Projets novateurs réalisés

Sequential optimal experiment design for an industrial polymerization process

This project aims to develop an accurate model of an industrial polymerization process using a minimal number of experiments. The approach combines basic physical understanding of the process with data-driven modeling techniques to efficiently capture process behavior. An adaptive experimental design strategy will be used, where each experiment is chosen based on results from previous ones to maximize learning while minimizing experimental effort. The resulting model will support process simulation, design, and optimization, helping improve efficiency in polymer production.

Medial Lateral Separation and Age Differences on Collision Avoidance Behaviours

Avoiding collisions with other pedestrians is a daily necessity for individuals across all age groups. Unfortunately, age-related declines in older adults’ motor and sensory systems can negatively impact their ability to safely navigate the environment. For instance, decrements in muscular strength, balance, and cognitive function can impair one’s ability to adapt locomotion to environmental constraints. These age-related declines in sensory-motor processing appear to be particularly prominent when navigating dynamic environments, where timely and accurate responses are critical to safe locomotion. The objective of the proposed study is to determine whether age-related changes to sensory-motor processing affects collision avoidance behaviours. It is believed that older adults’ avoidance behaviours (onset of avoidance and clearance at the time of crossing) will be more variable, suggesting a decrease in accurate sensory-motor processing. The proposed study will help identifying age-related differences in decision-making, contributing to the broader understanding of sensory-motor processing decline. Ultