commercial and open-source games using datasets of metadata, patch notes, postmortems, and source code. A central thread of my work investigates how procedural content generation (PCG), AI techniques, and automation are adopted in industry practice. My research informs the construction of taxonomies, datasets, and benchmarks that support reproducible study of how games are built and maintained, with attention to the needs of small and mid-sized studios.
Student roles: The intern will lead the empirical study of PCG adoption in commercial games, structured across the 12-week internship as follows.
In weeks 1–2, the intern will conduct a focused literature review of PCG research from 2013 onwards, with attention to PCGML, LLM-based generators, and constraint-based methods. They will internalize the Hendrikx six-layer taxonomy, identify candidate extensions for modern techniques, and produce an annotated bibliography. The intern will also onboard with the lab's existing datasets (PlayMyData metadata for nearly 100,000 Steam games; patch notes corpus; postmortems dataset).
In weeks 3–5, the intern will design and implement a classification pipeline to detect PCG use in commercial games from Steam metadata, game descriptions, reviews, and patch note evidence. They will define keyword lists, prompt LLM-based classifiers (such as Gemini or open-source equivalents) for harder cases, and validate labels through manual coding of a stratified sample with the supervisor.
In weeks 6–8, the intern will run the pipeline at scale, produce descriptive statistics of PCG adoption by genre, year, studio size, and platform, and identify candidate games for case studies. They will then perform deep case studies of 6–10 representative titles, documenting PCG techniques used, their role in gameplay, and developer communications about them.
In weeks 9–11, the intern will synthesize findings into an updated PCG taxonomy that augments the Hendrikx framework with modern categories. They will produce visualizations and analysis tables and validate the taxonomy by checking coverage across cases.
In week 12, the intern will deliver a technical report, contribute the dataset and pipeline as open-source artifacts, and assist in drafting a research paper. They will present results to the lab.
The intern will meet weekly with the supervisor and collaborate with lab members working on adjacent dataset projects.
Skills required: The student should have a computer science background with interest in empirical software engineering, game research, or AI-generated content. Strong programming skills in Python are required for data collection and analysis, including familiarity with pandas, scikit-learn, and visualization libraries. Experience with web scraping or APIs (e.g., Steam Web API), text analysis, and clustering methods is highly desirable. Familiarity with at least one game engine and a basic understanding of PCG concepts is a plus. The student should be comfortable reading academic surveys, conducting structured literature reviews, and writing well-organized technical reports in English.
701. Statistical analysis of brain and behavioral data
There is a profusion of open neuroscience data, providing analysts with the opportunity to pose and answer many important questions fairly rapidly. It is not possible to predict the precise project topic now in May 2026 for May 2027, but we are very open-minded and eclectic, as long as the question lies within our ability to at least move towards a solution within the time-frame of an internship. In general, we work in the areas of systems neuroscience, cognitive neuroscience, computational neuroscience, computational psychology, and neuro-AI. We are particularly interested in vision, hearing, eye-movements and gaze, behavior (all aspects), attention, memory, and heart-brain interactions.
Behavioral data collection is also a possibility in this project, with a focus on eye-tracking.
Research area, student roles & skills
Research area: Analysis of neuronal firing data and imaging/local field potentials; analysis of fMRI, EEG and MEG data; development of statistical methods; machine learning and neuro-AI; analysis of behavior; eye-tracking and analysis of gaze data; quantitative neuroscience; computational neuroscience; modeling.
Student roles: The student will take the lead in proposing/developing/extending a project (from the state in which it is in, in Summer 2027), and will work in close collaboration with the mentor.
Skills required: The student needs to be sufficiently fluent in programming (in any language, including Matlab/Python). The student also needs to have good general quantitative aptitude/skills, and an understanding of statistics (or ability and willingness to learn).
Prior knowledge of our specific research areas is not a requirement, as long as the student is willing to learn as needed.
702. Student engagement detection in online learning
Supervisor: Mohammad Ali Dewan
University: Athabasca University (Edmonton campus)
Location: Edmonton, Alberta
Start date: 2027-05-03 (flexible)
Disciplines: Computer Science, Engg-Computer, Engg-Software, Engg-Systems and Technology
Online learning plays a key role in today’s education system. Engagement detection in online learning is crucial as the student’s success in online courses heavily depends on his/her state of mind. In this paper, we will use student’s behavioral, emotional, and cognitive dimensions for engagement detection. Different machine learning and deep learning approaches will be used to analyze facial expressions and forum posts in online courses. We will perform empirical studies and develop prototype applications to test and explore theoretical concepts on student engagement detection in online learning platform.
Research area, student roles & skills
Research area: Artificial Intelligence in Education
Educational Data Mining
Affective Computing
Data Visualization
Machine Learning
Neural Networks and Deep Learning
Incremental Learning
Image Processing and Computer Vision
Student roles: Designing and developing system to apply in distance education Reviewing research articles, and implementing and analyzing systems for performance evaluation Writing report and research articles Doing project presentations in the progress meeting
Skills required: Programming experiences and skills in Python Experience and ability to design and build tools or systems Experience in designing user interfaces and data visualization Student must have knowledge in image processing, computer vision, machine learning, and pattern classification Experience in information collection, re-organization, summarization, and preparing reports Good communication skill
703. Study of Video Game Maintenance using Patch Notes
Supervisor: Cristiano Politowski
University: Ontario Tech University (Oshawa campus)
Video games are continually maintained after release through patches that fix bugs, add features, and adjust balance. Patch notes — developer-authored summaries published directly to players — provide a uniquely rich chronological record of game software evolution. The lab has recently constructed the largest publicly available classified patch notes dataset: 817,765 confirmed patch notes across 68,043 Steam games, processed into 8.3 million classified text fragments tagged as bug fixes (36.2%), feature additions (45.8%), or balance changes (18.0%). This dataset enables empirical study of game maintenance at unprecedented scale, but the analytical work to extract insights is still in its early stages.
This project will use the dataset to investigate questions about video game maintenance practices. The intern will pursue two interlinked research questions: (1) How do maintenance patterns differ across game genres, studio sizes, and game ages? (2) What recurrent bug categories, feature types, and balance changes emerge, and how do they map to existing software engineering taxonomies?
The intern will design and implement analysis pipelines on the dataset, applying clustering and topic-modelling techniques (sentence embeddings, BERTopic, LDA) to identify recurrent categories within each tag type. They will produce descriptive statistics of patch cadence, patch size, and tag composition over a game's lifecycle, and conduct case studies of representative games (indie roguelikes, AAA live-service titles, early-access games). The intern will collaborate with the supervisor on extending the taxonomy to capture game-specific maintenance concerns (balance changes, content drops, hotfixes) not well covered by traditional SE taxonomies.
Deliverables include reproducible analysis scripts, a dashboard for exploring patch-notes patterns by genre and studio size, a refined maintenance taxonomy, and a draft research paper targeting MSR, EMSE, or IEEE Transactions on Games.
Research area, student roles & skills
Research area: My research focuses on empirical software engineering for video games, with emphasis on how games evolve as software after release. I study video game maintenance through large-scale analyses of patch notes, postmortems, source code, issue trackers, and developer communications. My work produces datasets and taxonomies linking bug types, feature changes, and balance updates across genres, platforms, and studio sizes. By treating games as complex socio-technical systems, my research informs developer-facing tools for prioritizing maintenance, comparing live-ops practices, and predicting post-release issues. Recent outputs include datasets of nearly 100,000 multi-platform titles and over 800,000 classified patch notes from Steam.
Student roles: The intern will conduct an empirical study of video game maintenance based on the lab's patch notes dataset, organized across the 12-week internship as follows.
In weeks 1–2, the intern will familiarize themselves with the dataset, its schema, and the four-stage construction pipeline (data collection, keyword filtering, LLM classification, fragment extraction). They will review prior work in mining software repositories, patch note studies, and game maintenance research. The intern will define a focused research question with the supervisor and identify a stratified sample of games for deep analysis (e.g., across genres and studio sizes).
In weeks 3–5, the intern will implement an analysis pipeline producing descriptive statistics of maintenance behaviour: patch frequency, patch composition (bug fix / feature / balance proportions), patch size, and temporal patterns over a game's lifecycle. They will visualize these patterns at scale and identify outliers and notable cases.
In weeks 6–8, the intern will apply topic modelling and clustering techniques (sentence embeddings with BERTopic, or LDA) to each tag type, identifying recurrent subcategories within bug fixes, features, and balance changes. They will manually validate a sample of cluster labels and iterate on the methodology.
In weeks 9–10, the intern will conduct case studies of 5–8 representative games, qualitatively analyzing how maintenance evolves over each game's lifecycle. They will refine the maintenance taxonomy in collaboration with the supervisor.
In weeks 11–12, the intern will build a small interactive dashboard for exploring patterns, finalize the technical report, contribute documented code and analysis artifacts to the lab's open-source repository, and assist in drafting a research paper. They will present results to the lab.
The intern will meet weekly with the supervisor and collaborate with the student who built the dataset, who remains available for technical consultation.
Skills required: The student should have a computer science background with strong Python programming skills and experience in data analysis (pandas, numpy, scikit-learn). Familiarity with natural language processing techniques (embeddings, topic modelling, text classification) is required. Knowledge of full-text search systems (SQLite FTS, PostgreSQL, or Elasticsearch) is a plus, as is experience with web-based dashboards (Streamlit, Dash, or similar). The student should be comfortable working with large datasets, reading academic papers in empirical software engineering or mining software repositories, and writing technical reports in English.
704. Studying How AI Agents Are Changing Code Review in Open-Source Projects
Supervisor: Taher Ghaleb
University: Trent University (Peterborough campus)
Code review is a key part of software development, where contributors examine submitted code, leave comments, and decide whether to approve or reject changes. This human oversight process is central to maintaining software quality. However, AI agents are now both submitting pull requests and participating in code reviews, raising questions about how human oversight is affected. This project aims to investigate how AI agents influence code review behavior in open-source software. Using a large dataset of pull request review events with identifiers for human and AI contributors, the student will examine whether human reviewers change their behaviour when AI has already reviewed the code, whether AI reviewers treat AI-generated code differently from human-written code, and how the role of human maintainers evolves over time. The findings will contribute to a broader research programme on the impact of AI agents in open-source development.
Research area, student roles & skills
Research area: My research specializes in software engineering, with a focus on applying AI and data-driven techniques to improve software development processes. I work on mining software repositories, optimizing continuous integration and delivery (CI/CD), and leveraging large language models (LLMs) for tasks like code generation, testing, and configuration migration.
Student roles: - Process pull request review datasets, including review events, comments, and contributor labels - Measure changes in human reviewer engagement over time (e.g., comments, response time, approvals) - Compare how human and AI reviewers evaluate AI-authored versus human-authored pull requests - Investigate whether AI reviewers show bias or leniency toward AI-generated code - Analyze changes in maintainer activity as AI participation increases - Produce statistical summaries, tables, and figures for the paper - Contribute to writing the technical report and research paper
Skills required: - Proficiency in Python - Familiarity with Git and GitHub - Comfortable working in Linux and using the command line - Basic data analysis skills - Basic understanding of pull requests and code review workflows - Ability to read and understand research papers - Exposure to natural language processing (e.g., text classification or sentiment analysis) is a plus but not required
705. Studying Redundant Executions and Wasted Compute in CI/CD Pipelines
Supervisor: Taher Ghaleb
University: Trent University (Peterborough campus)
CI/CD pipelines are useful to automate the building, testing, and deployment of software, but they can also consume significant computing resources. In many projects, workflows may perform unnecessary work, such as repeating the same build jobs or steps, running tests multiple times across various workflows, or triggering pipelines when no relevant files have changed. These forms of redundant execution increase resource usage, energy consumption, and developer waiting time. This project aims to investigate redundant execution in CI/CD pipelines by analyzing GitHub Actions workflows from thousands of open-source repositories. The student will help identify different forms of redundancy, develop techniques to detect them automatically, and measure how frequently they occur and how much unnecessary computation they generate. The findings will provide new insights into the efficiency of real-world CI/CD practices and contribute to a research paper.
Research area, student roles & skills
Research area: My research specializes in software engineering, with a focus on applying AI and data-driven techniques to improve software development processes. I work on mining software repositories, optimizing continuous integration and delivery (CI/CD), and leveraging large language models (LLMs) for tasks like code generation, testing, and configuration migration.
Student roles: - Collect GitHub Actions workflow files and CI/CD run data from public repositories - Develop detection methods for different types of redundancy in CI/CD pipelines - Measure how often each type of redundancy occurs across projects and programming languages - Estimate wasted compute by analyzing run durations and usage of CI/CD minutes - Study relationships between redundancy and project characteristics such as size, age, and activity level - Analyze how redundancy changes over time within projects - Produce statistical summaries, tables, and figures for the paper - Contribute to writing the technical report and research paper
Skills required: - Proficiency in Python - Familiarity with Git and GitHub - Comfortable working in Linux and using the command line - Basic data analysis skills (e.g., pandas and statistical testing) - Ability to read and understand research papers - Basic understanding of software testing or CI/CD concepts is beneficial - Knowledge of YAML is a plus, but not required
706. Studying Security Aspects of CI/CD Pipelines
Supervisor: Taher Ghaleb
University: Trent University (Peterborough campus)
CI/CD pipelines automate the process of building, testing, and deploying software. However, they can also create security risks. For example, a poorly configured GitHub Actions workflow might expose secret credentials, give untrusted code too many permissions, or allow supply chain attacks where malicious third-party tools are introduced into the build process. These risks are becoming even more important as AI coding assistants and autonomous AI agents are increasingly able to contribute code, create pull requests, and modify workflows with limited human oversight. Although these problems are common, we do not yet have a clear understanding of how often they occur in real open-source projects, how effectively they can be detected, or whether they can be fixed automatically. This project aims to analyze thousands of GitHub Actions workflows to identify common security issues such as hardcoded secrets, unpinned third-party actions, overly broad permissions, and script injection vulnerabilities. The goal is to develop detection techniques using static analysis and/or large language models (LLMs), and then evaluate whether LLMs can generate secure fixes for the problems they find. The results will contribute to a research paper in software engineering.
Research area, student roles & skills
Research area: My research specializes in software engineering, with a focus on applying AI and data-driven techniques to improve software development processes. I work on mining software repositories, optimizing continuous integration and delivery (CI/CD), and leveraging large language models (LLMs) for tasks like code generation, testing, and configuration migration.
Student roles: - Collect GitHub repositories and their workflows at scale - Identify common CI/CD security anti-patterns from prior research and guidelines - Develop static analysis tools to detect security issues - Measure the prevalence and trends of vulnerabilities across projects - Evaluate LLMs for vulnerability detection and automated repair - Assess the correctness of generated fixes using labelled examples - Analyze results and produce figures, tables, and statistics - Contribute to the research paper and technical report
Skills required: - Proficiency in Python - Basic understanding of software security concepts (e.g., secrets, permissions) - Familiarity with Git, GitHub, and CI/CD pipelines - Comfort working in Linux/command-line environments - Basic knowledge of static analysis or program analysis concepts is a plus - Experience with LLM APIs is a plus, but not required - Ability to read and understand research papers
707. Studying Supply Chain Risks Introduced by AI Coding Agents
Supervisor: Taher Ghaleb
University: Trent University (Peterborough campus)
AI coding agents are increasingly contributing pull requests to open-source projects, where they write code, fix bugs, and modify configuration files. However, every change they make can affect the software supply chain, which includes the external code, tools, services, and automation a project depends on to build, test, and deploy software. AI-generated contributions may introduce supply chain risks that are difficult for developers to detect. These include adding vulnerable or unverified dependencies, using unpinned or third-party CI/CD actions, introducing licensing conflicts, or reusing code from unclear sources. Unlike human developers, AI agents may not fully account for project-specific security and maintenance constraints. This project aims to study how AI-authored pull requests introduce software supply chain issues to open-source projects. The student is expected to analyze a large dataset of real pull requests and compare AI and human contributions in terms of introduced dependencies, CI/CD configurations, licensing issues, and code provenance. The study will also examine whether these risks accumulate over time and contribute to a research paper in software engineering.
Research area, student roles & skills
Research area: My research specializes in software engineering, with a focus on applying AI and data-driven techniques to improve software development processes. I work on mining software repositories, optimizing continuous integration and delivery (CI/CD), and leveraging large language models (LLMs) for tasks like code generation, testing, and configuration migration.
Student roles: - Identify AI-authored pull requests using metadata and behavioral signals - Assess supply chain risks in each pull request (dependencies, CI/CD actions, licensing, provenance) - Compare risk profiles of AI-generated and human contributions at scale - Cross-check dependencies and components against vulnerability and license databases - Analyze whether AI-related risks accumulate or are corrected over time - Investigate reuse of external components across AI agents and projects - Produce statistical summaries, tables, and figures for the paper - Contribute to the technical report and research paper
Skills required: - Proficiency in Python - Familiarity with Git and GitHub - Comfortable working in Linux and using the command line - Basic data analytics skills - Ability to read and understand research papers - Basic awareness of software security or software supply chain concepts is beneficial but not required
708. Studying Test Optimization Strategies in Practice
Supervisor: Taher Ghaleb
University: Trent University (Peterborough campus)
Software projects use CI/CD pipelines to automatically run regression tests whenever developers make changes to a project. However, running the entire test suite can be expensive and time- and resource-consuming. Though many advanced test optimization techniques have been proposed in software engineering researche (e.g., test suite reduction, test case selection and prioritization, flaky test detection, etc.), little is known about how widely they are adopted in practice. This project aims to analyze thousands of open-source GitHub repositories and their GitHub Actions workflows to identify how tests are executed in CI/CD pipelines. Using a taxonomy of testing strategies, the study will investigate which approaches are most common, how adoption varies across programming languages and project sizes, and how testing practices have evolved over time. The findings will provide a large-scale empirical view of real-world CI/CD testing practices and contribute to a research paper.
Research area, student roles & skills
Research area: My research specializes in software engineering, with a focus on applying AI and data-driven techniques to improve software development processes. I work on mining software repositories, optimizing continuous integration and delivery (CI/CD), and leveraging large language models (LLMs) for tasks like code generation, testing, and configuration migration.
Student roles: - Analyze CI/CD workflow files from open-source repositories - Develop and refine techniques to classify testing strategies in GitHub Actions workflows - Measure the prevalence of different testing practices across projects and programming languages - Study how testing strategies have evolved over time - Investigate the adoption of third-party testing and analytics tools - Examine whether advanced test optimization techniques are used in practice - Analyze results and produce figures, tables, and statistical summaries - Contribute to the technical report and research paper
Skills required: - Proficiency in Python - Familiarity with Git and GitHub - Basic Linux and command-line skills - Basic data analysis and visualization skills (e.g., pandas and matplotlib) - Ability to read and understand research papers - Understanding of software testing concepts is beneficial - Familiarity with YAML is a plus, but not required
709. Supporting Mobile Security and Needs of Older Adults
Supervisor: Mary Oluwasola Adedayo
University: University of Winnipeg
Location: Winnipeg, Manitoba
Start date: 2027-05-31 (flexible)
Disciplines: Computer Science, Electronic Systems, Engg-Computer, Engg-Electrical, Engg-Software, Engg-Systems and Technology, Engineering, Management Information Systems, Mathematics, Physics, Science and Technology, Studies Science and Technology
The overall goal of this project is to address the lack of effective, accessible, and adaptive learning platforms to educate older adults on security and privacy in their use of mobile devices. Many cybersecurity and privacy programs focus on younger adults and do not consider the unique challenges that have been reported by older adults such as declining cognitive ability, and increased threats that specifically target older adults. This project will contribute to the design of a personalized adaptive learning platform that is tailored to address these challenges faced by older adults. This work will lead to an increased level of digital inclusiveness and reduced vulnerability among older adults. The project will build on existing findings and involve collaboration with a team of graduate and undergraduate working on developing the learning platform. Specifically, this project will focus on developing gamified components of the development to facilitate engagement and learning among older adults. While the focus of our work is on older adults, output from the research will be useful for other demographics.
Research area, student roles & skills
Research area: Ownership of mobile devices has grown significantly over the last two decades with these devices increasingly being used as a means of communication and accessing digital services in for example banking and healthcare. Despite some uptick in public awareness training regarding privacy and security. Older adults remain a particularly vulnerable group when considering mobile security and privacy. Our research focuses on how to mitigate some of the challenges faced by older adults in understanding using security features on mobile devices by working with older adults and developing personalized, adaptive learning platforms to ensure knowledge retention and mitigate vulnerability among them.
Student roles: The student will be required to do background research to understand the field of computer security, and some of the research advances in human-computer interaction. The student may be required to analyze interviews and surveys on requirement gathering and implement code in an applicable programming language (e.g., Java, Python, C, or C++). The student may be required to learn to use some open-source tools and apply them in the qualitative analysis of data, and prototype development. The student will be involved in writing up results and preparing them for publication. The student should be able to work well with others and be comfortable with working in a research area where the existing literature may be limited.
Skills required: We seek students with prior knowledge of computer security and privacy and an interest in human-computer interaction. The student must demonstrate excellence in courses such as Computer security, Data structures and algorithms, Software engineering and requirement analysis, Operating systems, Computer architecture, and system software, and be capable of working with or learning development tools as necessary. The student must be able to work with minimal direct supervision and be willing to ask questions as necessary. Strong communication skills (written and oral) and some experience working on technical projects and writing research articles are required.
710. Supporting Persistence in Artificial Intelligence Students
Building on the five-year research program that our team, spread across the University of Toronto, the University of Waterloo, and the Massachusetts Institute of Technology, has led, this project will extend the most recent survey findings. To improve persistence (e.g., students remaining in a specific field post-graduation) in Machine Learning/Artificial Intelligence (ML/AI), and to increase diversity within ML/AI development teams, this project will focus on how students understand their career fit in ML/AI. Our recent work found that students need more than programming skills to remain in ML/AI -- they need programming skills and a feeling that they fit within the career. Focusing on the "fit" portion, this project will use interviews with recent graduates to identify experiences within and outside the classroom that made them feel like they did or did not fit within ML/AI. Then, we will use web scraping to identify which core experiences are currently part of the standard ML/AI curriculum across institutions, and which gaps exist. The goal would be to eventually design curriculum support (in the form of short assignments or modules) that can be added to ML/AI university classes to help all students identify whether they would fit within the career.
Research area, student roles & skills
Research area: My research in this area investigates student career pathways in Machine Learning/Artificial Intelligence (ML/AI). This project has been ongoing since 2020, in which time we have built models to describe what factors affect whether students who take an ML/AI class choose to continue in the field, finding that aspects of feeling like your career benefits society, and feeling that you "fit-in" too the career, have a significant effect on career choice. We have also investigated how social belonging is built in ML/AI classrooms, and how early computer science programs support persistence (e.g., intentions to stay in an ML/AI career).
Student roles: The student will take the lead on one portion of this larger project. For example, they may be responsible for helping to conduct and analyze interviews with recent graduates, which would include designing the interview script, piloting the interview, editing the script, conducting the interviews, and then analyzing the interview data to identify themes. For a comparison of syllabi and ML/AI course descriptions, this would include identifying a data collection scope and strategy, collecting the data, and identifying themes across the descriptions/syllabi. The student will work closely with the PI, meeting at least once a week, and then meeting once every 2 weeks with the larger team to share early findings and get feedback. If interested, the student can help contribute to a conference or journal paper, published in the field of engineering or computer science education.
Skills required: Students should be interested in themes of engineering/computer science education, AI ethics, social implications of technology, and diversity, equity and inclusion. The project will mostly use social science research methods, such as interviews, surveys, and qualitative and quantitative (statistics) analysis. However, students do not need these skills to apply, only a willingness to learn them. Any background in reading research papers or conducting qualitative analysis (e.g., finding themes in a data set) is an asset, but is not required.
711. Surrogate models for uncertainty quantification in CFD
This project will develop machine-learning-enabled surrogate models to accelerate uncertainty quantification for high-speed and hypersonic flow applications. High-fidelity CFD simulations will be combined with data-driven methods to create physics-informed digital twins capable of rapidly predicting aerodynamic and aerothermal performance while quantifying uncertainty. The resulting framework will support real-time analysis, design exploration, and decision-making for advanced aerospace systems operating in uncertain environments.
Research area, student roles & skills
Research area: My research focuses on high-speed aerodynamics, aerothermodynamics, aeroacoustics and computational fluid dynamics (CFD), with an emphasis on aerospace applications. I develop advanced numerical methods, uncertainty-aware modelling frameworks, and physics-informed digital twins to predict aerodynamic, thermal, and propulsion performance in complex flow environments. My work integrates high-fidelity simulations, machine learning, turbulence modelling, and multi-fidelity optimization to support the design and operation of aerospace and defence systems, including hypersonic vehicles, propulsion systems, and uncrewed aerial platforms.
Student roles: The student will support research activities related to CFD, machine learning, and uncertainty quantification for high-speed flow applications. The ideal candidate will be self-motivated, capable of working independently, and eager to learn new computational and data-driven techniques. A strong willingness to develop technical skills, solve challenging problems, and contribute to research outcomes is essential.
Skills required: Applicants should have a strong background in fluid mechanics, thermodynamics, and numerical methods. Experience with Computational Fluid Dynamics (CFD), programming (Python, MATLAB, or C++), and data analysis is desirable. Familiarity with machine learning, uncertainty quantification, or scientific computing is considered an asset but is not required. The ideal candidate will have strong analytical and problem-solving skills, an interest in aerospace applications, and the ability to work independently as part of a multidisciplinary research team.
712. Sustainable E-Commerce Recommendation System for Reducing Carbon Footprint
Supervisor: Mahreen Nasir Butt
University: Algoma University (Sault Ste. Marie campus)
Consumer purchasing decisions across e-commerce platforms have a significant impact on environmental sustainability. This project focuses on developing a recommendation system that promotes environmentally responsible product choices while still aligning with individual user preferences. Students will incorporate sustainability indicators such as estimated carbon footprint, packaging impact, and supply chain sustainability scores into the recommendation process.
The system will generate product recommendations across various e-commerce categories that balance personalization with environmental impact, encouraging users to make more sustainable purchasing decisions. This project introduces students to sustainable recommender systems and multi-objective optimization techniques that integrate user satisfaction with ecological and ethical considerations.
Research area, student roles & skills
Research area: Dr. Nasir leads the Research in Artificial Intelligence and Data Analytics for Real-World Solutions, conducting interdisciplinary research that integrates AI, Machine Learning, Generative AI, and Quantum Computing to advance intelligent recommendation systems and address complex real-world challenges.
Student roles: The student will be responsible for designing and implementing an end-to-end sustainable e-commerce recommendation system. The work begins with exploring and preprocessing e-commerce datasets such as user-item interactions, product metadata, and ratings. The student will analyze the dataset to identify relevant features for building baseline recommendation models using collaborative filtering or hybrid approaches.
Next, the student will develop a sustainability augmentation layer by assigning environmental impact scores to products. These scores may include carbon footprint estimates, packaging impact, or category-based environmental weights. The student will then integrate these sustainability factors into the recommendation pipeline to build a multi-objective ranking system that balances user preference with environmental responsibility. The student will experiment with different strategies for combining recommendation scores and sustainability scores, such as weighted fusion or Pareto-based ranking. Model performance will be evaluated using standard recommendation metrics like Precision@K, Recall@K, and NDCG, along with a sustainability-aware metric that measures the average environmental impact of recommended items. In addition, the student will compare traditional recommendation outputs with sustainability-aware recommendations to analyze trade-offs between accuracy and environmental impact. The final deliverable includes a working prototype system, visualizations of results, and a short technical report summarizing methodology, experiments, and insights. Optionally, the student may develop a simple dashboard to demonstrate recommendations with sustainability indicators in an interpretable format.
Skills required: Students should have a solid understanding of Python programming and basic data science concepts, including data preprocessing and analysis using Pandas and NumPy. Familiarity with machine learning fundamentals, particularly recommendation systems such as collaborative filtering and content-based methods, is important. Exposure to scikit-learn or similar ML libraries is expected. Basic knowledge of evaluation metrics like precision and recall is helpful. Experience with data visualization tools and an interest in sustainability or real-world applications of AI will be an advantage.
713. Swarm Intelligence algorithms and routing in wireless sensor networks
Supervisor: Mohamed Tawhid
University: Thompson Rivers University (Kamloops campus)
Due to advances in low-power wireless communications, low-power analog and digital electronics, the development of low-cost and low-power sensor nodes that are small in size has received increasing attention. Sensor nodes have the ability to sense the environment nearby, perform simple computations and communicate in a small region. Although their capacities are limited, combining these small sensors in large numbers provides a new technological platform, called Wireless Sensor Networks (WSNs). WSNs provide reliable operations in various application areas including environmental monitoring, health monitoring, vehicle tracking system, military surveillance and earthquake observation
In this project, we will present a new protocol for WSN routing operations. The protocol is
achieved by using Swarm Intelligence algorithms (such as Hybrid Genetic Particle Swarm Optimization Algorithm, and Hybrid recent swarm Algorithms ) based optimization technique in network routing to optimize routing paths, providing an effective multi-path data transmission method to achieve reliable communications in the case of node faults. We will aim to maintain network life time at a maximum, while data transmission is achieved efficiently, so an adaptive approach is developed according to this goal. The proposed approaches are compared to a well known other algorithms such as ant based algorithm named EEABR using an event-based simulator. Our goal is to offer significant reductions of energy consumption which is used as a performance metric for different sized WSNs. Also, we will implement our approach on a small sized hardware component requiring minimum connections suitable for tiny node designs and we will develop an methods for handling the routing tasks by using the proposed approaches router chip. We will test the performace of the proposed algorithms by comparing the proposed algorithm with other existing swarm algorithm in the literature and applying non-parameteric statistical tests.
Research area, student roles & skills
Research area: Professor M. A. Tawhid
Mohamed A. Tawhid got his PhD in Applied Mathematics from the University of Maryland Baltimore County, Maryland, USA. From 2000 to 2002, he was a Postdoctoral Fellow at the Faculty of Management, McGill University, Montreal, Quebec, Canada. Currently, he is a full professor at Thompson Rivers University.
His research interests include nonlinear/stochastic/heuristic optimization, operations research, modeling and simulation, data analysis, and wireless sensor network. He has published in journals such as Computational Optimization and Applications, J. Optimization and Engineering, Journal of Optimization Theory and Applications, European Journal of Operational Research, Journal of Industrial and Management Optimization,
Student roles: Student is expected to know some skills of programming like MatLab and Metaheurtics algorithms (is able to learn it quickly if s/he does not know). Student will write many computer programs by using matlab software or Javaa and he will apply particale swarm optimization algorithm, genetic, differential evaluation algorithm, and other recent algorithms such as whale optimization, grey wolf, spider, etc. Hybrid Swarm Intelligence algorithms to routing in wireless sensor networks. He will compare these algorithms with other algorithms. Also, student will help in writing a report and graph as a research assistant.
Skills required: Writing extensive computer programming (Python, or Matlab software or R or others); Swarm Intelligence algorithms Writing report and using statistical nonparametric tests
714. Swarm Robotics for Recycling
Supervisor: Andrew Vardy
University: Memorial University of Newfoundland (St. John's campus)
The purpose of this project is to move swarm-robotic sorting from simplified laboratory objects toward real-world recyclable materials. Our current work uses robots to manipulate small, regular proxy objects, such as coloured 3D-printed pucks. This allows perception and control to be treated relatively simply: objects have predictable shapes and visual appearances, and their motion under robot contact can be modelled with reasonable accuracy. Real recyclables, by contrast, vary widely in shape, size, colour, mass, deformability, and surface properties. Adapting swarm sorting to these materials requires new perception strategies for detecting and classifying objects, as well as control strategies that can handle less predictable physical interactions.
The project will therefore focus on adapting our existing robots and algorithms so that they can perceive, move, and sort representative recyclable objects. Issues such as where material enters or leaves a facility remain important for eventual deployment, but in this project they will serve mainly as application context rather than the primary technical challenge. The immediate research goal is to identify which aspects of our current perception and control pipeline transfer to real recyclables, which must be redesigned, and how these changes affect collective sorting performance.
This work is motivated by the needs of smaller communities in Newfoundland and Labrador, which often ship recyclable materials to larger centres such as the facility at Robin Hood Bay in St. John's. Conventional sorting equipment can be large, costly, and difficult to scale down. A robotic swarm is potentially more flexible because sorting capacity can be adjusted by changing the number of robots. In the longer term, this may support smaller-scale, community-level recycling systems.
Research area, student roles & skills
Research area: Our main research area is swarm robotics, which can be defined as the study of multi-robot systems in which the individual robots use strictly local sensing and communication. In particular, the BOTS group studies how swarms of robots can manipulate their environment to achieve the user's objectives. Example tasks include object sorting, clustering, and construction.
Student roles: The role of the students will be to adapt our current perception and control approaches to operate with actual recyclable objects. Our current approach is based on manipulating small cylindrical objects (i.e. 3d-printed pucks) by pushing them with the robot's body. This simplifies both the perception and control aspects of the problem. For example, rather than detecting the complex structure of 3d objects, we just recognize the elliptical appearance and colour of the pucks. Similarly, when the robot pushes a cylindrical object, we have a good model of the object's response to the push. However, when transitioning to real-world recyclable objects, the interactions between robot and objects will be more intricate and less predictable. Hence, both the perception and control aspects of the problem become more challenging. This motivates the number of students requested. There need not be a fixed division of labour, but one student would specialize more on the perception side, while the other worked more on the control aspect.
The students would work to adapt our existing robots and algorithms. The robots used would be our ROS Orin robots from Hiwonder and the algorithms adapted make use of ideas from the doman of model predictive control (MPC). Machine learning is not a major factor in our current work, but could be utilized to bring a higher level of generalization to both object perception and control.
Skills required: The students should have had some exposure to robotics, either in the form of previous experience or through completing an undergraduate course. They should have advanced programming experience and have some basic electronics skills. Skills in CAD, fabrication technologies, and hands-on experience with tools would also be beneficial.
715. Synchronized GPR and Visual Sensor Payload on UAVs for Automated Bridge Soffit NDT
This project develops an integrated drone-based system for autonomous inspection of bridge soffits using Ground Penetrating Radar (GPR) and visual sensors (LiDAR and camera). These structures are difficult to access and inspect using traditional means, making UAVs an ideal solution for safe, efficient, and cost-effective inspection.
A custom payload comprising GPR, LiDAR, a camera module, and GPS has been developed and tested. The next phase involves final integration of this payload with a drone, synchronization of data streams, and enabling real-time autonomous navigation.
The student will assist in implementing a 3D structure mapping pipeline and developing algorithms that allow a user to define inspection trajectories directly on the map. The UAV must autonomously navigate from home point to a defined GPS rally point using GPS, and then switch to visual navigation for accurate path following in GPS-denied zones. Once the inspection is complete, the drone will return home autonomously.
Another key focus is developing a robust, low-latency system for remote data transmission, evaluating options such as 5G modems with ROS-based streaming or custom radio links. All sensing data (GPR and visual) must be synchronized and transmitted for remote monitoring and post-processing.
Research area, student roles & skills
Research area: Our research focuses on UAV-based sensing and autonomous inspection for structural health monitoring (SHM) of civil infrastructure. We integrate LiDAR and camera systems into drone platforms to enable detailed 3D mapping, visual navigation, and data-driven condition assessment in both indoor and outdoor environments. Key areas include sensor fusion, autonomous navigation in GPS-denied settings, and real-time, low-latency data transmission. Our interdisciplinary work bridges robotics, embedded systems, and real-time data processing to develop scalable, intelligent aerial systems for infrastructure monitoring and non-destructive testing applications.
Student roles: The student will play a key role in completing the integration of the sensor payload—comprising Ground Penetrating Radar (GPR), LiDAR, a camera, and GPS—with the UAV platform. A primary responsibility will be to ensure the synchronized acquisition, processing, and real-time transmission of multi-sensor data during flight operations. An important part of the role will involve designing and implementing a robust, low-latency communication pipeline to stream data from the drone to a ground station. The student will evaluate different transmission options, such as ROS-based streaming over 5G modems or custom radio links, ensuring that GPR and visual sensor data are reliably synchronized for remote monitoring and post-flight analysis. The student will also contribute to the development of autonomous flight algorithms. These algorithms will control the UAV’s movement from the home point to a GPS-defined rally point, and then transition to visual navigation for following a user-defined inspection trajectory in GPS-denied zones (e.g., under bridge soffits). After completing the inspection, the drone will autonomously return to the home point. The student may also participate in field testing and assist in debugging hardware and software during flight trials. They will help monitor system performance, record test results, and contribute to iterative design improvements.
Skills required: Applicants should have a background in robotics, mechatronics, or computer/electrical engineering. Experience with UAV platforms, ROS (Robot Operating System), and programming in Python and/or C++ is required. Familiarity with 3D mapping, SLAM, or visual navigation techniques is highly desirable. Knowledge of GPS systems, LiDAR, or embedded processors (e.g., NVIDIA Jetson) is a plus. The student should also be comfortable working with real-time data and communication protocols. Strong problem-solving skills and the ability to work independently in a hardware-focused environment are essential.
716. Tangible and Embodied Privacy Mechanisms in Social Robots
Supervisor: Leah Zhang-Kennedy
University: University of Waterloo
Location: Waterloo, Ontario
Start date: 2027-05-03 (flexible)
Disciplines: Computer Science, Engg-Mechanical, Engg-Computer, Engg-Systems and Technology, Engineering
Social robots are increasingly being developed to support users' physical, cognitive, and social well-being in domestic and care environments. However, their embodied, socially interactive, and mobile nature introduces privacy challenges that are not well addressed by current screen-based privacy notice and consent paradigms. We are exploring embodied privacy mechanisms grounded in embodied cognition, where privacy interactions become tangible, observable, and socially meaningful actions. For example, approaches may include tangible privacy tokens, gesture-based sensor modulation, configurable sensor covers, and spatial- or proximity-based privacy controls on social robots.
Humanoid social robots communicate through speech, gestures, gaze, touch, and locomotion, creating stronger perceptions of social presence and agency. Their ability to autonomously navigate domestic spaces introduces unique privacy concerns, as robots may enter sensitive areas, reposition themselves to observe interactions, or collect information across multiple contexts within the home. As a result, privacy interactions with HSRs are not limited to static device settings, but become ongoing, embodied, and socially negotiated experiences that unfold through movement and interaction in physical spaces.
Current privacy consent models are largely screen-based, making them difficult to implement on HSRs that lack traditional display interfaces. Current solutions are usually decoupled from the main system and provided as text-based privacy policies on companion mobile devices and websites, while the robot itself does not provide any privacy controls. As a result, users are less likely to read the notice and therefore may not realize that their expectations do not match the robots’ actual data practices. To address these issues, previous research highlights the need to develop privacy notices through the robots’ available communication channels and interaction capabilities
The intern will design, prototype, and evaluate the embodied privacy mechanisms, examining how well these approaches support users' understanding, trust, consent behavior, and sense of control in real-world settings.
Research area, student roles & skills
Research area: Prof. Zhang-Kennedy carries out research in human-computer interaction (HCI) at the University of Waterloo. HCI focuses on the design, evaluation and implementation of interactive computing systems. As a result, we make systems easier to use, more efficient, and more compelling. In pursuit of these goals, HCI draws upon various fields, including the behavioural sciences, social sciences, computer science, design and engineering. Prof. Zhang-Kennedy investigates interdisciplinary research questions at the intersection of HCI, privacy, digital literacy, and user experience. She directions the Safe Interactions lab: https://uwaterloo.ca/safe-interactions-lab
Student roles: Protoyping and development of tangible and embodied privacy mechanisms on the social robot, NAO. Participate in the recruitment of research participants. Conduct user studies to evaluate the prototypes. Review academic literature and publicly available documentation. Participate in the writing and dissemination of research findings. Assist with other duties relating to research. Attend weekly project meetings.
Skills required: Students with backgrounds in human-computer interaction (HCI), computer science, engineering or related fields are preferred. Some software and hardware development experience is required to carry out the prototyping and development aspect of the project. Demonstrated experience in an academic research environment is preferred. Some knowledge of HCI research methods is preferred. Some knowledge of qualitative and quantitative data analysis is an asset. Good oral skills and writing skills are an asset.
717. Task Relevant Active Learning
Supervisor: Ali Ayub
University: Concordia University (Montréal campus)
This project investigates Task Relevant Generalized Category Discovery (TR GCD), a machine learning framework that enables intelligent systems to identify both known and previously unseen categories that matter for a specific task. Traditional category discovery often produces clusters that are statistically valid but not useful for decision making; TR GCD bridges this gap by integrating relevance signals directly into the representation and clustering process. The research will develop models that jointly learn embeddings, relevance weighting, and category structure using semi and self supervised learning methods, and will evaluate them on image and interaction datasets where only a subset of categories is meaningful for downstream tasks. The goal is to create adaptive systems that can operate reliably in open world environments by discovering new, meaningful categories while ignoring irrelevant variability. Expected outcomes include novel algorithms, improved performance in open world classification and retrieval tasks, and open source implementations to
support reproducible research.
Research area, student roles & skills
Research area: My research falls in the broad areas of AI, ML, and Robotics, and specifically, it focuses on interactive machine learning. The goal of my research is to develop personalized
autonomous systems that can continually interact with and learn about their users’ individual preferences and unique environments to provide personalized assistance to
their users over the long term. Currently, I am interested in working on challenges associated with interactive continual learning both from the ML and human-robot interaction (HRI) perspectives to improve scalability, efficiency, and explainability of ML methods applied to robotics and autonomous vehicles.
Student roles: The student will implement and test TR-GCD models, conduct experiments on labeled-unlabeled datasets, and help design relevance-aware learning components. They will perform literature reviews, run ablation studies, analyze results, and contribute to writing research reports, code documentation, and publications. The student will receive close mentorship through weekly supervision meetings, code reviews, and structured research guidance. They will gain training in machine learning methods, experimental design, scientific writing, and reproducible coding practices, ensuring strong technical development and professional growth.
Skills required: 1) Strong fundamentals in computer programming (preferably Python). 2) Motivation to learn new tools and research methods in AI. 3) Background in software engineering is a plus. 4) Background in AI (through courses or projects) is a plus, but not required. 5) Some experience with Pytorch is a plus, but not required. 6) Knowledge of Linux based systems is a plus, but not required. 7) Knowledge of version control (git: GitHub, Gitlab) is a plus, but not required.
718. Technical and Legal Aspects of Software and AI Systems
Supervisor: Julia Rubin
University: University of British Columbia (Vancouver campus)
Location: Vancouver, British Columbia
Start date: 2027-05-03 (flexible)
Disciplines: Computer Science, Engg-Computer, Engg-Software, Law
This project explores technical, legal, and societal implications of security and privacy standards and regulations for software and AI systems, such as ISO 27701, CCPA, GDPR, and EU AI Act. The student working on this project will join an interdisciplinary team of graduate and undergraduate students focusing on ML trustworthiness. The student will explore a particular aspect of ML Trustworthiness, e.g., explainability, from both technical and regulatory perspective and will further apply the gained insights to improve AI-based systems in the areas of cybersecurity and law. The student will also work with research and industrial partners to map the main requirements of the regulations and analyze their implications on industrial practices.
Research area, student roles & skills
Research area: The UBC Research Excellence Cluster on Trustworthiness in Machine-Learning-Based Systems (TrustML) is comprised from members from six faculties at UBC (Applied Science, Science, Medicine, Forestry, Business, and Law), industry, and the BC government: https://trustml.ubc.ca/. TrustML facilitates development of trustworthy machine-learning-based systems, i.e., systems that are reliable, secure, explainable, and ethical. It examines trust-related requirements in different application domains, including healthcare, finance, and aerospace, and investigates solutions for building trustworthy systems that professionals and the general public can reliably adopt.
Student roles: The student will work with the PI to help analyze and classify existing work on ML explainability and evaluate its effectiveness. The student will also analyze existing US/Canada/EU regulations related to explainable and transparent AI, identify relevant existing approaches and gaps. The student will also help close the gap by implementing relevant approaches.
Skills required: The student is expected to: - have solid programming experience and good command of Java, C++, C, or a similar language; - be able to learn new programming languages and frameworks; - have excellent technical communication skills (written and oral); - have good time management skills and proven ability to meet deadlines; - be detail-oriented and thorough.
719. Temporal Bias in Cancer Prediction Models Using SEER Data
Supervisor: Munima Jahan
University: Thompson Rivers University (Kamloops campus)
Machine learning models developed for cancer prediction and survival analysis are commonly trained using historical clinical datasets such as the SEER database. However, medical practices, treatment protocols, diagnostic technologies, and patient demographics evolve significantly over time. As a result, models trained on older data may become less reliable or biased when applied to newer patient populations. Despite this risk, most existing studies assume that clinical datasets remain temporally stable and rarely evaluate how model performance changes across different time periods.
This project investigates temporal bias and reliability drift in cancer prediction models using SEER data. The study will develop machine learning models for tasks such as cancer survival prediction and evaluate their performance across multiple decades of patient records. Models trained on earlier cohorts will be tested on more recent cohorts to analyze how predictive accuracy, fairness, and reliability degrade over time.
The project will explore:
• Temporal changes in model performance
• Fairness disparities across demographic groups over time
• Feature drift caused by evolving treatment standards and diagnostic practices
• Reliability metrics for long-term deployment of clinical AI systems
The expected outcome is a framework for identifying and measuring temporal bias in healthcare machine learning models. This research aims to support the development of more robust, fair, and trustworthy AI systems for clinical decision support while contributing to the broader understanding of reliability in real-world AI deployment.
Research area, student roles & skills
Research area: My research focuses on software engineering, trustworthy artificial intelligence, and data-driven analytics to improve the reliability and quality of complex systems. I am particularly interested in machine learning for software engineering, distributed systems reliability, requirements engineering, and the verification of intelligent systems. My research also explores the application of AI and data analytics to healthcare, including the analysis of cancer datasets to develop reliable predictive models, identify biases, and support data-driven clinical decision-making. Through interdisciplinary collaborations, I aim to develop innovative, explainable, and trustworthy solutions that address real-world challenges in both software and healthcare domains.
Student roles: The student will participate in the following activities:
* Developing software algorithms for the tasks.
* Document, analyze, discuss, and represent the obtained results and the developed systems.
* Perform weekly meetings with the supervisor.
* Build systems (e.g., machine learning models) that analyze the source code and recommend the suitable implementation to the requested features.
* Write a final report that summarizes the background, approach, the obtained results, and the future directions.
* Make presentations, attend a conference.
Skills required: Basic knowledge in Biology Basic knowledge in machine learning skills Strong programming skills (preferred python and R) Basic background in Bioinformatics preferred, but not required
720. Temporal Explainability in Multi-Agent AI Architectures: Designing Deterministic Audit Trails for Autonomous Systems
As security operations centers adopt automated multi-agent AI architectures to handle high-velocity digital threats, verifying the reliability of decentralized decisions has become a critical compliance challenge. When multiple localized LLM agents interact to triage an incident, their collective decision-making pipeline can become an un-auditable "black box." If an autonomous system executes a faulty containment action, standard logs fail to capture when and why the reasoning shifted. This research project addresses this operational gap by developing a generalized Temporal Explainability Framework for Multi-Agent AI Security Systems.
The intern will construct an independent validation engine that monitors, parses, and timelines the interactions of an active multi-agent ecosystem. Rather than providing a static, single-point explanation, the framework will track the chronology of agent reasoning. It will capture how individual agents update their internal states, exchange data tokens, and adjust their risk perception as an incident unfolds over time. The core objective is to translate dynamic agent interactions into a deterministic, time-sequenced dependency tree that acts as a flawless, compliance-ready audit trail.
Integrating frameworks like LangGraph with XAI design patterns, the student will build a prototype system that evaluates automated decisions against established security baselines. The engine will explicitly measure metric drift and track the provenance of every data point driving the agents' final output.
The final deliverable is an open-source software repository and validation benchmark that can sit alongside any commercial multi-agent architecture, allowing organizations to deploy autonomous defense while maintaining absolute regulatory accountability.
Research area, student roles & skills
Research area: Our laboratory pioneers the integration of trustworthy technology, efficient operations, and capable people within cybersecurity. On the systems side, we study how to make complex, distributed computing environments more measurable, reliable, and compliant. Our current research investigates the intersection of multi-agent Large Language Model (LLM) architectures and Explainable AI (XAI). Specifically, we focus on temporal explainability, capturing how decentralized AI agents interact, reason, and evolve their decisions over time.
Student roles: The selected intern will function as an AI Systems Architect and Lead Developer, taking ownership of the design, implementation, and benchmarking of the temporal explainability module. Over the 12-week internship, the student will execute a rigorous, product-driven R&D lifecycle:
Initially, the student will design the data ingestion pipeline and state-tracking hooks required to intercept and record agent-to-agent communication in real time. They will write the Python backend code to normalize these telemetry logs into a standardized, chronological data structure. The core technical milestone involves engineering the XAI visualization and parsing engine, which reconstructs the step-by-step reasoning path of the multi-agent system, highlighting exactly when hypotheses were updated or actions were triggered.
The intern will then execute empirical stress-testing, passing simulated multi-stage security incidents through an orchestrated agent setup to evaluate the explainability engine's fidelity. They will gather performance telemetry, benchmarking the auditing tool's computational overhead and latency to ensure it can operate in high-throughput enterprise environments without creating operational bottlenecks.
The student will package the code to professional production standards. This includes containerizing the application using Docker, designing clean, scannable API endpoints for audit exports, and authoring comprehensive deployment guides so industry partners can integrate the engine into their existing operational pipelines. The student will actively defend their technical architecture and present data visualizations during weekly laboratory briefings.
Skills required: Applicants must be pursuing a degree in Computer Science, Data Science, or Software Engineering with a strong focus on artificial intelligence. Advanced proficiency in Python and experience with LLM orchestration frameworks (such as LangGraph, LangChain, or AutoGen) are required. Familiarity with state-management, structured data logs (JSON/YAML), and asynchronous programming is highly desirable.
721. Temporal Procedural Content Generation
Supervisor: Joseph Alexander Brown
University: Thompson Rivers University (Kamloops campus)
Location: Kamloops, British Columbia
Start date: 2027-05-03 (flexible)
Disciplines: Computer Science, Art, Information Studies
Procedural Generation currently focuses on creating either online assets, which are used in the system at the time of generation, or offline assets, which help developers. The idea would be to extend these generations into temporal development, in which an existing world is presented, and then, based on actions in the game within the narrative timeline, the generated assets will change online to reflect the new game world state. For example, trees growing over time or new buildings being built in cities depend on the actions of characters in the game. While many games contain such temporal changes, these are scripted interactions and do not emerge from the game's systems. This work would explore the use of finite-state machines and cellular automata to enable changes over time. The project will also look into how to simplify these systems to be applicable in modern game frameworks and will aim to provide an open source middleware to help developers.
Research area, student roles & skills
Research area: I work in game development, particularly in procedural content generation. This is the development of game assets with minimal designer or developer interaction to reduce the complexity and time required to produce games while maintaining the suitability of these assets in terms of narrative, mechanics, dynamics, and aesthetics.
Student roles: Design, Development, and Programming of the framework Evaluation and analysis of the outcome Publication of the code to an open source platform White paper production, including literature review
Skills required: Experience in Python or C# (Some experience with Unity, Unreal, or Godot is preferred) Linear Algebra Knowledge of artistic expression
722. Testing and Analysis of Vision-Language-Action Models in Autonomous Driving
Supervisor: Zhijie Wang
University: Concordia University (Montréal campus)
Vision-Language-Action (VLA) models have recently emerged as a promising paradigm for autonomous driving by integrating visual perception, language understanding, and driving decision making into a unified framework. Despite their strong capabilities, the reliability and robustness of these models under complex and safety-critical driving scenarios remain insufficiently understood. Existing autonomous driving simulation platforms provide realistic environments for evaluating VLA models, yet it is still unclear how effectively current testing approaches can assess model reliability, uncover unsafe behaviors, and expose failure cases under diverse driving conditions. This project will first investigate the reliability and robustness of VLA models within existing autonomous driving simulators through systematic testing and behavioral analysis. Based on the observations and limitations identified during this study, the project will further explore and develop new testing frameworks and methodologies to improve the efficiency and effectiveness of testing VLA-based autonomous driving systems.
Research area, student roles & skills
Research area: My research interest focuses on Software Engineering (SE) support for complex AI-based software systems (quality assurance for AI). My most recent works cover software testing, analysis, and improving for AI models and systems. Additionally, I am broadly interested in the intersection of Software Engineering, AI, and Human-Computer Interaction (HCI). I design and develop interactive interfaces to help developers interpret and debug AI-based software.
Student roles: The student will participate in research activities under the supervision and guidance of the faculty member. Responsibilities include conducting literature reviews, implementing and evaluating research prototypes, analyzing experimental results, and assisting with the preparation of research papers. The student is also expected to actively communicate progress, collaborate with other team members, and contribute to regular meetings and discussions.
Skills required: 1. Strong programming skills and solid foundations in computer science and/or software engineering. 2. Willingness and ability to learn new cutting-edge knowledge. 3. Good communication and collaboration skills. 4. Good English proficiency, particularly in reading technical papers. 5. Previous experience with autonomous driving simulators (e.g., CARLA) is preferred.
While unlearnable examples have been actively studied for classification-oriented data protection, how to design effective unlearnable examples for text-image generative models remains underexplored. In this project, we will address several fundamental aspects of unlearnable visual protection for diffusion-based image editing. Ideally, the project will end up with a paper published at a top-tier ML/CV conference, such as CVPR, ICML, or NeurIPS. The key strengths/objectives of our project include:
-- We will leverage the tools from representation learning and optimization principles to develop new theoretical and algorithmic insights for text-image generalization unlearnable examples. In other words, we will study how protected images can disrupt the correspondence between visual semantics and text-guided editing instructions, thereby motivating a principled protection framework for diffusion-based image editing.
-- Our proposed framework will also support robustness against diverse editing requests, which are common in real-world text-image applications. For diffusion-based image editing, there exist several challenging scenarios: 1) instruction-dependent editing, where different text prompts may require modifying different semantic attributes, such as object appearance, background, style, or local structure; 2) model-dependent generation, where different diffusion models may rely on different visual-text representations and editing mechanisms. We will design principled methods to address these issues, so that protected images remain visually natural to humans but become difficult for diffusion models to correctly edit or exploit under diverse text-guided editing conditions.
Research area, student roles & skills
Research area: My research centers on trustworthy machine intelligence, which is essential for deploying intelligent systems with any amount of versatility and reliability. I study the foundational pillars of trustworthiness, including fairness, privacy, and robustness, etc. Specifically, I design novel machine learning methods grounded in solid theoretical foundations across a wide range of scenarios. Over the years, our group has published (100+) papers intensively in top venues (e.g., NeurIPS, ICML, ICLR)
Student roles: Responsibilities and tasks The successful candidate will be responsible for: - Doing a literature review in the related areas - Implementation of novel machine learning algorithms for trustworthy AI - Contributing to efficient interdisciplinary research team collaboration - Writing a scientific report/paper - Completing at least one of the objectives listed in the project description
While the start date is flexible, in previous years, the selected students would work with me remotely once they were matched. Ideally, the project will result in a paper published in top venues (e.g., ICML, NeurIPS, ICLR).
Skills required: You may be a strong candidate if you have (at least three of the following): - a strong background in statistics, linear algebra - strong motivation and interest in machine learning research. - knowledge of machine learning, optimization - published academic work on the topics of machine learning - sufficient programming experience (e.g., Python) - some experiences on advanced machine learning paradigms (e.g., computer vision, adversarial examples, watermarking, etc.) - good English reading and speaking skills
Disabled applicants will receive priority in case they have equal qualifications.
724. The Effect of Mitochondrial DNA Variation on the Metabolome, Epigenome, and Transcriptome
The research project will involve the analysis of mtDNA, nuclear DNA methylation, and nuclear gene expression data from both available human cohort data and in-house cell-culture data.
The objectives of this research project therefore are as follows:
1. Characterize the relationship between mtDNA-CN variability, nDNA methylation, and nDNA gene expression.
This study will identify CpG sites/regions and genes associated with mtDNA variation including the characterization of methylation/gene expression in the context of disease outcomes.
2. Determine mechanisms by which mtDNA alters nDNA methylation/gene expression.
We will use in silico functional enrichment methods to determine biological mechanisms which may be driving these associations.
To accomplish these research aims, the student will perform association analyses using a linear modeling framework adjusted for relevant confounding variables. Several in-house algorithms as well as publicly-available software tools and packages will be employed, such as those to facilitate differential methylation and mediation analyses. All data analyses will be conducted using the R programming language through high-performance computing environments in a shared computing environment.
Research area, student roles & skills
Research area: Mitochondria are important organelles crucial for proper cell functioning as they are involved in many physiological processes including providing cellular energy. Human mitochondria have their own circular DNA molecules which are present in multiple copies in each organelle. The quantity of these mitochondrial DNA molecules, termed mitochondrial DNA copy number (mtDNA-CN), represent proxies of mitochondrial function given that the quantity of mitochondrial DNA is correlated with a cell’s energy demands. Recently, our group and others have found wide-spread associations between mtDNA-CN and nuclear DNA methylation. However, the precise mechanisms driving these associations remain unclear.
Student roles: The student will be responsible for coding, analysis, data visualization, and summarizing of the results into written form.
Skills required: The student should have experience with unix, R, and/or another programming language and willingness to learn/adapt to a new language. Some experience with genomics is also an asset.
725. ThermalMesh-AI: A Physics-Informed Spatio-Temporal Graph Neural Network for Continuous Skin Temperature Field Reconstruction from Sparse Thermal Manikin Sensors
Supervisor: Abdelatif Merabtine
University: École de Technologie Supérieure (Montréal campus)
Thermal manikins are widely used in thermal comfort research to simulate human heat exchange with the surrounding environment. However, their practical deployment is often limited by the sparsity of embedded discrete sensors, which typically cover only 15–20 discrete measurement points across the entire body surface. This sparse sampling is fundamentally insufficient to capture:
• Spatial heterogeneity of skin temperature distribution across hundreds of anatomically distinct surface zones
• Transient dynamics, such as sudden localized warming near a heat source or progressive cooling of extremities
• Thermal asymmetry, such as the difference between the sun-exposed and shaded side of the body
As a consequence, critical micro-scale thermal gradients that are known to influence human thermal perception are completely lost. Standard interpolation techniques (e.g., bilinear or kriging) are inadequate because they are purely mathematical and ignore the underlying physics of bioheat transfer — they do not account for skin vasodilation, metabolic heat generation, or convective heat exchange at the body surface.
The Core Problem: How can one reconstruct a high-fidelity, physically consistent, and time-resolved continuous 3D skin temperature field from a handful of sparse sensor measurements on a thermal manikin?
Research objectives are :
- Design and validate a 3D graph-based mesh representation of the human body surface suitable for spatio-temporal ML inference
- Develop a Physics-Informed Spatio-Temporal Graph Convolutional Network (PI-STGCN) that reconstructs the continuous skin temperature field from sparse sensor inputs
- Incorporate the Pennes Bioheat Transfer Equation as a physics-based regularizer within the ML loss function to enforce thermodynamic consistency
- Validate the reconstructed temperature field against high-resolution Infrared (IR) thermography ground truth under various dynamic thermal environments
Research area, student roles & skills
Research area: The supervisor's work sits at the intersection of building physics, environmental engineering, and human physiological science. Their primary scientific focus is understanding and quantifying how the built thermal environment affects human occupants ranging from the physical measurement of heat exchange between the human body and its surroundings to the psychological and physiological responses that these exchanges generate in terms of thermal sensation, comfort, and health.
His recent research focuses on the development and application of advanced machine learning architectures that integrate physical knowledge, spatial structure, and temporal dynamics applied to complex engineering and scientific systems where data is sparse,
Student roles: The student occupies a central and autonomous role in this project. They are not a support technician, they are the main researcher responsible for the scientific output working under the guidance of two supervisors (one thermal engineering expert, one ML expert). Below the main requirements and the proposed timeline.
Requirements : 1. Develop a validated PI-STGCN model capable of reconstructing continuous 3D skin temperature fields from sparse inputs 2. A curated, high-resolution spatio-temporal skin temperature dataset 3. A benchmarking report comparing PI-STGCN against classical interpolation and non-physics-informed ML approaches 4. At least one peer-reviewed publication on physics-informed reconstruction of thermal fields on human body surfaces
Timeline (12 weeks) : weeks 1–2: Literature review weeks 3–5: Preprocessing existing experimental data weeks 5–8: PI-STGCN design, PINN loss integration, training weeks 9–10: Model evaluation, benchmark comparisons weeks 11–12: Writing / publication
Skills required: The candidate must have a solid proficiency in : - Processing time-series temperature sensor data: filtering, gap-filling, synchronization - data preprocessing pipelines for multi-sensor systems (calibration, drift correction) - handling noisy, sparse, irregular measurement data - Supervised learning, loss function design, and backpropagation- - Practical experience with PyTorch or TensorFlow for model development - Understanding of Physics-Informed Neural Networks (PINNs): embedding PDE residuals into the training loss - Familiarity with automatic differentiation (autograd) as used in PINN training
Exploring advanced sort clustering techniques, three-way clustering, including both evaluation-based and
operation-based approaches. Applying the examined models to medical, financial, security applications.
Research area, student roles & skills
Research area: Machine learning, Clustering, Three-way decisions.
Student roles: Work with me to refine there-way clustering models.
Skills required: Analyzing, Math proving, programming
This project will advance Tilt-Translate Tangibles, a real-time tangible interaction system for detecting and interpreting spatial patterns in complex scientific datasets. The system uses 3D-printed LEGO-like blocks tagged with fiducial markers. Users assemble physical configurations that represent meaningful scientific concepts, then tilt the blocks onto a MultiTaction tabletop display for recognition. The system converts each configuration into a graph-based digital representation by analyzing connected components, spatial arrangement, orientation, and proximity. These physical patterns are then translated into machine-readable encodings that can be used to query structured datasets.
The current implementation has been tested in the context of DNA replication biology, where scientists used the system to explore biological data through guided and open-ended tasks. Initial evaluation with six biologists showed that the system supported faster data exploration, intuitive interaction, and collaborative scientific reasoning.
The proposed project will refine the system’s recognition pipeline, improve the mapping between tangible patterns and semantic queries, and expand its applicability to broader scientific datasets. The research will contribute to human-computer interaction, tangible analytics, and interactive scientific visualization.
Research area, student roles & skills
Research area: Our specialized research area is human-computer interaction, tangible user interfaces, and interactive data visualization for scientific discovery. Our work focuses on designing physical-digital systems that allow users to explore complex datasets through spatial interaction, embodied reasoning, and collaborative inquiry. We are particularly interested in semantic querying, tangible pattern recognition, and interaction techniques that make complex biological and scientific data more accessible to interdisciplinary research teams.
Student roles: The student will contribute to the design, development, and evaluation of the Tilt-Translate Tangibles system. Their role will include improving the tangible block recognition workflow, refining how physical block configurations are converted into graph-based representations, and supporting the translation of these representations into machine-readable semantic queries. The student may assist with software development, interface refinement, testing of fiducial marker recognition, and integration with structured scientific datasets. The student will also participate in research activities related to evaluation and analysis. This may include preparing study materials, conducting pilot tests, observing user interaction sessions, analyzing qualitative and quantitative data, and helping interpret findings related to usability, collaboration, and scientific data exploration. Depending on their background, the student may also support the design of new tangible patterns, documentation of the system architecture, and preparation of figures or written material for academic dissemination. Through this role, the student will gain experience in tangible interaction design, interactive visualization, user-centered evaluation, and interdisciplinary research involving computing and biological data.
Skills required: The student should have a background in human-computer interaction, computer science, information visualization, or a related field. Experience with interactive systems, prototyping, user interface development, or tangible/embodied interaction is desirable. Programming skills in Python, JavaScript, C#, or similar languages are useful. Familiarity with computer vision, fiducial marker tracking, graph representations, or biological data analysis would be an asset. The student should also be comfortable working in an interdisciplinary research environment.
728. Towards Unified Explainable Machine Learning Models for Medical Image Analysis
Medical images are widely used for diagnostic, monitoring, or research purposes while often analyzed by machine learning (ML) tools. However, many advances from core ML cannot be directly transferred to medical images as their analysis poses unique challenges: (1) High spatial dimensionality (e.g., a 3D brain magnetic resonance imaging - MRI - scan contains millions of image elements), (2) temporal components (e.g., scans repeated over time to assess changes), and (3) multimodality (e.g., multiple MRI sequences available for a subject). This results in a massive amount of heterogeneous data to be processed, requiring specialized and efficient methods. Such ML models not only need to be accurate, but explainability of their decisions are also important to ensure trust by clinicians and patients and to aid meaningful image-driven biomarker discovery. Existing methods are mostly task- and modality-specific, unable to jointly handle multiple modalities with temporal components, and lack built-in explainability capabilities. This results in many untapped diagnostic and research potentials, given the ever increasing amounts of imaging and also non-imaging data (e.g., biosignals from smart devices) acquired. We will research ML methods for medical images that do not suffer from the mentioned restrictions to improve image-driven diagnostics and biomedical research. Our main idea is to tackle the mentioned challenges in a unified way with self-explainable hybrid generative-discriminative ML models. While standard discriminative ML models map images to a diagnostic score, our models will solve the opposite problem of learning the distribution of images and scores in a probabilistic way. Being able to generate data equips our models with built-in self-explainability mechanisms that naturally removes the common black box problem in today’s ML landscape.
Research area, student roles & skills
Research area: I am a computer scientist and an expert in machine learning-based medical image analysis. My research centers around the development of machine learning solutions for health data science and precision medicine problems. I am specifically interested in developing and advancing machine learning methods that accurately model the complex dynamics and variations of normal or pathological processes in the human body by integrating and combining diverse, large-scale medical data (e.g., images, clinical data, text reports).
Student roles: The student will work on implementing and testing hybrid generative-discriminative deep learning models for medical image analysis tasks. The student will work in close collaboration with several Biomedical Engineering graduate students and will be exposed to the newest techniques in deep learning-based medical image analysis (e.g., diffusion models, GANs, normalizing flows). The student will utilize state-of-the-art Python-based deep learning frameworks such as PyTorch and MONAI to implement, train, and test the developed models on top-notch hardware (i.e. NVIDIA RTX4090 GPUs).
Skills required: The student for this project should have: - A background in Biomedical Engineering, Computer Science, Electrical Engineering, or a closely related field - Very good math skills - First experience in machine learning-based medical image analysis or a related field (e.g., computer vision, machine learning) - Some hands-on experience with modern software frameworks for deep learning like PyTorch - Programming skills in Python
729. Towards better optimizers for training neural networks
In this project, we will explore better optimization methods for training neural networks. This includes designing scalable second order optimization and improving modern optimizers like Muon. We will explore schedule-free optimization methods, optimizers that can work even without any additional tricks like skip connections and normalization layers.
Research area, student roles & skills
Research area: My team works in Artificial Intelligence and Machine Learning. Specifically, we focus on Deep Learning, Reinforcement Learning, Lifelong Learning, Optimization, Foundation Models, and AI for Science. For more details, please check out the lab website: https://chandar-lab.github.io/
Student roles: Study existing optimizers; implement existing optimizers and benchmark them; design new optimization methods; implement and benchmark the newly proposed algorithm; write a report.
Skills required: Strong background in deep learning and optimization; strong coding skills; proficiency in Python and PyTorch; knowledge of multi-GPU and distributed training.
730. Towards dynamic motion intelligence for different levels of autonomy in avatar systems..
Supervisor: Francisco Andrade Chavez
University: Thompson Rivers University (Kamloops campus)
The student will be helping develop software infrastructure for developing control frameworks to be used in large and small humanoid robots. This infrastructure will then be leveraged to develop robotic applications such as avatar systems and socio-physical human robot interaction. With a bigger focus on avatar systems and different levels of shared autonomy.
Research area, student roles & skills
Research area: My research aspires to develop robot capabilities that show potential to provide people with improved workplace safety, greater job satisfaction, deeper personal development, and overall better quality of life. For example, in health care, robots could care for people who require quarantine conditions. Healthcare professionals would be safer from risky exposure or repetitive work that might injure them, freeing up their time to focus on the quality of their interactions and patient assessments. In such scenario, Dynamic Motion Intelligence, the capacity to exploit robot dynamics to adapt in a rapidly changing scenario while seamlessly interacting with humans, is crucial.
Student roles: The students role is to contribute to the lab by testing new and novel ideas along with adapting known implementations to the lab infrastructure. He is to be and active member of the group discussions and be unafraid to ask questions and try things. It also means learning fast, reviewing scientific literature and improve coding skills to convert theory to implementation.
Skills required: Student should have experience coding as well as the ability to learn a new programming language. Ideally with engineering background Student should have a well developed problem solving skills. A curiosity driven mindset. Be unafraid to try, fail and learn. Code is mainly in Python so experience with python is a must, but C++ and Julia are a good to have and a plus. ROS 2 experience is encouraged, but not required. Passionate about robotics is very useful.
731. Towards ethically sourced code generation
Supervisor: Shin Hwei Tan
University: Concordia University (Montréal campus)
Students will investigate several problems existing in building ethically sourced code generation and evaluate existing code generation models.
Research area, student roles & skills
Research area: My research interests include ethically sourced code generation, open-source software and software ethics.
Student roles: The student will first conduct a literature review on existing problems in open-source code generation models, then experiments will be designed to evaluate existing models. The final goal project to enhance existing models towards ethically sourced code generation.
Skills required: Students with prior background in using code generation models for daily development is a plus. Students with background in human aspects and psychology are also highly desirable.
732. Tracking Snowpack Loss and Freshwater Availability in the Canadian Rockies Using Satellite Data and AI
Supervisor: Vijal Jain
University: Thompson Rivers University (Kamloops campus)
Water is the most critical natural resource on Earth, and for millions of British Columbians, it begins as snow.
Every winter, the Canadian Rockies accumulate vast quantities of snow that slowly melt through spring and summer, feeding the rivers, reservoirs, and aquifers that supply drinking water to cities, irrigation water to farms, and flow to ecosystems across British Columbia and Alberta. But climate change is quietly disrupting this ancient cycle. Snowpack levels across western Canada have been declining for decades, and the consequences, including water shortages, agricultural stress, and ecosystem damage, are already being felt.
The problem is that we cannot monitor snowpack everywhere. Ground-based snow measurement stations are sparse, expensive to maintain, and completely absent from many remote mountain areas. Satellites fill this gap.
This project uses freely available satellite data from Sentinel-1, Sentinel-2, and MODIS to monitor seasonal snowpack changes across the Canadian Rockies. The intern will build a machine learning model that estimates snow cover extent and water equivalent from satellite observations, producing water availability forecasts that are directly useful for agricultural planning, urban water management, and climate research.
The project is grounded in real Canadian data. Satellite observations will be validated against open ground-truth measurements from Environment and Climate Change Canada and the BC River Forecast Centre, ensuring outputs are accurate and meaningful.
All data is freely available and all tools are open-source, meaning the intern can begin contributing from day one with no barriers to access. Kamloops sits at the gateway to the Canadian Rockies, making this research locally relevant and giving the intern a unique opportunity to study one of Canada's most important natural resources from their own backyard.
Findings will be prepared for journal submission, giving the intern a co-authored publication from their 12 weeks at Thompson Rivers University.
Research area, student roles & skills
Research area: I specialize in making satellite radar imagery useful for the real world. My research develops AI-driven signal processing pipelines that clean, interpret, and extract meaning from Synthetic Aperture Radar (SAR) data imagery that sees through clouds, smoke, and darkness where optical satellites fail. Using deep learning, computer vision, and classical ML, I transform raw radar signals into actionable maps of flooded regions, deforested areas, and land-cover change. My award-winning DFT-based speckle reduction framework, validated across multiple satellite frequencies, forms the methodological foundation of this applied remote sensing research.
Student roles: The role is structured across three progressive phases designed to build skills and confidence while delivering tangible research outcomes within the 12 week internship. Phase 1 (Weeks 1 to 4): Onboarding and Data Preparation. The intern will begin by building foundational knowledge of snow hydrology, satellite remote sensing, and the Canadian Rocky Mountain water cycle through guided reading and hands-on exploration. They will download and preprocess satellite datasets from Sentinel-1, Sentinel-2, and MODIS covering multiple winter and spring seasons across the Canadian Rockies. Ground-truth snow measurement data from Environment and Climate Change Canada and the BC River Forecast Centre will be collected and merged with satellite observations. This phase ends with a clean, analysis-ready dataset spanning multiple years and locations. Phase 2 (Weeks 5 to 9): Feature Engineering and Model Development. The intern will extract meaningful features from satellite data including snow cover extent, surface reflectance, and radar backscatter values sensitive to snowpack depth and water content. They will train and evaluate machine learning models to estimate snow water equivalent and forecast seasonal water availability at locations beyond existing ground station coverage. Model performance will be validated against held-out ground measurements. Deep learning approaches will be explored where time permits. Phase 3 (Weeks 10 to 12): Visualization, Interpretation, and Writing. The intern will generate maps showing seasonal snowpack trends across the Canadian Rockies, interpret findings in the context of climate change and water security, and contribute directly to drafting a peer-reviewed journal paper. Throughout the internship the student will attend weekly one-on-one meetings with the supervising professor, maintain well-documented code in a shared GitHub repository, and deliver a final research presentation summarizing their findings. The role is designed to give the intern complete ownership of a meaningful research workflow from raw satellite data to co-authored publication.
Skills required: The ideal candidate is an undergraduate student in Computer Science, Software Engineering, Environmental Sciences, or Earth Sciences with working knowledge of Python and a basic understanding of machine learning concepts. Familiarity with GIS tools, climate datasets, or hydrological data is an asset but not required. No prior experience with satellite data is necessary as the project is structured to onboard students from the ground up. Curiosity about climate change, water security, and environmental sustainability matters as much as technical background. All software and datasets used are free and open-source.
733. Trajectory Datasets: Developing the Atlas of Human Movements
Video analysis is now used in many transportation applications for traffic data collection: the most common outputs are road user counts and their speeds. Trajectory datasets like highD (https://levelxdata.com/highd-dataset/) are more and more available, and can be generated using open source software. Yet there is no standard format for trajectory data, nor to link the data spatially and temporally to make it discoverable and enable generalized analyses of road user behaviour.
This project aims to develop a framework to unify trajectory datasets in time and space, linked to road network data (like OpenStreetMap). Starting with public trajectory datasets (https://www.polymtl.ca/wikitransport/index.php?title=Public_Transportation_Datasets#Traffic_Data) and data collected for previous projects at Polytechnique Montreal, data formats and code will be developed to store, retrieve, process and display trajectory data. An example applications would be to analyze all interactions between right-turning cars and cyclists at intersections.
Research area, student roles & skills
Research area: My research area is intelligent transportation, with a focus on road safety, street functions and traffic data collection. The general goal is to develop computational tools that provide ways to perform more accurate and efficient data collection and road safety analysis. For that purpose, we develop algorithms and software to demonstrate these new methods.
Student roles: As part of a research team developing automated methods for road traffic monitoring and safety analysis, the student will develop the code and test it on real data. The project steps are the following: 1. Review of existing data formats for road user trajectories and methods for their spatio-temporal analysis 2. Selection of datasets and technology (GIS, SQL database, mapping) 3. Development of the code to store, query, process and display trajectory data 4. Testing with several datasets and an application (eg, safety analysis across different cities/countries) and presentation in a conference and public repository (e.g. ArXiv) The student is expected to present the work in several internal and public forums to get feedback and improve the results.
Skills required: The student is expected to have good programming skills, i.e. at least the working knowledge of an object-oriented programming language like C++ or Java. A high-level language, like Python, is desirable. It is also desirable that the student has good knowledge of computer vision and machine learning, as well as good software engineering habits (version control, compilation process using Make or CMake, etc.). Knowledge of Linux and the OpenCV library is a plus. Although the student does not have to be a software engineer, he/she should be dedicated to developing quality software that can be easily maintained. Interest for transport
734. Transforming Online Education with Private Custom LLM-Based Intelligent Learning Systems
Supervisor: Ghazanfar Latif
University: Thompson Rivers University (Kamloops campus)
Students at large universities such as Thompson Rivers University (TRU), often struggle to efficiently access, interpret, and utilize the vast amount of academic, administrative, and support information required for successful learning and well-being. This challenge can negatively affect academic performance, decision-making, and timely access to essential university services, especially in an online learning environment. To address this issue, this project proposes the development of an AI-based educational platform powered by private, custom Large Language Models (LLMs) to enhance the online learning experience.
The research focuses on testing, fine-tuning, and evaluating local open-source LLMs such as Meta’s LLaMA family, Microsoft Phi-4, and DeepSeek-R1 using publicly available TRU data, including academic calendars, program descriptions, course outlines, and student support resources. These models will be integrated within a Retrieval-Augmented Generation (RAG) framework to enable accurate, context-aware responses grounded in institutional knowledge. A web-based prototype application will be developed to support interaction through text and voice interfaces, allowing students to ask questions, explore personalized degree pathways, and access academic and wellness resources.
The system will also include an intelligent advisor capable of answering frequently asked questions, guiding students in course selection, and supporting navigation of mental health and campus services. Performance evaluation will be conducted by comparing locally deployed LLMs with proprietary models such as GPT-4o and Google Gemini across key metrics, including response accuracy, latency, task completion effectiveness, and user satisfaction.
A mixed-methods evaluation approach will be used, combining quantitative performance metrics with qualitative user feedback. Statistical analysis will assess comparative model performance, while thematic analysis will capture student experience and usability insights. Emphasis will be placed on privacy-preserving deployment strategies by ensuring that sensitive data remains within locally hosted environments. This research aims to establish best practices for secure, scalable, and student-centered LLM deployment in higher education, ultimately improving accessibility and learning.
Research area, student roles & skills
Research area: I have published over 110 research articles since 2017 and hold 14 US patents in AI, Assistive Technologies, Medical Imaging, Computer Vision, IoT, Robotics, and Deep Learning. I supervised over 30 undergraduate projects, which resulted in 30+ co-authored publications with those students. I received multiple international awards, including a Gold Medal at the 49th Innovation Exhibition in Geneva and first place at the NEOM Oxagon Hackathon in KSA for assistive tech related to the proposed project. I was named among the Top 2% Scientists by Stanford University in 2024, 2025 and also became an IEEE Senior Member in 2025.
Student roles: Under my supervision, the student will contribute to the research, design, development, and evaluation of an AI-based educational platform powered by private custom Large Language Models (LLMs) to enhance student learning at Thompson Rivers University. The student will assist in collecting, preprocessing, and organizing publicly available university data, including program descriptions, course outlines, academic policies, and student support resources, to build a structured knowledge base for the system. The student will support the development and fine-tuning of open-source LLMs such as LLaMA, Phi-4, and DeepSeek-R1, and their integration within a Retrieval-Augmented Generation (RAG) framework. This includes data preprocessing, prompt engineering, embedding generation, vector database construction, and model evaluation. The student will also contribute to optimizing model performance for tasks such as question answering, academic advising, and personalized degree pathway generation. In addition, the student will assist in developing a web-based prototype application that enables interactive access through text and voice-based interfaces. Responsibilities include integrating backend AI services with frontend components, implementing API connections, and supporting system testing and debugging. The student will also participate in evaluating and comparing local LLM performance against proprietary models such as GPT-4o and Google Gemini using both quantitative metrics (accuracy, latency, task completion rate) and qualitative user feedback. Furthermore, the student will contribute to analyzing user interactions, documenting system behavior, and refining iterative improvements based on evaluation results. The role may also include supporting visualization of model outputs and enhancing system usability for students. This position provides hands-on experience in natural language processing, large language models, retrieval-augmented generation, and full-stack AI application development. Through this work, the student will gain advanced research and engineering skills while contributing to an innovative, privacy-preserving educational AI system aimed at improving student success and accessibility in higher education.
Skills required: I am seeking a motivated intern student with a background in Computer Science, Artificial Intelligence, Software Engineering, Data Science, or related fields. Ideal candidates should have experience in machine learning, natural language processing, and large language models (LLMs). Proficiency in Python and familiarity with frameworks such as PyTorch, TensorFlow, or Hugging Face Transformers are desirable. Experience with Retrieval-Augmented Generation (RAG), web development, and API integration is an asset. Strong analytical, problem-solving, communication, and teamwork skills are essential. An interest in AI for education, human-centered design, and privacy-preserving systems is highly encouraged for this research-focused project.
735. Transforming Real-Time Transit Data into Visual Insights
Public transportation agencies generate large volumes of real-time bus operations data through Automatic Vehicle Location (AVL) systems, which record bus locations, speeds, and arrival times. These data have strong potential to support analysis of transit reliability, equity, and system performance. However, they are often difficult for students, researchers, and transportation planners to access and analyze beyond static reports or high-level dashboards that provide only aggregated summaries. This project builds on an ongoing research effort in Dr. Saidi’s lab that has collected and stored real-time transit data from multiple cities in open formats. While this infrastructure supports fast querying and large-scale analysis, there remains a need for user-friendly tools that allow interactive exploration and comparison of these datasets. The purpose of this project is to develop an interactive, browser-based dashboard for visualizing large-scale bus movement data and comparing transit reliability across Canadian cities. The dashboard will transform complex, high-resolution transit data into clear visual insights that support greener, more sustainable, equitable, and reliable mobility. The project objectives are to design an efficient approach for working with very large transit datasets, create interactive visualizations that reveal reliability patterns such as delays and travel-time variation, and enable comparisons across cities, routes, and time periods. Reliability measures will be calculated from recorded bus movements over time, and the dashboard will be developed using open-source, browser-based visualization tools. The project outcome will be a practical visualization tool that improves access to complex transit data and supports evidence-based research and planning for more reliable and equitable public transportation systems.
Research area, student roles & skills
Research area: My research in general is finding smart short-term and long-term solutions to mobility challenges specially in the area of public transportation considering sustainability, equity, and emerging technologies. My research goal is to create actionable intelligence through developing decision support systems and models to improve the efficiency, quality, and sustainability of societies’ current and future mobility. My main research focus is on public transportation mobility system diagnostics to better monitor, control, and manage services and operations through the lens of sustainability.
Student roles: Help with data collection and data processing from various sources, working data integration and analysis from various sources, develop computer program to test model, help with ridership estimation model development and test the performance using historical observations, delivering presentations and write reports. In addition to the faculty member, the student will receive support from a senior PhD student to complete these tasks.
Skills required: Applicants in computer science is preferred. The project requires strong analytical and technical skills, including experience working with large datasets, basic programming skills, and familiarity with data processing and visualization. Experience with Python, JavaScript, SQL, or browser-based visualization tools would be an asset. Knowledge of GIS, web mapping, or transit data formats such as AVL or GTFS would also be beneficial, but is not required.
The student should be comfortable learning new tools, working independently, and translating complex data into clear and meaningful visual outputs. Strong attention to detail, problem-solving skills, and the ability to communicate results clearly are important.
736. Trustworthy AI for Data Privacy: Protecting Personal Data in the Era of Pretraining
Artificial intelligence and machine learning, especially deep learning, have advanced significantly in recent years, giving rise to powerful pretrained models built on massive amounts of data. However, this progress also raises serious concerns over unauthorized data usage, where collected data may be exploited without consent and thereby compromise the privacy of data owners. To address this issue, this project focuses on the privacy dimension of trustworthy AI and studies Unlearnable Examples (UEs), a promising privacy-preserving strategy that protects data by injecting small perturbations, so that models trained on the protected data fail to generalize. Ideally, the project will result in a paper published at a top-tier ML/CV conference (e.g., ICLR, CVPR). The key strengths and objectives of our project include:
-- Understanding how pretraining priors undermine privacy protection. We focus specifically on protecting personal data against unauthorized exploitation by large-scale pretrained models. We will investigate how the semantic representations acquired during pretraining affect the protection offered by UEs, yielding new theoretical and empirical insight into the interaction between data privacy and pretraining.
-- Designing protection mechanisms robust to pretraining priors. We will develop principled optimization frameworks that disrupt the prior-driven alignment between data and semantics, compelling models to rely on the injected protective signals rather than authentic features. We will evaluate these methods across diverse pretraining and finetuning scenarios, including computer vision models and large language models (LLMs), to assess their real-world applicability.
Research area, student roles & skills
Research area: My research centers on trustworthy machine intelligence, which is essential for deploying intelligent systems with any amount of versatility and reliability. I study the foundational pillars of trustworthiness, including fairness, privacy, and robustness, etc. Specifically, I design novel machine learning methods grounded in solid theoretical foundations across a wide range of scenarios. Over the years, our group has published (100+) papers intensively in top venues (e.g., NeurIPS, ICML, ICLR)
Student roles: Responsibilities and tasks The successful candidate will be responsible for: - Doing a literature review in the related areas - Implementation of novel machine learning algorithms for trustworthy AI - Contributing to efficient interdisciplinary research team collaboration - Writing a scientific report/paper - Completing at least one of the objectives listed in the project description
While the start date is flexible, in previous years, the selected students would work with me remotely once they were matched. Ideally, the project will result in a paper published in top venues (e.g., ICML, NeurIPS, ICLR).
Skills required: You may be a strong candidate if you have (at least three of the following): - a strong background in statistics, linear algebra - strong motivation and interest in machine learning research. - knowledge of machine learning, optimization - published academic work on the topics of machine learning - sufficient programming experience (e.g., Python) - some experiences on advanced machine learning paradigms (e.g., computer vision, adversarial examples, watermarking, etc.) - good English reading and speaking skills
Disabled applicants will receive priority in case they have equal qualifications.
737. Trustworthy Large Language Models for Sustainability Reports Analysis
Sustainability and climate-related disclosures have become increasingly important for investors, regulators, financial institutions, and the public. Large organizations now publish extensive sustainability reports describing environmental, social, governance (ESG), and climate-related risks and opportunities. However, these reports are often hundreds of pages long, highly unstructured, and difficult to analyze efficiently.
Recent advances in Large Language Models (LLMs) provide new opportunities for automating sustainability disclosure analysis. However, current LLM systems still face major challenges, including hallucination, inconsistent reasoning, lack of transparency, and limited reliability when processing long technical documents.
This project aims to develop trustworthy and explainable LLM-based methods for climate and sustainability disclosure analysis. The intern will contribute to building AI pipelines capable of extracting, organizing, and evaluating sustainability-related information from long corporate reports. The project will investigate retrieval-augmented generation (RAG), prompt engineering, structured reasoning, and explainability techniques to improve the reliability and interpretability of AI-generated outputs.
The project may include benchmarking LLM-generated analyses against established sustainability disclosure frameworks such as IFRS S2 and GRI. The developed tools can support more transparent and accessible sustainability intelligence for researchers, businesses, investors, and policymakers.
Research area, student roles & skills
Research area: My expertise is AI and Machine Learning. My current research focuses on multi-omics and sequence data analysis which provides novel information on the mechanisms of the biological process. We develop computational tools for complex and high-dimensional data. I'm particularly interested in applying deep learning and AI to complex data and solve real-life problems.
Student roles: The intern will participate in several aspects of the research project, including: • Collecting and preprocessing sustainability and climate disclosure reports • Reviewing literature on LLMs, explainable AI, and sustainability analytics • Developing and evaluating retrieval-augmented generation (RAG) pipelines • Experimenting with prompt engineering and structured information extraction • Benchmarking different LLMs for disclosure analysis tasks • Supporting annotation and dataset development for sustainability-related questions • Evaluating explainability, consistency, and reliability of AI-generated outputs • Creating visualizations and documentation for experimental results • Assisting with research reports, open-source documentation, and potential publications
Expected outcomes of the project include: • An annotated dataset for sustainability disclosure analysis • Benchmarking results comparing multiple LLM approaches • A prototype explainable AI pipeline for sustainability report analysis • Technical documentation and reproducible research workflows • Potential conference or journal publication contributions
Skills required: The student is expected to have: • Strong programming skills in Python • Experience with machine learning or natural language processing • Familiarity with deep learning frameworks such as PyTorch or TensorFlow • Experience with data preprocessing and visualization • Basic understanding of Large Language Models (LLMs) is preferred • Experience with Git/GitHub is an asset • Good analytical, communication, and technical writing skills Background knowledge in sustainability, finance, or business analytics is beneficial but not required.
738. Understanding AI misinformation
Supervisor: Ifeoma Adaji
University: University of British Columbia (Kelowna campus)
Artificial intelligence (AI) systems are increasingly used to answer questions, provide information, and support decision making across domains such as health, education, climate change, finance, and public policy. While these systems offer significant benefits, they can also generate inaccurate, misleading, or fabricated information. However, little is known about how the occurrence and nature of AI misinformation varies across subject areas and AI models.
The goal of this project is to investigate misinformation generated by contemporary AI systems across multiple knowledge domains. Specifically, the project will compare responses produced by leading AI models when answering questions from areas such as health, climate change, education, technology, finance, and public affairs. The study will examine the frequency of misinformation, the types of errors generated, and differences in performance across models and subject areas.
To support this work, we will develop a benchmark dataset consisting of carefully curated questions spanning diverse domains. Responses generated by multiple AI models will be collected and evaluated to determine their accuracy and reliability. This dataset will enable the identification of patterns in AI misinformation and highlight domains that may be particularly vulnerable to inaccurate AI-generated information.
The project will also explore factors that contribute to misinformation generation and examine whether prompting strategies can reduce the likelihood of incorrect responses. By comparing models and domains, the study will provide insights into the strengths and weaknesses of current AI systems and identify opportunities for improving reliability.
The findings will contribute to research on AI trustworthiness, responsible AI deployment, and AI literacy. The project will generate a publicly available benchmark dataset, empirical evidence regarding AI misinformation across domains and models, and recommendations for helping developers and users better understand and mitigate misinformation generated by AI systems.
Research area, student roles & skills
Research area: My research is in the general area of Human-Computer Interaction. In particular, my research is in the user modelling theme of HCI. My research involves developing applications and games to influence the attitude or behaviour of people for their good and that of society. In addition, my research consists in analysing user-generated data from social media to understand users' behaviour in order to predict future behaviour.
Student roles: The student will assist with the design and execution of the research study. Specifically, the student will conduct literature reviews on AI misinformation and AI trustworthiness, help develop a benchmark dataset of questions across multiple subject areas, and collect responses from various AI models. The student will participate in data cleaning, annotation, and analysis to identify patterns of misinformation across domains and models. In addition, the student will assist in evaluating model outputs, preparing research reports, and disseminating findings through presentations and publications. Throughout the project, the student will receive training in AI evaluation, data analysis, and research methodologies.
Skills required: The ideal student would be an undergraduate Computer Science student with a strong foundation in programming, data analysis, and artificial intelligence. Through coursework and project experience, the student has developed skills in Python programming, data processing, and problem solving. The student has a strong interest in AI and its societal impacts, particularly issues related to information quality, trust, and responsible technology use. Participation in this project will allow the student to further develop skills in AI evaluation, experimental design, data analysis, and research communication while contributing to research on misinformation generated by AI systems.
739. Understanding Female Creator Intentions, Audience Motivations, and Reception
Supervisor: Afsoon Soudi
University: Toronto Metropolitan University
Location: Toronto, Ontario
Start date: 2027-05-25 (flexible)
Disciplines: Computer Science, Sociology, Statistics, Journalism, Physics, Science and Technology, Mathematics, Management Information Systems
Digital platforms are increasingly central to civic and political discourse, with female creators playing a significant role. This project contributes to a broader understanding of their influence in a larger study: systematically analyzing the content these creators produce and the public sentiment it generates. The undergraduate researcher will assist in examining how influential female creators frame civic issues and how audiences respond, laying crucial groundwork for understanding the mechanisms of their impact.
The primary goal is to conduct a foundational analysis of selected female creators' civic and political content and the associated audience reactions. Specific objectives include:
Assisting in the systematic qualitative content analysis of a purposively selected sample of content (videos, posts, articles) from influential female creators across platforms like YouTube, TikTok, Instagram, X, and blogs. This involves helping to identify key themes, framing strategies (linguistic and visual), and calls to action.
Supporting the quantitative sentiment analysis of audience comments and discussions related to the selected content and topic modeling of the content. This includes aiding in the collection of comment data, its preparation for analysis, and the preliminary interpretation of sentiment trends to gauge public reception.
The students will support a mixed-methodological approach:
Qualitative Content Analysis Support: The student will assist in applying a predefined coding scheme to content, identifying illustrative examples of framing techniques, and contributing to discussions on emerging themes. This may involve working with textual, visual, and audiovisual data.
Quantitative Sentiment Analysis and Topic Modeling Support: The student will assist in the ethical collection of public audience comments from relevant platforms using research API. They will help clean and prepare this textual data under supervision, and assist with sentiment analysis of comments and topic modeling of videos and help visualize patterns related to specific content or creators.
Research area, student roles & skills
Research area: Dr. Afsoon Soudi is an Assistant Professor at Toronto Metropolitan University and the Director of Creative AI hub at The Creative School. She led machine intelligence teams to develop an audience analytics platform and recommender systems enabling discoverability and personalization in digital media products. Dr Soudi completed her Ph.D. in Physics at Washington State University. She holds multiple patents and published numerous peer-reviewed papers in high-impact journals. Her current research interests intersect responsible machine learning, sustainability and media.
Student roles: The undergraduate students will act as a Research Assistant (Topic modeling & Sentiment Analysis), providing vital support for the analytical phase of a study on female creators' influence on civic engagement. Working under the guidance of senior researchers, the student's responsibilities will include: Assisting in the identification and curation of a sample of influential female creators and their relevant civic/political content from diverse platforms. Systematically reviewing and applying coding schemes to selected content (text, images, video transcripts) as part of the qualitative content analysis. Ethically collecting publicly available audience comments and discussion data related to the analyzed content. Cleaning and preparing textual comment data for quantitative sentiment analysis. Preparing video for Topic modeling through transcription tools. Conduct basic sentiment analysis under supervision to process comment data. Assisting in organizing, visualizing, and interpreting preliminary findings from both the content analysis (e.g., common frames, themes) and sentiment analysis (e.g., dominant sentiment scores, emotional tenor). Maintaining meticulous records of analyzed content, data sources, and analytical steps. Participating in regular team meetings, contributing to methodological discussions, and potentially presenting initial findings to the research group. Potential Deliverables will include: A documented and partially coded sample of female creators' civic/political content. A cleaned and organized dataset of audience comments corresponding to the analyzed content. A preliminary report summarizing identified content themes, framing strategies, and initial sentiment patterns, co-developed with the research team.
Skills required: We are looking for a motivated students who possesses: Excellent research and critical thinking skills with Knowledge of data extraction using research APIs, data modeling, and managing github repository Knowledge of content analysis, sentiment analysis and topic modeling is a plus Effective communication skills: The ability to communicate the research findings clearly and effectively both written and orally. Time Management: Proper time management skills to meet deadlines and complete the study within the specified timeline. Attention to detail and ability to follow guidelines and work under supervision Strong interest in media and audience theories, public behaviour and perception
Large language models often produce step-by-step "reasoning" before giving an answer, and users increasingly trust these explanations. But a model's stated reasoning may not reflect what actually drove its answer. It can rationalize after the fact, hide the cues it truly relied on, or be steered by signals it never mentions. This gap, known as reasoning unfaithfulness, is a serious problem for safety, trust, and oversight of modern AI.
This project investigates when and why LLM reasoning becomes unfaithful. The student will design controlled experiments that probe whether a model's chain-of-thought genuinely explains its behavior, measure faithfulness across tasks and model families, and explore how factors like post-training and adversarial manipulation affect it. The work combines hands-on experimentation with open-source models, careful evaluation design, and analysis of results.
The intern will gain experience in trustworthy machine learning, LLM evaluation, and research methodology and contribute to findings suitable for a workshop or conference submission.
Research area, student roles & skills
Research area: My research centers on trustworthy machine learning, with a focus on understanding and defending against threats that arise when models learn from data in the wild. I study data poisoning attacks and defenses, machine unlearning, and memorization in generative models, asking how we can protect both training data and the models built on it. Recent work spans poisoning attacks across LLM post-training stages, detection and attribution for problematic data in agent memory, and proofs of training data membership. My goal is to build the technical foundations and practical tools that let people trust how their data shapes modern AI.
Student roles: The student will work as a research intern within the lab, taking an active role in the project under the supervision of the professor and in collaboration with graduate students. They will begin by reviewing relevant literature on LLM reasoning, chain-of-thought, and faithfulness to build a working understanding of the problem, then set up and run open-source language models while writing clean, reproducible code for experiments. The core of their work will involve designing and implementing controlled experiments that test whether a model's stated reasoning reflects its actual behavior, and measuring and analyzing faithfulness across different tasks, prompts, and model families. As the project develops, the student will investigate how factors such as post-training or adversarial manipulation affect reasoning faithfulness, documenting their results throughout. They will meet regularly with the supervisor and lab members to discuss progress, troubleshoot, and refine the research direction, and will contribute to a written report and, where possible, to a workshop or conference submission. Overall, the intern is expected to take ownership of their part of the project, work independently on day-to-day tasks while seeking guidance when needed, and engage actively in the lab's research community.
Skills required: - Solid programming skills in Python, including experience with a deep learning framework such as PyTorch - Foundational knowledge of machine learning and neural networks (coursework or projects) - Familiarity with how large language models work at a high level, including concepts like tokens, training, and inference - Comfortable working in a Linux or command-line environment and using Git - Strong analytical and problem-solving skills, with the ability to design and interpret experiments carefully - Good written and spoken English for documenting results and discussing research
741. Understanding and Extracting Critical Traces in Agentic AI Systems
Supervisor: Mahtab Sarvmaili
University: University of New Brunswick (Fredericton campus)
Large language model based agents are increasingly used to complete multi-step tasks by planning, calling tools, retrieving information, using memory, and interacting with external environments. However, their execution traces are often long, noisy, and difficult for humans to inspect. A final answer does not reveal which evidence was used, whether tool calls were appropriate, where an error originated, or which steps were most important for the outcome.
This project will study how to understand and extract the critical parts of an agent’s execution trace. The goal is to move from raw logs to compact, structured trace summaries that help users, developers, and auditors understand agent behaviour. The project will focus on questions such as: What should be recorded in an agent trace? Which trace events are most informative? How can we identify decisive tool calls, missing evidence, unsupported assumptions, plan changes, memory use, retries, or failure points?
During the 12-week internship, the student will build or adapt a small agentic workflow, such as a retrieval-augmented research assistant, document question-answering agent, or multi-agent reviewer. The workflow will be instrumented to collect structured traces, including goals, plans, tool inputs and outputs, retrieved evidence, intermediate decisions, and final responses. The student will then design lightweight methods to filter, rank, and summarize important trace segments. Possible approaches include rule-based extraction, evidence-linking between final claims and sources, comparison of successful and failed runs, and simple failure-attribution heuristics.
The expected outcomes include a prototype agentic workflow, a structured trace schema, a critical-trace extraction method, example case studies, and an evaluation comparing raw logs with extracted trace summaries. The project contributes to responsible AI by improving transparency, debugging, accountability, and trust calibration for increasingly autonomous AI agents.
Research area, student roles & skills
Research area: My specialized research area is Responsible AI, with a focus on explainability, interpretability, reliability, privacy, and control of modern machine learning and large language model systems. My work investigates how complex AI models make decisions, how their predictions can be explained in data-centric and model-centric ways, and how their behavior can be evaluated and improved for trustworthy deployment. I am particularly interested in interpretable and aligned LLM-augmented systems, including retrieval-augmented generation and agentic AI, where transparency, safety, and robustness are essential. My broader goal is to develop practical methods that make AI systems understandable, reliable, and accountable in real-world applications.
Student roles: The student will play an active role in all stages of the project, with guidance from the supervisor. In the first stage, the student will review recent literature on agentic AI, trace-based evaluation, explainability, execution provenance, and failure attribution. Based on this review, the student will help define a small but meaningful agentic workflow that can be completed within the internship period.
In the second stage, the student will implement or adapt the selected workflow. This may involve building a simple retrieval-augmented assistant, document question-answering agent, coding helper, or multi-agent reviewer. The student will instrument the system to record structured traces, including the task goal, plans, tool calls, retrieved evidence, intermediate decisions, memory use, errors, retries, and final answer. The student will also create a small set of controlled tasks that produce both successful and failed runs.
In the third stage, the student will design and test methods for extracting critical trace elements. The focus will be on identifying the events that matter most for understanding the agent’s behaviour, such as decisive tool calls, unsupported claims, missing evidence, incorrect assumptions, failed actions, or changes in plan. The student may compare rule-based, evidence-based, and failure-oriented extraction strategies.
In the final stage, the student will run experiments, analyze successful and failed agent runs, and evaluate whether extracted trace summaries are more useful than full raw logs. The student will prepare documentation, example case studies, and a final report summarizing the prototype, methodology, findings, and limitations. The project is designed to be feasible for a strong undergraduate student and does not require training large models. The student’s main contribution will be system implementation, trace design, experimental analysis, and responsible-AI evaluation, while regular meetings with the supervisor will help keep the scope realistic and research-focused.
Skills required: The ideal student should have strong Python programming skills and basic knowledge of machine learning, natural language processing, or software engineering. Experience with large language models, retrieval-augmented generation, APIs, or agent frameworks would be helpful, but is not required. The student should be comfortable reading research papers, implementing small prototypes, running experiments, and analyzing results. Familiarity with data structures, logging, visualization, or evaluation metrics would be an asset. Most importantly, the student should be curious about responsible AI, explainability, debugging, and how complex AI systems can be made more understandable, reliable, and accountable.
742. Understanding and Reducing Cognitive Load in AI-Assisted Development Environments
Supervisor: Sanaa Alwidian
University: Ontario Tech University (Oshawa campus)
The rapid adoption of AI-powered coding assistants such as GitHub Copilot and similar large-language-model-based tools is fundamentally reshaping how software developers work. While these tools promise productivity gains, there is growing concern that they may inadvertently increase cognitive load, the mental effort required to process, evaluate, and integrate AI-generated suggestions alongside one's own reasoning about code correctness, security, and intent. Despite the widespread deployment of these tools in industry, there has been little rigorous empirical investigation into how they affect developers' cognitive processes during real programming tasks.
This project addresses that gap through a controlled empirical study combining physiological measurement, behavioral observation, and self-report methods to characterize how AI coding assistants affect developer cognitive load. Using eye-tracking technology, the NASA Task Load Index workload questionnaire, and structured think-aloud sessions, the study will expose participants to three conditions: programming without AI assistance, programming with standard inline AI suggestions, and programming with a modified interface featuring confidence indicators and suggestion deferral controls. Participants will include developers across experience levels, from novice to expert, recruited from the university community.
The study will identify which types of AI-generated suggestions, such as boilerplate code, algorithmic logic, or security-critical components, are associated with the highest cognitive load, and evaluate whether specific interface design choices can mitigate these effects. Findings will be synthesized into a set of evidence-based design heuristics for next-generation AI coding assistant interfaces.
Beyond its immediate contributions, this research has direct implications for the responsible design of AI developer tools being adopted at scale across Canada's technology sector, including in regulated domains such as healthcare and financial software where AI-assisted coding errors carry significant consequences. Results will be prepared for submission to a leading software engineering or HCI venue such as CHI or ICSE.
Research area, student roles & skills
Research area: I am the director of the ARiSE Research Lab (https://arise-lab.ca/). My research focuses on human-centric requirements and software engineering, with an emphasis on developing methodologies and frameworks that place human needs, values, and experiences at the center of the software development process. I investigate how requirements are elicited, represented, and validated in complex socio-technical contexts, and how software systems can be designed to better serve diverse user populations — including those with accessibility needs. My work spans empirical software engineering, participatory design, and AI-enabled requirements engineering, contributing to the development of more reliable, inclusive, and user-responsive software systems.
Student roles: The intern will serve as a junior researcher within the ARiSE Lab at Ontario Tech University, taking an active and ownership-driven role across all phases of this empirical study. The internship is structured to provide progressive responsibility, with the intern transitioning from structured onboarding to independent execution of research tasks as the project advances. In the opening weeks, the intern will conduct a focused synthesis of existing literature on cognitive load theory, developer experience, and AI coding assistant usability. They will assist in finalizing the experimental task battery, configuring the eye-tracking hardware and software environment, and preparing the study protocol for any required ethics amendments. During the pilot phase, the intern will recruit and run sessions with a small group of participants to refine the protocol, validate the behavioral coding scheme, and calibrate the eye-tracking setup. This phase is critical for ensuring data quality and procedural reliability before the main study begins. In the main data collection phase, the intern will independently recruit and run sessions with 20–30 participants, administer the three experimental conditions, collect eye-tracking and task performance data, and conduct think-aloud sessions. They will perform daily data quality checks and run behavioral coding in parallel with ongoing data collection. Following data collection, the intern will lead the quantitative analysis using repeated-measures statistical methods in R or Python, and conduct thematic analysis of think-aloud recordings using qualitative coding software. They will then synthesize findings into a set of evidence-based design heuristics for AI coding assistant interfaces. In the final weeks, the intern will co-author a draft research manuscript, prepare and deliver a formal oral presentation to the lab group, and submit a comprehensive project report documenting all methods, findings, and recommendations.
Skills required: Applicants should be in their 3rd or 4th year of an undergraduate degree in Computer Science, Software Engineering, or Computer Engineering. Required skills include proficiency in Python or JavaScript, familiarity with modern software development environments and IDE-based tools, and a genuine interest in empirical research and human-computer interaction. Experience with experimental design or human subjects research is an asset, as is prior exposure to eye-tracking or other biometric measurement tools. Strong analytical skills are essential for processing both quantitative and qualitative data. Coursework or demonstrated interest in HCI, developer experience, or AI-assisted software development is considered a strong advantage.
743. User Interface Development for a Multiaxial Neuromodulation Array
Supervisor: Laura Curiel
University: University of Calgary
Location: Calgary, Alberta
Start date: 2027-05-03 (flexible)
Disciplines: Computer Science, Engg-Computer, Engg-Biomedical, Engg-Software, Engg-Systems and Technology
This project focuses on the development, integration, and validation of the user interface (UI) for a multiaxial array for transcranial ultrasound neuromodulation, with particular emphasis on usability, safety, and regulatory compliance. The student will work closely with ultrasound pressure model developers to support integrating targeting and model predictions into a UI for Neuronavigation, ensuring seamless interaction between software controls, user interaction and the underlying hardware system. The project includes conducting UI validation activities to assess functionality, usability, and alignment with user and system requirements, followed by iterative redesign based on validation findings. In addition, the student will prepare validation results and supporting documentation suitable for regulatory review. The main goal is to deliver a robust and validated UI that supports accurate, intuitive, and compliant operation of the neuromodulation device. Through this project, the student will develop skills in human–machine interface design, software-hardware integration, usability validation, design iteration, and regulatory-oriented documentation within the context of advanced neurotechnology development.
Research area, student roles & skills
Research area: This research area focuses on user-centred interface (UI) development and validation for advanced transcranial ultrasound neuromodulation systems. It integrates human–machine interface design, software–hardware integration, and neuronavigation workflows to support safe, accurate, and intuitive device operation. Emphasis is placed on usability engineering, system validation, and regulatory-aligned documentation to ensure compliance with medical device standards. The work bridges neurotechnology, software engineering, and clinical usability considerations, addressing how complex targeting algorithms and pressure models can be effectively translated into user-facing controls. This research supports the translation of neuromodulation technologies into reliable and clinically deployable systems.
Student roles: The student will play a key role in designing, integrating, and validating the user interface for a multiaxial transcranial ultrasound neuromodulation system. Drawing on a background in mechatronics, software, or biomedical engineering, the student will collaborate closely with ultrasound model developers and system engineers to integrate targeting, control, and neuronavigation functionality into an intuitive and safe UI. They will plan and conduct usability and validation testing, evaluate alignment with user and system requirements, and translate user feedback into iterative design improvements. The student will also document design decisions, validation results, and supporting evidence in a manner suitable for regulatory review, contributing to the development of a robust, user-centred medical device interface.
Skills required: Suggested student background: • Background in mechatronics, software or biomedical engineering • Background or interest in software development, UI/UX, or human-computer interaction • Familiarity with validation testing or usability evaluation concepts • Ability to collaborate closely with engineers and domain experts • Experience documenting design decisions and test results • Attention to usability, safety, and workflow design • Strong communication skills for translating user feedback into design improvements • Interest in neurotechnology or medical device interfaces
744. User Interfaces for Software Tool Sharing and Discovery among Natural Scientists
Prior work has shown that natural scientists (e.g., marine scientists, geologists) often create their own scripts or tooling to do data analysis, which can make it harder to verify correctness of scripts and tooling, and can result in duplication of efforts across labs. In this project, we will explore user interface designs that enable natural scientists to more easily find and share scripts and tools for data analysis in ways that can be easily integrated into existing scientific software workflows. This will involve interviewing natural science researchers to better understand their needs, and prototyping software user interfaces to facilitate script sharing and integration into scientific workflows.
Research area, student roles & skills
Research area: Within the broader discipline of human-computer interaction (HCI), my research investigates, designs, and evaluates non-game forms of playfulness in software and electronic user interfaces. This includes interactions such as personalizing digital devices to facilitate self-expression, using software to play around with data analyses or visualizations, and engaging with joyful features like Easter eggs (hidden features or media).
Student roles: The student will conduct formative interviews with approximately 10 natural science researchers to identify: (1) potential moments during user interface interactions that could permit the sharing and receiving of scripts and tools, and (2) what natural science researchers would like to see in the design of a user interface for searching for others' shared scripts and tools.
The student will also assist with designing and developing user interface prototypes that help satisfy scientists' needs identified from the interviews. Interfaces to be designed may include one or more of the following: (1) Plugins for graphical software such as ArcGIS or code-based software such as RStudio that proactively suggests scripts or tools to incorporate, or facilitates searching of existing scripts or tools. (2) A web interface or desktop application representing a database of tools and scripts that facilitates searching, reviewing, and producing variants of such scripts or tools.
Skills required: Required: Experience using graphical or code-based scientific visualization or analysis software (e.g., Jupyter Notebook, RStudio, GraphPad Prism, ArcGIS, among many other examples).
Ideally: - The student will have a multidisciplinary background in both computer science and a natural sciences area (e.g., marine science, atmospheric science, geology). - The student will have experience creating custom scripts or tooling (e.g., simple interfaces) to facilitate scientific data analysis and/or visualization.
Beneficial, but not required: Prior experience conducting user studies with human participants (e.g., interview studies).
745. Using Electrical Muscle Stimulation to Reduce Performance Decline During Virtual Reality Selection Tasks
This project investigates whether Electrical Muscle Stimulation (EMS) can help reduce performance decline during repeated target-selection tasks in virtual reality. The study will use a VR-based Fitts’ Law task, where participants repeatedly select targets of varying sizes and distances. Such tasks are commonly used in Human-Computer Interaction to evaluate motor performance, pointing accuracy, movement time, and interaction efficiency.
During prolonged VR interaction, users may experience fatigue, reduced precision, slower responses, or decreased motor control. This project will explore whether carefully timed EMS feedback or assistance can support the user’s movement and help maintain performance over time. The objective is not to replace voluntary movement, but to examine whether subtle neuromuscular stimulation can reduce performance attenuation and improve interaction stability.
The project will involve designing a VR experimental environment, implementing a Fitts’ Law selection task, integrating EMS stimulation conditions, and comparing user performance with and without EMS. Performance metrics may include movement time, error rate, throughput, accuracy, fatigue perception, and subjective workload. The results could contribute to the design of future assistive, adaptive, and embodied interaction systems in VR, particularly for applications involving prolonged interaction, training, rehabilitation, or precision-based tasks.
Research area, student roles & skills
Research area: Prof. Pascal E. Fortin has been affiliated with the Computer Science and Mathematics Department of Université du Québec à Chicoutimi since 2022. He holds a PhD from McGill University. His main research expertise lies in human-computer interactions, with a specific focus on multisensory and physiological interaction techniques. Prof. Fortin’s current projects tackle the mechanisms of cybersickness in extended reality (XR) environments to design more comfortable, immersive systems. His work not only pushes the boundaries of interactive technology but also fosters interdisciplinary collaboration through strategic partnerships with local industries and governmental entities, driving innovation and producing tangible benefits.
Student roles: The candidate will focus on identifying when, where, and how EMS should be applied during a VR-based Fitts’ Law selection task. Their responsibilities will include reviewing the literature on EMS, motor assistance, fatigue, and VR interaction; defining relevant movement phases where EMS could be applied; helping determine suitable stimulation locations on the arm or hand; designing experimental EMS conditions; assisting with the calibration and safe use of EMS; supporting data collection; analyzing how EMS timing and placement affect performance, accuracy, fatigue, and workload; and contributing to the preparation of a research report or manuscript draft.
Skills required: The candidate should have basic knowledge of Human-Computer Interaction, virtual reality, experimental design, and quantitative data analysis. Experience with Unity, C#, Python, statistics, or biosignal/haptic technologies would be an asset. Interest in motor performance, wearable technologies, human factors, and user studies is strongly encouraged.
746. Using ML/AI Methods to Investigate Brain States
The data stream of electrical activity collected from the brain (EEG dataset) is incredibly complex time series and each individual's data consists of anywhere from 27 to 256 time series representing activity across the brain. We are collecting data across a variety of conditions and tasks. Two active areas are: people who have suffered concussions and people placed under situations that require spontaneous creative activity. The intern will work closely with the senior research scientists in the team to do one of the following: (1) develop machine learning tools to cluster datasets and analyze the clustering to see if they coincide with behavioural and performance measures; (2) analyze new and existing datasets and AI/ML outcomes for secondary dependencies (e.g. age, sex, etc.); (3) develop graphical neural networks to leverage information flow measures to construct maps of brain networks; (4) investigate whether novel measures (e.g. criticality measures) can be used to understand changes in brain state; (5) develop insightful visualisation tools that can be used to gain further insight into how the brain functions. The specific project available to the research intern will depend on the intern's background, experience and skill sets as well as our priorities at the time.
Research area, student roles & skills
Research area: The human brain is made up of more than a hundred billion neurons connected to each other via a hierarchy of networks. How do these networks move information around? How is this information is processed? How does the resulting activity leads to perception, creativity, emotions, etc. are all open questions. Similarly, how injuries like concussions, PTSDs and dementia disrupt the normal functioning of the brain is also an open question. I work with a team led a neuroscientist, who uses EEG to record brain activity, to investigate these various issues using novel techniques like information flow and AI/ML tools.
Student roles: As noted above, the student intern will work closely with members of the research team to develop and/or refine existing Python codes and scripts to analyze the the EEG datasets. Depending on the project specifics, this may involve developing and refining existing innovative data analysis scripts that we have developed; exploring the use of CNNs/GNNs/RNNs etc. to analyze/cluster the data; or writing new code -- creating a new analysis package -- based on a novel approach that we want to try out. The project will inevitably involve making (or creating scripts that can make) high-quality plots, graphics, and even movies. The intern will be expected to provide at least once-a-week updates to the research group. This is an excellent opportunity to learn and hone presentation skills. And, if the project advances sufficiently, the intern will participate in drafting a manuscript for publication.
Skills required: This project is at the intersection of neuroscience and data/computer science. A suitable intern is expected to be a motivated, independent researcher, a go-getter who is willing to learn new techniques, investigate new directions, think out-of-the-box, and be persistent. The intern will work closely with members of the research team on the project. Regardless of project specifics, the intern will require and must have excellent Python programming (and debugging) skills, experience with AI/ML methods (CNNs, GNNs, etc.), as well as good writing skills.
747. Using generative AI to streamline content generation for secondary math teachers
Supervisor: Nikki Lobczowski
University: McGill University (Montréal campus)
Location: Montreal, Québec
Start date: 2027-05-31 (flexible)
Disciplines: Computer Science, Design, Industrial Design and Technology
This research project sits at the intersection of artificial intelligence and secondary mathematics education, focusing on advancing a locally hosted large language model (LLM) designed to support math teaching. This internship will extend an existing project by incorporating design improvements and structured input from target users (secondary math educators). Responsibilities include supporting updates to both the front-end interface and the back-end system, assisting with user engagement activities, organizing and synthesizing feedback, and contributing to benchmarking efforts to evaluate the quality of model-generated math explanations. The role combines research and light technical tasks, including testing model outputs and supporting iterative improvements to the web-based tool. This position offers experience working on applied AI in an educational context, including hands-on exposure to LLM integration within interactive applications, in a collaborative research environment.
Research area, student roles & skills
Research area: My research focuses on designing and studying context-sensitive interventions for collaborative settings that incorporate cutting-edge technologies and analytic methods using a social regulation of learning framework. This project will draw on my previous research working with local teachers to understand how to use genAI for creating secondary math content. My related experience includes a PhD in Learning Sciences, two postdocs in human-computer interaction projects for secondary math, and ten years teaching high school math.
Student roles: The intern will support the ongoing development and evaluation of a locally hosted LLM designed for secondary mathematics education. Responsibilities will include: 1) Supporting system refinement (assist with implementing and documenting design changes to improve model outputs, help test and validate improvements in clarity, correctness, and instructional usefulness) 2) Web application support (assist with basic updates to the application [e.g., interface tweaks, feature testing, workflow improvements], support integration between the LLM back-end and user-facing components) 3) Benchmarking and evaluation (review and compare model-generated outputs using established evaluation criteria, identify patterns in performance, including strengths, errors, and gaps in reasoning, assist in recording and summarizing evaluation results) 4) User engagement and feedback (help coordinate sessions with secondary math educators, organize and synthesize qualitative feedback related to usability and pedagogical value, contribute to identifying actionable insights from user input) 5) Design-Based Research support (assist in documenting experimental iterations and changes over time, collaborate with the research team on ongoing tasks, assist with maintaining clear documentation and project organization) 6) General research support (support basic quantitative and qualitative analysis of results, contribute to summaries and reports used for research outputs)
Skills required: A background in education, mathematics, computer science, or a related field. Interest in AI and learning technologies. Experience with Python, web development (front-end or back-end), and/or working with AI tools. Designing/developing web-based applications Experience in human-centered design (e.g., UI/UX methods, human-computer interaction, co-design, etc.) Working on diverse, interdisciplinary teams (experience preferred, interest required) Experience using qualitative and quantitative methods is preferred Strong organizational skills, attention to detail, and the ability to work independently on structured tasks
748. Using machine learning methods to represent potential energy surfaces in sum of products form: Zundel cation
his research project applies methods in machine learning to molecular physics and addresses the motion of atoms in molecules and during reactions. To understand, at a detailed level, the motion of atoms, one must apply the laws of quantum mechanics and solve the Schrödinger equation by representing wavefunctions in terms of basis functions and using methods of linear algebra to compute observables. To do this one must make a potential energy surface (PES). The student who works on this project will use methods of machine learning to make PESs. In particular, we shall focus on making potential energy surfaces in sum-of-products (SOP) form. The SOP form makes it much easier to solve the Schrödinger equation to compute dynamical properties. To achieve the SOP form, we shall use exponential neurons, a neural network, and a decomposition of the PES into lower-order terms. A combination of ideas is already known to work well for molecules with five atoms. During the summer these ideas will be re-worked to devise a systematically improvable neural-network method for fitting SOP PESs of molecules with about a dozen atoms. It has never been applied to something as large as the Zundel cation (a 15-dimensional PES). There is a lot of interest in the Zundel cation because of the importance of liquid water and water solutions.
Research area, student roles & skills
Research area: I develop numerical and computational methods to study the motion of atoms in molecules and during
reactions.
Student roles: The student will spend most of her/his time reading scientific papers, talking to me and others in the group in order to understand them, and writing and debugging computer programs. I shall communicate regularly with the student to be sure that he or she is making good progress. I shall spend time with the student teaching her or him the basic theory and numerical analysis required to complete the project. I shall also try to enrol the student in CAC (the Queen’s Centre for Advanced Computing) courses on scientific computing. At the end of the summer the student will submit a report summarizing the theoretical/computation methods he/she has used and the results obtained. The student will get a foretaste of research. She or he will learn to use computers to apply quantum mechanics to a real problem. The student who works with me will learn essential elements of computational science and numerical analysis. Computers are used to guide the discovery of new drugs, design airplanes, study global warming, predict the weather etc. Computers are becoming so inexpensive that many companies are using them to model processes related to their products. The student who works with me will learn a great deal about computational chemistry but also about numerical analysis and computers. The training she/he receives will be an excellent preparation for careers both inside and outside academia. Inevitably there will be both thrilling and frustrating moments and the student will learn to enjoy the thrill and persevere through the frustration. The opportunity to learn to use cutting edge computers and to write new computer programs to solve his/her own scientific problem will give the student a firm basis from which to continue his development as a scientist.
Skills required: The student who works with me will spend most of his/her time writing and debugging computer programs, reading scientific articles and books, and formulating equations. The skills in this list will help a student working with me. Although all are desirable none of these skills are essential prerequisites. Lack of any of these skills can be compensated for by desire to learn and native ability. List: Desire to work hard to acquire new knowledge; knowledge of linear algebra; experience with numerical methods; experience with machine learning methods
749. Using machine learning to identify noise aspects in log files
Supervisor: Fabio Petrillo
University: École de Technologie Supérieure (Montréal campus)
Software systems in production generate data, services, and facilities for their users. But as it runs, it also needs to create other information about the system execution. The log files hold this information, which will be helpful to DevOps engineers, monitoring and observability systems, ensuring the good health of the system, preventing attacks, and even as a basis for investigating any system failures. What are the scenarios, settings, and data capable of generating noise in log files? What data characterize the information desired in the log files, and what information is unwanted?
From a knowledge base about logs, one way to reach this objective is to apply Machine Learning (ML) techniques to analyze the lines of the log files to find the noises of these contained in them. In this project, we will research existing log libraries, interview system operators to understand the scenarios in which logs are used, and what information is desired and unwanted from these scenarios.
Thus, we will build a Machine Learning (ML) model to evaluate the log files and their noise level from the collected responses. The project product will support students in the context of UQAC's bachelor with a major in software with knowledge in monitoring, logging, and machine learning.
Research area, student roles & skills
Research area: Fabio Petrillo is an associate professor at the University of Quebec at Chicoutimi. He obtained his Ph.D. in Computer Science in 2016 from Federal University of Rio Grande do Sul (Brazil) and was a postdoctoral fellow at Concordia University. Dr. Petrillo has worked on Empirical Software Engineering, Software Quality, and Architecture, Debugging, Service-Oriented Architecture, RESTful analysis on Cloud, Agile methods. He has been recognized as a pioneer on Computer Games and Software Engineering.e was the creator of Swarm Debugging, a new collaborative approach to support debugging, and he has developed new approaches to logging analysis using Machine Learning techniques.
Student roles: The student role is to project, analyze and implement a prototype, extract and process data from logs papers using text mining and Machine Learning approaches.
Skills required: The student’s required skills are Python and JavaScript programming languages as well as web technologies such as Node.js and concepts on databases such as SQL. Solving problem capability, communication curiosity and proactivity are desired for this project. The student has to be interested in text manipulation and Artificial Intelligence (AI) approaches like Machine Learning.
750. Using markerless motion capture to drive music generation and develop neuroscientific/psychological theories of music creativity and music-movement-dance interactions
Supervisor: suresh krishna
University: McGill University (Montréal campus)
Location: Montreal, Québec
Start date: 2027-05-03 (flexible)
Disciplines: Computer Science, Electronic Systems, Engg-Computer, Engg-Biomedical, Engg-Electrical, Engg-Software, Engineering, Engg-Systems and Technology, Music
This project aims to use the recent AI-based advances in markerless motion capture software (e.g. MediaPipe from Google) to create a working framework whereby movements can be translated into sound with short-latency, allowing for example, gesture-driven new musical instruments and dancers who control their own music while dancing. The project will require familiarity with Python, and ability to interface with external packages like MediaPipe. Familiarity with low-latency sound generation, image processing, and audiovisual displays is an advantage, though not necessary. The development of such tools will facilitate both artistic creation, as well as scientific exploration of multiple areas, including for example - how people engage interactively with vision, sound, and movement and combine their respective latent creative spaces. Such a tool will also have therapeutic/rehabilitative applications in populations of people with limited ability to generate music and in whom agency and creativity in producing music have been shown to produce beneficial effects.
The project is being actively developed this year through the Google Summer of Code and was also part of a MITACS project last year (2025) and Google Summer of Code projects in 2025 and 2024.
Research area, student roles & skills
Research area: Gesture recognition, music generation, neuroscience, agency, sensorimotor integration, eye-tracking
Student roles: The student will take the lead in extending the project (from the state in which it is in, in Summer 2027), and will work in close collaboration with the mentor.
Skills required: Comfortable with Python and modern AI tools. Experience with gesture to speech models, and/or with music, and/or with AI-related music generation, and/or with image/video processing and using deep-learning based image-processing models. Familiarity with C/C++ programming, low-latency sound generation, image processing, and audiovisual displays, as well as MediaPipe are all advantages, though not necessary.
751. Utiliser l’intelligence artificielle pour rendre les prévisions de débit des rivières plus fiables et réalistes
L’intelligence artificielle est de plus en plus utilisée pour prédire les débits des rivières à partir de données météorologiques et hydrologiques. Cependant, certains modèles d’IA peuvent produire des prédictions statistiquement précises, mais peu réalistes du point de vue physique. Par exemple, un modèle peut apprendre des relations qui ne sont pas cohérentes avec les processus hydrologiques réels. Cela limite la confiance envers les outils d’IA appliqués à la gestion des ressources en eau. Ce projet vise à explorer des façons de rendre les modèles d’intelligence artificielle plus fiables et plus cohérents avec les mécanismes gouvernant le cheminement de l'eau. L’étudiant ou l’étudiante travaillera avec des bases de données hydrologiques en accès libre contenant des informations sur les précipitations, la température, la neige et les débits des rivières pour plusieurs bassins versants nord-américains. L’objectif sera de tester différentes approches permettant de guider les modèles d’apprentissage automatique afin que leurs prédictions respectent davantage les relations physiques connues. Le projet comprendra la préparation des données, l’implantation de modèles en Python, les tests de performance et l’interprétation des résultats. L’étudiant ou l’étudiante comparera des modèles d’IA standards avec des modèles intégrant certaines contraintes physiques ou causales. Les performances seront évaluées à l’aide d’indicateurs hydrologiques classiques et d’outils permettant de mieux comprendre les prédictions des modèles. Ce stage convient particulièrement à une personne intéressée par l’intelligence artificielle, les données hydrométéorologiques, les changements climatiques et la gestion des ressources en eau. À la fin du projet, l’étudiant ou l’étudiante aura acquis une expérience pratique avec des données réelles et des méthodes modernes d’apprentissage automatique appliquées à l’hydrologie.
Research area, student roles & skills
Research area: Mes recherches portent sur le développement d'une nouvelle génération de modèles d'intelligence artificielle visant à mieux représenter les systèmes hydrologiques et à faciliter la gestion des ressources en eau. Plus précisément, je développe des modèles hybrides qui combine les connaissances physiques des systèmes hydrologiques avec des méthodes d'apprentissage automatique.
Student roles: L'étudiant contribuera au développement et à l'évaluation de modèles d'IA destinés à la prévision des débits fluviaux. Au début du stage, il se familiarisera avec les concepts clés liés à la modélisation hydrologique, à l'apprentissage automatique et à l'intelligence artificielle fondée sur la physique. Il participera ensuite à la préparation d'ensembles de données hydrologiques, notamment des séries chronologiques de données météorologiques et de débits fluviaux.
La tâche principale consistera à mettre en œuvre et à tester des modèles d'apprentissage automatique en Python. L'étudiant comparera des modèles d'IA standard à des modèles conçus pour mieux respecter les relations physiques en hydrologie. Cela peut inclure des contraintes simples ou des ajustements de modèles qui contribuent à garantir que les prévisions restent réalistes d'un point de vue environnemental.
L'étudiant analysera les performances des modèles à l'aide de mesures standard et produira des résumés visuels des résultats. Il pourra également utiliser des outils d'interprétabilité pour mieux comprendre quelles variables influencent les prévisions des modèles. Tout au long du stage, l'étudiant participera à des réunions régulières avec le superviseur et l'équipe de recherche, discutera des résultats et améliorera progressivement le processus de modélisation.
À la fin du stage, l'étudiant devra produire du code documenté, des figures et un bref rapport de recherche résumant la méthodologie, les résultats et les principales conclusions. En fonction des résultats, ce travail pourra contribuer à une publication scientifique ou à un futur projet de recherche de troisième cycle.
Skills required: L’étudiant ou l’étudiante devrait posséder des bases en programmation Python et démontrer un intérêt pour l’apprentissage automatique, la science des données ou la modélisation hydrologique. Une expérience avec des bibliothèques comme NumPy, pandas, scikit-learn, PyTorch ou TensorFlow constitue un atout, sans être obligatoire. La personne devra être à l’aise avec les données numériques et intéressée à apprendre les concepts de base en hydrologie. La curiosité, l’autonomie et les habiletés de communication sont importantes. Une expérience en analyse de séries temporelles ou en réseaux neuronaux serait un avantage.
752. VERAM: Verification-Oriented Evaluation of Retrieval-Augmented Methods for Trustworthy Scientific Code Generation
This project will develop and evaluate VERAM, a verification-oriented framework for assessing retrieval-augmented methods in scientific code generation. While recent AI coding systems can generate programs that compile and execute, this alone is not sufficient for scientific computing. In numerical and high-performance applications, small errors, race conditions, unstable outputs, or nondeterministic behavior can lead to unreliable scientific conclusions. The project investigates three questions: when AI-generated scientific code is correct, what failure modes occur despite successful compilation, and which verification checks most effectively identify unreliable outputs.
VERAM shifts the emphasis from generation to verification. Instead of asking only whether AI can produce faster parallel code, this project asks whether the generated code preserves correctness, numerical stability, runtime safety, and reproducibility.
The project will use representative scientific computing workloads drawn from sparse numerical computing and graph analytics, focusing primarily on sparse Jacobian computation and graph coloring, with triangle counting and clique cover considered as extension cases. Retrieval-augmented pipelines will be used to generate candidate scientific and OpenMP-parallel code variants. Each generated program will be evaluated against trusted reference implementations using correctness tests, numerical equivalence measures, compiler diagnostics, sanitizer outputs, regression testing, repeated execution, and performance-stability metrics.
The expected outcomes include a reusable verification workflow, benchmark datasets, a classification of common failure modes in AI-generated scientific code, reproducible experimental artifacts, and a technical research report suitable for future publication development. The project will provide strong training for students in AI-assisted software engineering, scientific programming, parallel computing, and empirical research methods.
Research area, student roles & skills
Research area: My research focuses on artificial intelligence for scientific software engineering, with emphasis on retrieval-augmented generation, high-performance computing, parallel programming, compiler technologies, numerical correctness, and reproducible computing. I study how large language models can be guided, evaluated, and verified when generating performance-critical scientific code. This work addresses a major limitation of current AI coding tools: generated programs may compile and run, but still produce incorrect, unstable, or nondeterministic results. My research develops verification-oriented methods, benchmark workflows, and empirical evidence to support trustworthy AI-assisted programming for scientific computing, scientific software verification, and parallel software development.
Student roles: The student will contribute to the implementation and evaluation of the VERAM framework under close supervision. The role is hands-on and research-oriented, with emphasis on controlled experimentation, verification, benchmarking, and documentation. The student will help prepare scientific computing workloads, organize reference implementations, execute AI-assisted code generation pipelines, and evaluate generated programs using predefined correctness, reliability, and reproducibility criteria.
A major responsibility will be to help organize benchmark datasets from sparse Jacobian computation and graph-based workloads. The student will help identify input cases, expected outputs, reference results, and evaluation conditions. They will run sequential and manually verified parallel baselines, then compare these trusted implementations with AI-generated variants produced through retrieval-augmented methods.
The student will also assist in verification and validation. This includes compiling generated programs, running automated tests, comparing outputs against reference results, checking runtime errors, and repeating experiments to detect nondeterministic behavior. Where appropriate, the student will use debugging tools, sanitizer reports, timing results, and structured logs to support failure analysis.
Another important part of the role will be performance and stability analysis. The student will collect and analyze data on execution time, parallel efficiency, and run-to-run variability. These results will be organized into tables, figures, and summaries that help explain when AI-generated scientific code succeeds and when it fails. The student will participate in regular meetings with the supervisor, read selected research literature, maintain experimental notes, and contribute to reproducible workflows. Depending on progress, the student may also help prepare reproducible research artifacts, documentation, and presentation materials summarizing benchmark results. Through the project, the student will receive structured training in AI-assisted programming, scientific software verification, parallel computing, empirical software engineering, and reproducible research practices.
Skills required: The ideal student will have strong programming, analytical, and problem-solving skills, with interest in artificial intelligence, software engineering, scientific computing, or high-performance computing. Experience with Python, C, C++, Linux, Git, debugging, testing, or data analysis is desirable. Familiarity with large language models, retrieval-augmented generation, parallel programming, compiler diagnostics, numerical methods, or benchmarking is an asset but not required. Students from computer science, information technology, software engineering, data science, computational mathematics, or related disciplines are encouraged to apply. The student should be careful, reliable, willing to document experiments, and comfortable working in a collaborative research environment.
753. VR Game Safety Training for Hazard Identification
Workplace incidents in oil and gas facilities remain a critical occupational health challenge, often resulting from inadequate hazard identification skills. Traditional safety training methods fail to adapt to individual learner needs, providing the same experience regardless of trainee performance or prior knowledge. This project investigates the design and development of an adaptive difficulty system embedded within a VR-based safety training platform for oil and gas environments, where scenario complexity dynamically adjusts based on real-time trainee performance.
Building on prior groundwork establishing the VR environment and core hazard scenarios, the Globalink intern will explore how trainee behaviour signals within a VR environment can be used to infer competency levels and drive adaptive scenario difficulty. The central research question is: what adaptive strategies most effectively support hazard recognition learning in immersive VR safety training?
The intern will investigate adaptive system design approaches, develop and test proof-of-concept implementations, and document findings in a technical research report. Depending on progress, the intern may also explore user testing with industry stakeholders. This work contributes to the fields of intelligent tutoring systems, serious games, and occupational safety training.
Research area, student roles & skills
Research area: Workplace accidents in oil and gas facilities remain a critical occupational health challenge, often resulting from inadequate hazard identification skills. Traditional safety training methods fail to adapt to individual learner needs, providing the same experience regardless of trainee performance or prior knowledge. This project investigates the design and development of an adaptive difficulty system embedded within a VR-based safety training platform for oil and gas environments, where scenario complexity dynamically adjusts based on real-time trainee performance.
Building on prior groundwork establishing the VR environment and core hazard scenarios, the Globalink intern will explore how trainee behaviour signals within a VR
Student roles: The intern will investigate and prototype an adaptive difficulty system for a VR-based safety training platform. Activities will include: familiarizing themselves with the existing VR environment and hazard scenario framework; conducting a literature review on adaptive learning systems and intelligent tutoring; exploring design approaches for monitoring trainee performance metrics (e.g., hazard detection rate, response time, interaction patterns); developing proof-of-concept implementations within the Unity environment; and documenting findings, design decisions, and recommendations in a final research report. The scope of implementation will be guided by the intern's progress and emerging research insights. The goal is meaningful research contribution rather than a fully delivered system. The intern will meet weekly with the supervising professor and work alongside graduate students. This project suits a self-motivated student who thrives in an exploratory research environment.
Skills required: Applicants should have experience in Python and/or C#, and familiarity with Unity is required. Background in game development, adaptive systems, or applied machine learning is a strong asset. Experience with user analytics, performance tracking, or interactive application development is beneficial. An interest in serious games, intelligent tutoring systems, or human factors is welcomed but not required. The ideal candidate is a self-motivated learner comfortable working within an existing codebase in an exploratory research environment.
The MOST programming language (Kavanagh and Pientka, 2024) extends traditional session-typed languages with concurrent type-level computation to more precisely specify communication protocols. In particular, protocols specifications are very restricted concurrent programs that generate a communication protocol based on past communications in a system. MOST users write both a program and its protocol specification, and MOST's type system ensures that the program respects its specification.
This project will use MOST to specify and verify concurrent algorithms and data structures, including stacks, queues, and concurrent hash-tries. Since concurrent algorithms and data structures are often described in terms of shared memory, they will need to be redesigned to be implemented as a collection of interacting processes. We will draw inspiration from the literature and from open-source software projects to produce other case studies. These case studies will inform future development of MOST, form a reusable library of verified concurrent algorithms, and help develop principles for specifying concurrent algorithms and data structures in this setting.
Research area, student roles & skills
Research area: Type systems syntactically specify or classify programs in order to statically rule out classes of software defects and to enforce abstraction. Traditional type systems classify programs according to the kinds of values they produce, for example, strings or integers. In contrast, behavioural types specify permitted behaviours or interactions between programs. For example, session types encode communication protocols (rules that specify which messages may be sent or received) at the type level, and they ensure that messages passed between well-typed programs respect the specified protocols. This rules out protocol violations and ensures session fidelity.
Student roles: The student will spend the first several weeks reading and learning about session-typed languages and concurrent algorithms. Afterwards, they begin their implementation efforts. In each case, the student will specify the correctness properties they wish to ensure. They will redesign the data structure to implement it as a collection of message-passing processes. They will also specify the required communication protocols in MOST's type system.
The student and I will meet at least weekly to collaborate on their project, and they will regularly interact with a PhD student working on MOST's development. The student will join a large and vibrant mathematics and computer mathematics lab (the LACIM).
Skills required: The student must be comfortable reading and writing mathematical proofs. They must also be comfortable with reading and understanding algorithms. Ideally, the student will have experience with functional programming (Haskell, OCaml, Standard ML, or other).
755. ViLLM-Seg: Vision-Integrated Large Language Models for Explainable Medical Image Segmentation
Despite strong empirical performance in medical image analysis, deep learning-based segmentation systems remain clinically underutilized due to limited transparency and contextual grounding. Clinicians require not only accurate segmentation masks but interpretable, language-grounded reasoning aligned with anatomical structures and clinical knowledge. This project introduces a novel framework, Vision-Integrated Large Language Models (ViLLM-Seg), that fuses large language models with advanced visual encoders and segmentation architectures to deliver context-aware, explainable medical image segmentation. Unlike conventional segmentation approaches that operate purely on pixel-level features, ViLLM-Seg integrates LLM-derived semantic embeddings with visual representations to enable language-guided segmentation decisions. Clinical text descriptions, radiology reports, and anatomical ontologies are encoded by LLMs such as BioGPT or MedPaLM and fused with visual features extracted by Vision Transformers and ResNet encoders. This combination is connected to segmentation decoders including U-Net and DeepLabV3, producing outputs that are grounded in both visual evidence and clinical language understanding. Attention mechanisms link segmented image regions to corresponding textual clinical concepts, delivering transparent and auditable reasoning paths for healthcare professionals. The research program proceeds through three interconnected phases. In the first phase, the ViLLM-Seg architecture will be designed and implemented, establishing the multimodal fusion pipeline that aligns LLM text encoders with visual segmentation backbones. Benchmark datasets including the Medical Segmentation Decathlon will be used for training and evaluation. In the second phase, the language-guided interpretability pipeline will be refined, enabling natural-language querying of segmented regions and automated generation of textual explanations for segmentation outputs. In the third phase, cross-domain validation will assess generalizability across MRI, CT, and X-ray modalities in radiology and oncology applications. The project will quantify gains in segmentation accuracy, semantic alignment, and explanation quality compared to state-of-the-art baselines. Ablation studies will isolate the contribution of LLM grounding versus vision-only approaches. Findings will be disseminated through top-tier venues such as MICCAI.
Research area, student roles & skills
Research area: This research area focuses on equitable, explainable, and automated disease intelligence in medical imaging using reasoning large language models, reinforcement learning, and multimodal learning. The work integrates model-driven, data-driven, and knowledge-driven computing paradigms to advance transparent AI systems for radiology and oncology. By embedding domain priors into deep learning architectures-including hierarchical level sets, variational formulations, and graph-based registration-the program addresses clinical limitations of existing black-box models. Research is validated through multi-organ benchmarks, clinical collaboration, and open-source dissemination, targeting resource-constrained healthcare settings and underserved communities with scalable, interpretable diagnostic tools.
Student roles: The student will play an active and substantive role in developing and validating the ViLLM-Seg framework for language-guided explainable medical image segmentation. Their primary technical responsibilities will include implementing and integrating the LLM text encoder component—using models such as BioGPT or MedPaLM—with visual encoder backbones based on ResNet and Vision Transformer architectures, and connecting these with segmentation decoders such as U-Net and DeepLabV3. The student will design and code the multimodal fusion module that aligns language embeddings with visual feature maps, and will prepare and preprocess multi-organ benchmark datasets, particularly the Medical Segmentation Decathlon, ensuring data quality, alignment, and reproducible train/validation/test splits. A central responsibility will be running training experiments, monitoring convergence, and conducting systematic ablation studies that isolate the contribution of LLM grounding versus vision-only segmentation baselines. The student will implement evaluation metrics assessing both segmentation quality—Dice score, Hausdorff distance—and semantic alignment between textual explanations and segmented regions, enabling rigorous comparison against state-of-the-art methods. They will also contribute to cross-domain validation experiments spanning MRI, CT, and X-ray modalities. Beyond technical development, the student will engage in knowledge mobilization activities. This includes assisting in drafting scientific manuscripts targeting MICCAI, IEEE transactions, and related venues, preparing visual presentations of experimental results, and documenting all methodologies, experimental configurations, and findings in a reproducible and well-organized format. These scholarly communication activities will develop the student's academic writing and conference presentation skills. Throughout the project, the student will acquire advanced technical skills including LLM fine-tuning, multimodal fusion, Vision Transformer integration, and benchmark evaluation in medical imaging. They will simultaneously develop critical soft skills including independent research management, scientific communication, collaborative problem-solving, and adaptability to new methodological challenges. This experience will prepare the student for graduate studies or careers in AI-driven healthcare, medical imaging research, or applied machine learning.
Skills required: The ideal student should have foundational knowledge in machine learning and deep learning, with some experience in Python programming and familiarity with libraries such as PyTorch or TensorFlow. Basic understanding of convolutional neural networks, attention mechanisms, and image processing concepts is beneficial. Exposure to large language models, transformer architectures, or natural language processing is an asset. Prior experience with medical imaging or clinical data is helpful but not required. The student should be motivated to work at the intersection of AI, computer vision, and healthcare, capable of independent problem-solving, and eager to learn multimodal fusion techniques. Backgrounds in computer science,
756. Video-Based Procedural Skill Assessment for Instructional and Vocational Training
Supervisor: Yan Song
University: Thompson Rivers University (Kamloops campus)
This project will investigate automated assessment of procedural skill execution from video, with applications in instructional and vocational training contexts. As skilled trades face growing workforce shortages and traditional apprenticeship models become harder to scale, there is increasing demand for intelligent systems that can observe a learner performing a task and provide meaningful, actionable feedback, without requiring an expert to be physically present. A system that can watch a trainee perform a procedural task and identify where they went wrong and why has direct value for trades education, culinary training, medical procedure training, and similar domains where hands-on skill acquisition is both critical and difficult to scale.
Recent advances in video understanding and vision-language models present a timely opportunity to address this gap. However, most existing work on automated skill assessment focuses narrowly on sports performance scoring, leaving procedural and trade-oriented skills — cooking, mechanical repair, assembly, and similar hands-on tasks — largely unexplored. The project is organized around three open research questions that will guide the investigation over the project period:
RQ1 — Recognition: Can video models reliably recognize and segment the procedural steps of a skilled task, and distinguish correct from incorrect execution at the step level?
RQ2 — Assessment: What signals in video, e.g., pose, motion, object interaction, temporal ordering, are most informative for automated procedural skill assessment, and how do these differ across task domains?
RQ3 — Feedback: To what extent can vision-language models translate video-based assessments into natural language feedback that is useful to learners and instructors?
The project will leverage large-scale publicly available video datasets of skilled human activity, including procedural and instructional domains, as primary experimental testbeds. The methodological approach combines video backbone models for temporal understanding with vision-language models for feedback generation.
Research area, student roles & skills
Research area: My research focuses on vision-based analysis and content understanding using machine learning techniques, with a particular focus on human activity recognition and detection. Research topics include developing deep learning models and pipelines that interpret visual data to identify human activities in various fields and domains, aiming to achieve robust performance in real-world environments. A key application area is the analysis of everyday actions in settings such as smart homes, elderly healthcare, and education, where accurate, efficient activity recognition and understanding can support safety, well-being, and independent living.
Student roles: The student will work closely with the supervising researcher to build and evaluate deep learning pipelines on large-scale video datasets. This is a hands-on, implementation-focused role with the opportunity to contribute to a research publication. 1. Conduct a literature review on existing models and pipelines. 2. Understanding the dataset's annotation structure, writing data loaders, verifying label quality, and preparing train/validation splits for model training. 3. Implement and run experiments using pretrained video backbone models, fine-tuning them on the project's target tasks under the supervisor's guidance. 4. Evaluate model performance against established benchmarks, produce results tables and visualizations, and document findings clearly. 5. Connect the video analysis pipeline with downstream components, including a vision-language model feedback layer. The RA's role here is primarily integration and testing rather than designing the language component. 6. Maintain clear documentation of the codebase, experimental configurations, and results. Regular progress updates will be shared with the supervising researcher in weekly meetings. 7. Draft a manuscript.
Skills required: 1. Demonstrate strong programming skills with a solid foundation in Python. Experience with deep learning frameworks is essential, and experience in PyTorch is highly desirable. 2. Having experience working with images/videos is highly desirable. 3. Comfortable with data preprocessing, training, and evaluating models. 4. Has taken a computer vision or deep learning course. 5. Have skills in literature review, including the ability to read, summarize, and compare existing models and methods. 6. Strong problem-solving skills, the ability to work independently, and an interest in human activity recognition or applied AI in health and smart environments are important.
757. Vision Language Action Based Human Robot Interaction for Legged Wheeled Robots
Supervisor: Meng Cheng Lau
University: Laurentian University (Sudbury campus)
This research project investigates Vision Language Action (VLA) systems and Human Robot Interaction (HRI) using a ROS-based legged wheeled robotic platform. The project aims to explore how artificial intelligence can enable robots to understand visual information, interpret human instructions, and perform assistive tasks in shared environments. The research focuses on practical robotic deployment and interaction rather than purely theoretical simulation.
The project will be conducted at the Laurentian Intelligent Mobile Robotics Lab (LIMRL) at Laurentian University. Interns will work with existing robotic platforms equipped with onboard sensors, cameras, and AI-based perception systems. The project is designed as a technical undergraduate research experience suitable for students with prior exposure to robotics, ROS, Python, Linux, or artificial intelligence.
During the internship, students will assist in developing and testing robotic interaction workflows using Robot Operating System (ROS). Tasks may include integrating computer vision models, processing sensor data, implementing navigation or interaction behaviors, and evaluating robot performance during experimental trials. Students will also explore multimodal interaction methods such as visual understanding, language-guided actions, and simple autonomous decision-making.
The project emphasizes hands-on experimentation, software integration, and collaborative research. Interns will gain practical experience with robotics software development, AI-based perception systems, and real-world robotic testing. Expected outcomes include functional ROS modules, experimental evaluation results, technical documentation, and research presentations. The project may also contribute to ongoing research activities related to autonomous robotics, HRI, and Vision Language Action systems within LIMRL and the LU Snobots robotics team.
Research area, student roles & skills
Research area: My research focuses on intelligent mobile robotics, artificial intelligence, and autonomous systems, particularly Human Robot Interaction (HRI) and Vision Language Action (VLA) models. I study how robots can perceive, understand, and interact with humans using AI based vision, machine learning, and sensor integration. My work involves humanoid and legged wheeled robotic platforms using ROS based systems for navigation, perception, and assistive tasks in real world environments. Current applications include industrial robotics, autonomous mobile systems, and educational robotics. The research emphasizes practical robotic deployment, computer vision integration, embedded AI systems, and real time interaction between humans and autonomous robotic platforms.
Student roles: The students will serve as undergraduate research interns and contribute to the development, integration, and testing of AI enabled robotic systems using a ROS based legged wheeled robotic platform. Their primary role will be to assist with technical research activities related to Human Robot Interaction (HRI), Vision Language Action (VLA) systems, robot perception, and autonomous interaction workflows.
During the internship, students will participate in software development, robotic experimentation, and system integration tasks within the Laurentian Intelligent Mobile Robotics Lab (LIMRL) at Laurentian University. Students will work closely with the supervisor and graduate researchers to configure ROS environments, develop robot control modules, integrate AI based perception systems, and evaluate robotic behaviors in experimental settings.
Specific responsibilities may include collecting and processing sensor data, integrating computer vision models, testing navigation or interaction functions, debugging software modules, and assisting with robotic demonstrations. Students will also help evaluate system performance through experimental trials and document technical findings throughout the project. Depending on their background and progress, students may contribute to multimodal interaction tasks involving visual understanding, language guided actions, or assistive robotic behaviors.
The role requires active participation in weekly meetings, collaborative discussions, and research reporting activities. Students are expected to maintain organized documentation, contribute to technical presentations, and assist in preparing materials for potential conference papers or demonstrations. The internship emphasizes hands-on learning and practical robotics research, allowing students to gain experience with AI integration, ROS-based robotics systems, and real-world robotic deployment challenges.
By the end of the internship, students are expected to contribute to a functional prototype or experimental workflow, technical documentation, and a final research presentation summarizing their contributions and findings.
Skills required: Applicants should have a background in Computer Science, Software Engineering, Robotics, Artificial Intelligence, or related fields. Students should possess basic programming experience in Python or C++, familiarity with Linux systems, and strong problem-solving skills. Prior exposure to Robot Operating System (ROS), robotics, computer vision, or machine learning is preferred. The project is suitable for technically motivated undergraduate students interested in hands-on robotics research and software development. Students should be comfortable working collaboratively in a lab environment, learning new technologies independently, troubleshooting software and hardware systems, and participating in experimental testing, technical discussions, and research documentation activities.
758. Visual Abnormality Detection and LLM-based Analysis for Remote Structural Monitoring
Supervisor: Chul Min Yeum
University: University of Waterloo
Location: Waterloo, Ontario
Start date: 2027-05-02 (flexible)
Disciplines: Computer Science, Engg-Civil, Engg-Software, Engg-Systems and Technology, Engg-Computer
This research project aims to develop a smart visual abnormality detection system for remote structural monitoring. It involves comparing structural models created at different timestamps to automatically detect changes such as added, removed, or altered structural components. The system leverages advanced image analysis techniques combined with Large Language Models (LLMs) to interpret detected changes and generate intuitive, user-friendly reports. Specifically, the LLM provides clear descriptions of anomalies, assesses their significance, and supports decision-making through interactive, conversational interfaces. The goal is to enhance structural safety, streamline inspection processes, reduce maintenance costs, and enable data-driven decisions in managing infrastructure across various industries.
Research area, student roles & skills
Research area: My research focuses on developing a visual abnormality detection system for remote structural monitoring. The system compares structural models from different timestamps to detect changes such as added, removed, or altered components, and leverages Large Language Models (LLMs) to provide intuitive reports, assess the significance of anomalies, and support decision-making. This approach aims to improve efficiency, enhance structural safety, and facilitate data-driven maintenance strategies.
Student roles: 1. Data Preparation and Analysis: - Assist in gathering and organizing structural visual data from multiple time points. - Perform data preprocessing tasks, including data cleaning, model alignment, labelling, and preparation for visual comparison. - Collaborate on enhancing and refining algorithms for 3D model comparison and automated visual change detection.
2. LLM Integration and Report Generation: - Assist in integrating Large Language Models (LLMs) with the visual anomaly detection system. - Help develop methods for translating detected changes into intuitive natural-language reports and interactive Q&A interfaces to support user decision-making.
3. System Testing and Evaluation: - Evaluate system performance using real-world structural datasets or simulated scenarios. - Identify potential improvements, contribute to algorithm refinement, and support usability enhancements based on feedback.
Skills required: The ideal intern should have a solid background in computer science, particularly in computer vision and machine learning. Experience with Python programming and familiarity with image analysis tools or visualization libraries is strongly recommended. Interest or prior experience with large language models (LLMs) such as GPT or Hugging Face frameworks is beneficial. A proactive learner with strong analytical, problem-solving, and communication skills is essential. Background knowledge of civil engineering or structural monitoring is helpful but not required.
759. Visual Motion Analysis from Images and Videos
Supervisor: Li Cheng
University: University of Alberta (Edmonton campus)
This project focuses on analyzing visual behaviors of articulated objects (human and animals) from video feed, a problem that plays a crucial role in many real-life applications ranging from natural user interface to autonomous driving. As a member of the Vision and Learning Lab at the University of Alberta (https://vision-and-learning-lab-ualberta.github.io/), you are expected to work with a graduate student/Postdoc researcher, get familiar with state-of-the-art deep learning, quantum machine learning and computer vision techniques, and gain hands-on research experience on benchmark and home-grown datasets. There are ample opportunities to be involved in exciting research topics in computer vision and to publish
Research area, student roles & skills
Research area: Computer vision and applications
Student roles: You are expected to work with a graduate student/Postdoc researcher, get familiar with state-of-the-art deep learning and computer vision techniques, and gain hands-on research experience on benchmark and home-grown datasets. There are ample opportunities to be involved in exciting research topics in computer vision and to publish research findings at top-tier conferences and journals.
Skills required: Necessary image processing background and working knowledge of calculus and linear algebra; Good programming skill in python.
Large language models (LLMs) are changing how people interact with data and visualizations. In particular, natural language interaction is becoming increasingly common, allowing users to ask questions, describe analytical goals, and request changes to visualizations through text. However, natural language may not always be the most effective or appropriate way to interact with visualizations. In many cases, direct interaction with visual elements, such as selecting, filtering, highlighting, comparing, or manipulating visual representations, may provide a more intuitive and precise experience.
This project explores interaction design for visualization in the era of LLMs. The student will investigate how different interaction methods can support visualization-related tasks when LLMs are involved, and how natural language interaction can be combined with more direct forms of visual interaction. The project may examine when language-based interaction is useful, when direct manipulation is more effective, and how hybrid interaction techniques can better support data exploration, interpretation, and communication.
Research area, student roles & skills
Research area: My research interests lie in the intersection of Human-AI Teaming, Visualization, and Human-Computer Interaction.
Student roles: The student will contribute to research activities such as reviewing related work, analyzing interaction scenarios, designing and prototyping interaction techniques, and evaluating alternative designs.
Skills required: 1. Strong interest in HCI/UX/Visualization 2. Familiar with data analysis 3. Good communication and interdisciplinary collaboration skills 4. Highly self-motivated
761. Visualization of educational data in online learning
Supervisor: Mohammad Ali Dewan
University: Athabasca University (Edmonton campus)
Location: Edmonton, Alberta
Start date: 2027-05-03 (flexible)
Disciplines: Computer Science, Engg-Computer, Engg-Software, Engg-Systems and Technology
Educational institutions capture huge amounts of educational data, especially in online learning. Data mining techniques have shown promises to interpret these data using different patterns. However, understanding the mining patterns and extracting meaningful information from the data require reasonable skills and knowledge for the users. Information visualization, due to its potential to display large amount of data, may fill this gap. In this research project, we will analyze various visualization techniques to represent meaningful information from the educational data, especially dealing with monitoring student performance, understanding learning style, analyzing course and program status, and dropout prediction. We will develop a learner- or instructor-focused effective visualization system for online learning environment.
Research area, student roles & skills
Research area: Artificial Intelligence in Education
Educational Data Mining
Affective Computing
Data Visualization
Machine Learning
Neural Networks and Deep Learning
Incremental Learning
Image Processing and Computer Vision
Student roles: Designing and developing system to apply in distance education Reviewing research articles, and implementing and analyzing systems for performance evaluation Writing report and research articles Doing project presentations in the progress meeting
Skills required: Programming experiences and skills in Python, Java, and D3 Experience and ability to design and build tools or systems Experience in designing user interfaces and data visualization Student must have knowledge in data mining, machine learning, and pattern classification Experience in information collection, re-organization, summarization, and preparing reports Good communication skill
762. Voice-Based Extended Reality for Older Adults
Supervisor: Jaisie Sin
University: Carleton University (Ottawa campus)
Location: Ottawa, Ontario
Start date: 2027-05-03 (flexible)
Disciplines: Computer Science, Science and Technology, Studies Science and Technology, Design, Engineering, Information Studies, Communication, Engg-Computer, Humanities, Journalism, Media Studies, Psychology, Sociology, Social Work, Accounting, Architecture, Art, Banking, Business, Canadian Studies, Classics, Engg-Systems and Technology, Fine Arts, Finance, Film Studies, International Business, Philosophy
This project aims to design/develop an application based on extended reality and voice input to support the needs of older adults (people 60+ years old). Augmented reality (AR, e.g., with Google's Android XR Glasses) will be the primary focus of this project, but this may change depending on evolving needs of the project and users. AR and voice input have individually been explored in the past to support aging, but not as much investigation has been made to combine the two into a natural means of enhanced interaction. Potential use cases of the AR/voice application include but are not limited to: wayfinding, navigation, low-vision support, or social participation assistance. For instance, voice conversation may be used to let the older adult ask naturally for AR to provide magnification, contrast enhancement, object outlines, or arrows to support their daily tasks.
This project focuses on skills development and collaborative research to design immersive solutions for various real-world applications from finance to healthcare, to community engagement and education. This research sits at the intersection of user experience design, digital inclusion, aging, and technology. Aspects core to the project include inclusive design, accessibility, iterative design, and design thinking. The application will first start as a preliminary prototype and refined through several iterations into a developed application. The created application will be evaluated with stakeholders, including older adults, their caregivers, and members of the project team. Evaluation criteria will include but not limited to: usability, accessibility, inclusiveness, acceptability, and user perceptions.
Research area, student roles & skills
Research area: Older adults represent a significant and growing segment of the population, doubling from an anticipated 1.4 billion people (1 in 6) in the world aged 60 years in 2030 to 2.1 billion by 2050. They form a growing segment of users of emerging technologies such as extended reality. However, their unique needs are often ignored in the design and development of these new technologies. Our research aims to explore new ways of understanding and developing technology to ensure digital spaces remain accessible, inclusive, and relevant to aging populations.
Student roles: The student’s primary task will be to create an extended reality application designed for use by older adult users. The student’s responsibilities are divided into the following three tasks:
1) Build the application. The student will conduct market research and literature review to determine a use case that could have valuable research and societal contributions. After settling on an idea with the supervisor, the student will implement a preliminary version of the application.
2) Iteratively refine and improve the application. This will start with internal evaluation with the project team before moving to evaluation with external stakeholders (e.g., adult users). The student will be in charge of conducting internal evaluation focusing on user experience metrics including usability, user perceptions, accessibility, and acceptance. The student will assist with any evaluation with external stakeholders.
3) Report Writing & Project Wrap Up. The student will report on their process and findings in a structured manner (e.g., in a report). The student will prepare project files (e.g., the application source code) in a manner that can be used by the project team for further research and development.
Skills required: REQUIREMENTS: - Studying a relevant discipline (e.g., computer science, human-factors engineering, psychology, user experience design, cognitive science, communications). - Experience using relevant programming languages and environments such as Python, Java, C#, Unity, GitHub, etc. - Willingness to learn the programming languages and environments necessary to design/develop the AR application.
ASSETS: - Prior experience working with extended reality environments, prior experience with head-mounted VR, etc. - Prior experience/expertise with user experience design. - Prior experience conducting research and building products based on user requirements. - Demonstrated teamwork skills. - Experience working with older adults.
763. Voice-Based Health Interfaces - Exploring Limits to Auditory Perception of Health Data
Supervisor: Bradley Rey
University: University of Winnipeg
Location: Winnipeg, Manitoba
Start date: 2027-05-03 (flexible)
Disciplines: Computer Science, Industrial Design and Technology, Information Studies, Studies Science and Technology
Voice-based interfaces are becoming increasingly common in wearable and health technologies. Earbuds, headphones, smartwatches, and mobile devices can now provide spoken feedback while people exercise, commute, or go about their daily activities. These systems have the potential to act as personal health coaches, delivering real-time information about performance, progress, and goals without requiring users to look at a screen. Despite this potential, little is known about how much health-based auditory information people can effectively process, especially while physically active.
This project investigates how voice-based health coaching systems should communicate information during physical activity. Specifically, the project will examine how different levels of information detail affect users' ability to understand and remember details given during spoken feedback. For example, a system might provide a simple update such as “Your pace is improving,” a moderately detailed message including numerical performance information, or a more comprehensive summary that includes trends, comparisons, and recommendations. Finding the limits to these is important for recommending design guidelines for voice-based interfaces.
The project sits at the intersection of human-computer interaction, wearable computing, and personal health informatics. The student will develop a study tool capable of voice-based delivery of personal health data. The resulting system will serve as a research platform for studying information delivery while static and while in motion and for identifying design guidelines for future conversational health interfaces. More broadly, the project contributes to ongoing efforts to create health technologies that support informed decision-making without requiring users to interrupt their activities or divide their attention between exercise and on-screen displays.
Research area, student roles & skills
Research area: My research focuses on designing and evaluating intuitive, real-time, interfaces for personal health data exploration on mobile and wearable devices. We develop and study novel interaction techniques and visualization methods that support on-the-go querying and proactive sense-making of data. Moreover, we explore limits to this exploratory capability while on-the-go. Finally, my research also aims to understand underlying dimensions of personal health data which impact perception and understanding of said data. Overall, we aim to empower users to better understand and manage their health data in everyday contexts for increasingly positive and more personalized health outcomes.
Student roles: The student will play a central role in the design, implementation, and evaluation of a study tool capable of voice-based delivery of personal health data. Working closely with the research team, they will develop a prototype capable of delivering spoken health and fitness feedback that can be used to investigate how people engage with conversational health technologies while exercising.
Responsibilities will include implementing the software infrastructure required to generate and deliver voice-based coaching messages and developing mechanisms to present spoken feedback at varying levels of detail. The student will help design alternative coaching messages that differ in information content, complexity, and presentation style, allowing us to investigate how information density influences user understanding, recall, and overall experience. This work may include implementing logging systems, creating study materials, and developing mechanisms to capture participant responses and performance measures.
A significant component of the project will involve designing and conducting a user study. The initial phase will focus on an online evaluation in which participants experience and assess different coaching messages across a variety of exercise scenarios. If time permits, the student may also assist with a follow-up in-person evaluation involving participants engaged in real or simulated physical activity. The student will help develop study protocols, experimental tasks, questionnaires, recruitment materials, and data collection procedures needed to assess comprehension, workload, recall, preferences, and user experience. They will also assist with piloting, participant recruitment, data analysis, and the interpretation of findings.
Throughout the project, the student will gain experience in conversational user interface design, study development and execution, human-computer interaction research methods, and quantitative and qualitative data analysis. The resulting prototype
Skills required: Students should have an interest in human-computer interaction, conversational user interfaces, wearable computing, AND/OR personal health technologies. Applicants should have experience developing software through coursework, research projects, internships, or personal projects. Experience with either web OR mobile application development (e.g., Java, Swift, etc.) is required. Experience with user-centered design, usability evaluation, or research methods is an asset but can be learned throughout the project. Most importantly, students should be motivated to design technologies that support people in everyday health and fitness activities.
764. When Screens Fade Outdoors: Translating Colour Perception into Display Accessibility
Have you noticed that a phone screen becomes hard to read in bright sunlight, even at maximum brightness? Subtle colour differences that are obvious indoors can vanish outside. This affects everyone, not only people with visual impairments, and it matters wherever screens are used outdoors - messaging and navigation apps, drone operation, and agricultural or construction equipment displays.
Our lab studies how environmental conditions - bright sunlight, and the brightness of the area surrounding a screen - affect how well people tell colours apart on mobile devices. These situational visual impairments affect everyone using digital devices outdoors, yet there are still no design guidelines for keeping interfaces readable across a broad range of lighting conditions.
Your project closes part of that gap using data and models the lab already has. You will work with our existing colour-perception datasets and models of how colour discrimination changes with surround brightness to produce concrete tools for designers. These may include design guidelines, such as which colour combinations to avoid, or software that flags potentially problematic colour combinations in user interfaces.
Depending on which studies are active when you arrive, you may also work hands-on with the lab's custom apparatus - which measures the smallest colour difference a person can reliably detect under controlled light - and join ongoing data collection in chromatic (hue) or achromatic (greyscale) discrimination, with an outdoor component if conditions allow. Any new data you help gather would feed directly into these tools.
You will work closely with the supervisor and other students in an active HCI and accessibility research group. Training and the relevant ethics approvals are in place, and the work is primarily lab-based and fully in-person at the University of Guelph.
Research area, student roles & skills
Research area: This research investigates situational visual impairments (SVIs): the everyday difficulty people have reading emissive screens in bright environments, such as a phone in sunlight or equipment displays on agricultural and construction machinery. Our lab works in human-computer interaction, accessibility, and perceptual colour science, studying how well people distinguish colours under different viewing conditions. We turn these findings into concrete, practical tools - design guidelines and software - that help designers of mobile and outdoor displays build interfaces that stay readable and accessible across a wide range of lighting conditions.
Student roles: You will join the lab as a full member of the research team for the internship, working in person at the University of Guelph.
In your first few weeks, you will meet the team, learn how our lab measures colour discrimination, and become familiar with our existing datasets, models, and software. Together with the supervisor, you will then decide on the specific focus of your project, shaped to your strengths and the lab's active work.
The core of your work will be turning the lab's colour-perception data and models into something designers can use to create more accessible interfaces. Depending on your focus, this may mean building software that takes a user interface and flags colour combinations that may be hard to tell apart outdoors, or analyzing our discrimination data and using it to propose design guidelines for outdoor-friendly apps - or a combination of the two. You will design, build, test, and document your contribution, with regular feedback from the supervisor and graduate students.
You may also spend time with the lab's custom apparatus and take part in ongoing studies, helping run sessions and collect data under controlled lighting. This would give you first-hand understanding of where the data come from and how perceptual measurements are made.
Throughout this project, you will meet regularly with our team to share your progress. By the end of the internship, you will present your contribution — a tool, an analysis, or a set of proposed guidelines — to the team with a short report. You may also have an opportunity to represent your work at a poster session. You will leave with hands-on research experience, a concrete portfolio piece, and a clearer sense of how perceptual science and human-computer interaction come together.
Skills required: We are looking for a motivated undergraduate in computer science or a related field. You should be comfortable programming in a general-purpose language (such as Python or JavaScript/TypeScript) and able to organize, analyze, and clearly present data. Curiosity, attention to detail, and the ability to work independently and in a team matter more than prior experience. Coursework or projects in HCI or accessibility would be a strong asset, but no background in these areas is required - we will help you learn. Good written and spoken English is important for documenting your work and collaborating with the team.
765. Which Memory Signals Make an NPC Remember? An Ablation Study of Emotion-Grounded Retrieval
Supervisor: Cristiano Politowski
University: Ontario Tech University (Oshawa campus)
Millington's "illusion of intelligence" still applies in 2026, but in LLM-driven NPCs the illusion collapses at memory: the character forgets what mattered, remembers what didn't, and reacts the same way whether the last interaction was a gift or a betrayal. A returning player who befriended a tavern keeper is greeted as a stranger one week later. LLMs are stateless, so a retrieval system must pick which past events to feed back. Feed too little, the NPC forgets the player's name; feed everything, the context window saturates and the middle is ignored.
Park et al.'s Generative Agents (2023) blend recency, importance, and relevance into one score, and every follow-up (MemGPT, MemoryBank, EM-LLM, MOOM) inherits the blend without ablating it. Shipped systems (Inworld, Convai, NVIDIA ACE) publish no ablations either. Psychology suggests the blend is wrong: emotion strengthens consolidation (Cahill & McGaugh 1998) and mood biases recall (Bower 1981), yet neither signal lives in current LLM-NPC memory.
This project decomposes the blended score into five separable retrieval signals — recency, affect intensity, event-type gate, hybrid (BM25 + cosine) relevance, and mood congruence — and wires the NPC's emotional state into both retrieval and generation. The system runs locally on an 8 GB consumer GPU using Ouro 1.4B with a per-turn loop of write, state-update, score, inject, generate.
The intern will answer three research questions: (RQ1) which memory signals drive retrieval quality; (RQ2) whether emotional state changes what the NPC says or only what it remembers; (RQ3) whether composite scoring opens new attack surfaces under 20 MINJA-adapted probes and holds under 600 ms per turn. Deliverables: open code, 30 hand-labeled scenarios, dialogue traces, and a paper draft for AIIDE Demo Track or FDG.
Research area, student roles & skills
Research area: My research focuses on empirical software engineering for video games, with particular emphasis on AI-driven non-player characters (NPCs), gameplay QA, and automated testing of interactive systems. I investigate how large language models (LLMs) can be integrated into game NPCs to deliver lifelike dialogue and adaptive behaviour, while addressing barriers including latency, hallucination, long-term consistency, and designer control. My work spans systematic literature reviews of LLM-NPC architectures, persona and emotion engineering, and empirical evaluation of dialogue quality and persona drift. Outcomes inform research practice and commercial adoption, with attention to safety, reproducibility, and the needs of small to mid-sized studios.
Student roles: The intern will lead an ablation study of emotion-grounded NPC memory across 12 weeks of structured work.
Weeks 1–2 (Onboarding and scenario design): Study the lab's systematic mapping of 42 LLM-NPC papers and read Park et al., MemGPT, Emotional RAG, MOOM, and MINJA in depth. Configure the local Ouro 1.4B stack on an 8 GB GPU and finalize a two-character roster with the supervisor: Dawn Whitmore, an invented tavern keeper providing full ground-truth control over a trust → betrayal → reconciliation arc, plus one canonical character (Kenny from Telltale's *The Walking Dead*, with a Stardew Valley villager as IP-safe fallback).
Weeks 3–5 (Architecture implementation): Build the five-step per-turn loop, implement the SQLite memory schema with affect tags, code the five separable scoring signals (recency, affect intensity, event-type gate, hybrid relevance, mood congruence), and implement Park's blended baseline as control. Pilot the protocol with Dawn.
Weeks 6–8 (Experimental phase): Run RQ1 with seven scoring variants across three retrieval modes (42 configurations per character); run RQ2 across three emotional-state conditions with memory and input held fixed, measuring both Jaccard distance between top-k sets and tone shift via NPC valence-arousal classifier and blinded LLM-judge; run RQ3 with 20 MINJA-adapted attacks plus per-stage latency profiling at p50 and p95.
Weeks 9–11 (Analysis and iteration): Compute 95% confidence intervals and Bonferroni-corrected comparisons across variants. Apply the blinded LLM-judge protocol for the RQ2 generation shift. Iterate prompt construction on RQ3 failure modes.
Week 12 (Writing and release): Deliver the final technical report and lab presentation, contribute to a paper draft for the AIIDE Demo Track or FDG, and prepare an open-source release including code, the 30 hand-labeled scenarios, and dialogue traces.
The intern will meet weekly with the supervisor and collaborate with lab members on adjacent NPC tasks.
Skills required: Python proficiency, including LLM APIs and prompt engineering, with experience running local inference (e.g., via Hugging Face). Working knowledge of retrieval-augmented generation, embeddings, vector stores, and BM25. Statistical literacy: confidence intervals and multi-variant corrections. Bonus: dialogue systems, conversational AI, affect computing, or game design. The student should be comfortable reading academic literature, designing controlled experiments, and writing technical reports in English.
766. World Models for Airport Operations: From Prediction to Counterfactuals
Supervisor: Yassine Yaakoubi
University: Concordia University (Montréal campus)
Location: Montreal, Québec
Start date: 2027-05-03 (flexible)
Disciplines: Computer Science, Aviation, Engg-Industrial, Engg-Systems and Technology, Mathematics
Today's airport digital twins are hand-built discrete-event simulators (expensive to construct, slow to calibrate, and stale the moment operations change), and today's best predictors are correlational: they forecast delay cascades without telling an operator why a disruption occurred or what would have happened under a different action. Sequence pretraining has reached the airport surface (Amelia, 2024, a transformer trained on 9.4 billion tokens of SWIM-derived trajectories), but it predicts next positions only; no generative world model yet covers heterogeneous operational events with calibrated conditional rollouts. Causal methods in aviation model delay propagation between airports (CausalNet, 2024) rather than root-cause attribution, with no counterfactual evaluation of interventions and no benchmark carrying ground-truth causal structure.
The intern will build that world model, the twin that closes the loop: (1) design a tokenization scheme for heterogeneous airport event streams (gate changes, turnaround milestones, de-icing, weather), assembled from public corpora (Amelia, published schedules), our turnaround simulator, and synthetic operational logs; (2) pretrain transformer-based event-sequence models for next-event and time-to-event prediction with calibrated uncertainty, evaluated against discrete-event-simulation and temporal-point-process baselines; (3) implement root-cause attribution and counterfactual estimators on top of the learned twin ("had we pre-positioned one more de-icing truck at Pad 3, would the cascade have been avoided?"), benchmarked against correlational GNN baselines; and (4) release the first open benchmark for causal delay attribution (the simulator supplies the ground-truth causal structure the field lacks), with a what-if API and demo dashboard.
The work addresses the top-ranked use case in our airport partner's innovation roadmap and our most distinctive scientific thread, with a demonstration planned for the partnership's first research showcase. Expected outputs: an open event-stream benchmark with pretrained baselines, the causal-attribution benchmark, and a co-authored submission (NeurIPS workshops, AAAI, or TRISTAN); the methods generalize to ports, rail, and power grids.
Research area, student roles & skills
Research area: Our group builds cyber-physical digital twins that do not just simulate but learn, optimize, and explain. Within our research partnership with a major Canadian airport (whose top-priority use case is a self-learning digital twin of its operations), we develop generative world models of operational event streams that stay calibrated against live data, support what-if rollouts for planners, and attribute disruptions to their root causes. The approach combines graph and sequence learning (building on our work with Boeing and EUROCONTROL data).
Student roles: The intern works closely with the PhD students on our airport program, with weekly 1:1 supervision and group meetings, and owns the project in stages: the data pipeline and pretrained twin first, the causal layer on top of it second. Weeks 1-2: survey event-sequence modeling and causal-attribution literature; specify the tokenization schema with the team. Weeks 3-5: build the data pipeline and classical baselines (discrete-event simulation, temporal point processes). Weeks 6-9: pretrain and calibrate the event-sequence model on our compute clusters (cluster and cloud-credit allocations are in place); implement attribution and counterfactual estimators over the learned twin. Weeks 10-12: rollout-fidelity and intervention-policy evaluation, benchmark packaging, the what-if API and demo dashboard, and a co-authored manuscript draft; final presentation to the group and, if timing allows, a partner-facing demo (public and synthetic data only). Deliverables: open benchmarks with pretrained baselines, the what-if API, and a manuscript draft. Strong interns continue remotely toward publication and are prioritized for funded graduate positions on this research program.
Skills required: Strong Python and PyTorch; experience training transformer models (coursework or projects); solid probability and statistics. Familiarity with causal inference (structural causal models), temporal point processes, discrete-event simulation, or flight-data sources is an asset; none is required. Software-engineering maturity (APIs, testing, Git) matters: the model must become a tool others use. Students from aviation or air-transport engineering, aerospace, industrial engineering, or computer science are all welcome; the project needs a bit of each, and no single background covers it.
767. World Models for Reinforcement Learning and Robotics
Supervisor: Igor Gilitschenski
University: University of Toronto (Mississauga campus)
The primary barrier in making RL practical is low sample efficiency which arises from the sparse learning signal available from rewards. World models help circumvent this by providing the much richer supervision signal in the dense world modelling objective. We have several projects aimed towards this direction:
- Self-supervised learning objectives in model-based RL. The most sample efficient model-based RL algorithms leverage some form of dense prediction of next observations from current ones.
- Quickly adapting pretrained world models for new embodiments and environments. Pretrained action-conditioned video models provide a good starting point for dynamics prediction but are unreliable in embodiments and enviroments
Research area, student roles & skills
Research area: Our research aims to leverage world models to improve sample efficiency for RL. We are working on developing model-based RL algorithms that apply techniques from self-supervised learning. We also aim to leverage large scale models pretrained on video for sample efficient real robot learning.
Student roles: The student will be conducting research in a fast moving field. This includes reading papers, defining research questions and solution ideas together with the graduate student supervisors, implementing the proposed solutions, implementing tests and experiments, and presenting their findings to the lab.
Students will be using cluster compute resources for model training and simulation eval. We also have a selection of real robots that may be used for experiments.
Skills required: Required: - Strong programming skills in Python - Close familiarity with software engineering tools such as IDEs, remote ssh, and Git - Introductory course on deep learning
Bonus/nice to have - Familiarity with reinforcement and imitation learning - Experience working with simulators or real life robotics
768. Zero-Trust Network Architecture in Hybrid Multi-Cloud Environments: A Measurement-Driven Evaluation
Supervisor: Sina Keshvadi
University: Thompson Rivers University (Kamloops campus)
This project investigates the design and real-world performance of Zero-Trust Network Architectures across hybrid multi-cloud environments (AWS + Google Cloud + simulated on-premises).
The student will implement a zero-trust networking model using modern cloud-native technologies (such as AWS PrivateLink, Google Cloud Private Service Connect, service mesh, and identity-aware proxies). They will then conduct extensive measurements to evaluate the impact of zero-trust policies on latency, throughput, scalability, and operational overhead.
The project aims to provide practical guidelines for organizations adopting zero-trust models in multi-cloud settings and to produce research results suitable for publication in a top-tier networking or cloud security venue.
Research area, student roles & skills
Research area: My research focuses on computer networking and network measurement, with a particular interest in modern cloud networking architectures. My expertise lies in evaluating the performance, reliability, and security of large-scale networks, including hybrid and multi-cloud environments. I combine theoretical knowledge from my PhD with practical cloud skills (AWS and GCP) to study how real-world cloud networks behave under different conditions.
Student roles: The student will take a leading hands-on role and will be responsible for:
Designing and deploying a zero-trust networking architecture across AWS and Google Cloud Implementing and configuring security controls (mutual TLS, identity-aware access, micro-segmentation) Developing automated measurement frameworks to evaluate performance and security overhead Running extensive experiments under various traffic and failure scenarios Analyzing results and creating visualizations Contributing significantly to the research paper (methodology, experiments, results, and discussion) Participating in weekly project meetings and presenting findings
Skills required: Solid understanding of computer networking fundamentals and basic security concepts (firewalls, encryption, authentication) Experience with Linux and command-line tools Good programming skills in Python (scripting and data analysis) Interest in cloud security and modern network architectures Strong analytical and technical writing skills
This project will develop and test new image processing paradigms on ophthalmic images, namely OCT and Fundus photos. Tius will be used to do ophthalmic diagnosis
Research area, student roles & skills
Research area: My work is interdisciplinary ranging from optical science and engineering to visual perception. I am particularly interested in applying AI for diagnosis of ophthalmic disorders.
Student roles: The student working with the PI and a clinical consultant will develop algorithms to process images of various parts of the eye. The student will analyze the results in terms of various metrics Do extensive literature surveys write draft of a paper for publication and/or presentation at a major conference.
Skills required: 1. good programming skills (Python, C++, etc.) 2. familarity with AI software wuch as UNET, RESNET, GANs, etc. 3. Knowledge of math especially linear algebra 4. image processing
770. generative modeling for robot policy learning
The intern will design and improve robot policy learning methods by comparing generative modeling approaches such as Diffusion Policy and flow matching. The goal is to evaluate their effectiveness in generating robot actions and determine which approach achieves superior performance on robotic manipulation and control tasks.
Research area, student roles & skills
Research area: I am interested in the use of generative modeling for robot policy learning. Generative approaches, such as Diffusion Policy, can effectively leverage multimodal data—including images, language, and robot states—to learn flexible policies that generalize across a wide range of environments and tasks. Part of my research focuses on developing generative models for robot policy learning and investigating how the resulting policies can be used for robot control.
Student roles: The intern will work on robot policy learning. The intern will develop the code pipeline for developing and testing different robot policy learning schemes using generative modeling. They will compare two classes of generative models: 1. diffusion-based policies (e.g., Diffusion Policy) 2. flow matching / flow-based generative methods. The comparison will be based on how well they support robot action generation for control purposes.
Skills required: Helpful background: Experience with PyTorch and GPU computing, and some familiarity with the mathematical concepts underlying generative modeling.
771. mFieldtrip application development with RAG-Chatbot
Supervisor: Qing Tan
University: Athabasca University (Edmonton campus)
This project aims to upgrade an open mobile application used for a self-guided field course in physical geography. The field course involves real excursions across Alberta, where students are guided by a mobile application in the absence of direct instructor supervision. The app supports learning by drawing students’ attention to key geographical features and providing associated instructional content and field-based activities.
The project will require the student to use a cross-platform development environment, such as Flutter, to develop and enhance the mFieldtrip mobile application (iOS/Android). Advanced features will be implemented using up-to-date technologies, including data analytics, optical character recognition (OCR), and field-based tools to support data collection and analysis.
In addition, the project will focus on finalizing the mobile application by enhancing usability and refining the user interface design to improve accessibility, intuitiveness, and overall user experience, ensuring the app is both easy and pleasant to use in real field conditions.
Research area, student roles & skills
Research area: I earned my PhD in Cybernetics Engineering for Robotics from the Norwegian Institute of Technology. My research interests include Location-Based Technologies, Cyber-Physical System, Telepresence and Robotics, Mobile Computing, Cloud Computing, Machine Learning, Augmented Analytics, and Big Data Analytics. Recently my research mainly focuses on the Telepresence Robot Empowered Smart Lab, Machine Learning in Open Learning Content Creation with Blockchain, and the Privacy Issues and Societal Impacts of emerging technologies.
Student roles: The student will work closely with the PI and Co-PI, Prof. Frédérique Pivot, and will be directly involved in the advanced development of the mobile application under the PI’s supervision. The student will use a cross-platform development environment to develop and upgrade the mFieldtrip mobile app for both iOS and Android platforms. The role also includes contributing to the experimental implementation of new features and participating in field testing to evaluate app performance and usability in real-world field conditions.
Skills required: A student should have strong programming skills in Android mobile application development, including familiarity with cross-platform tools such as Flutter, Firebase, C++, Java, and JavaScript, as well as AI tools like LM Studio and MCP. Experience in mobile and web app development is an asset, along with knowledge of QGIS. The student should also have solid UI/cosmetic design skills to create an intuitive and visually appealing user experience, and experience publishing apps to both the Apple App Store and Google Play Store is highly desirable. Strong English communication skills are required. Suitable backgrounds include Computer Science, Computer Engineering, and Geoinformatics.
772. self-supervised anomaly detection in images and videos
Supervisor: Mohand Said Allili
University: Université du Québec en Outaouais (Gatineau campus)
This project aims to develop novel self-supervised learning techniques for anomaly detection in images and videos without requiring manually annotated abnormal samples. The central objective is to learn robust representations of normal visual patterns from large collections of unlabeled data and identify anomalies as deviations from the learned normality model. The proposed research will investigate contrastive learning, masked visual modeling, and temporal prediction mechanisms to capture both spatial and temporal regularities in visual scenes. A particular focus will be placed on exploiting contextual relationships between image regions and motion patterns across video frames to improve the detection and localization of subtle anomalies. The innovation potential of the project lies in moving beyond traditional reconstruction-based approaches by integrating self-supervised representation learning with graph-based relational modeling and uncertainty-aware anomaly scoring. Such a framework could enable the detection of previously unseen anomalies while providing interpretable localization maps that highlight abnormal regions or events. The proposed methods will be evaluated on applications such as industrial inspection, deepfake detection, intelligent surveillance, disaster monitoring, and medical image analysis, where abnormal events are inherently rare and diverse. By eliminating the dependence on costly anomaly annotations and leveraging the vast amount of unlabeled visual data available today, the project has the potential to significantly advance scalable and deployable anomaly detection systems capable of operating in dynamic real-world environments.
Research area, student roles & skills
Research area: My specialized research area is artificial intelligence, machine learning, and computer vision, with a particular focus on developing robust and interpretable methods for image and video understanding. My research investigates deep learning, graph neural networks, vision-language models, and multimodal learning for applications including semantic segmentation, anomaly detection, activity recognition, multimedia forensics and medical image analysis. A central theme of my work is the design of learning algorithms capable of operating under limited supervision, domain shifts, and uncertain real-world conditions.
Student roles: The student will actively participate in all stages of the research project, including literature review, algorithm development, experimental evaluation, and dissemination of results. The primary responsibilities will involve implementing and testing self-supervised learning models for anomaly detection in images and videos, preparing and preprocessing datasets, conducting experiments, and analyzing the obtained results. He/she will contribute to the design and evaluation of novel anomaly detection strategies, compare different learning approaches, and document findings through technical reports and research presentations. He/she will work closely with the research supervisor and other team members, participate in regular meetings, and contribute to the preparation of scientific publications when appropriate.
Skills required: The ideal student should have a strong background in computer science, artificial intelligence, or a related discipline. Prior knowledge of machine learning is highly desirable, particularly convolutional neural networks, transformers, and representation learning. Experience with Python and deep learning frameworks such as PyTorch is recommended. Familiarity with computer vision, image processing, or video analysis techniques would be an asset. The student should possess good analytical and problem-solving skills, an interest in research and experimentation, and the ability to read and understand scientific literature.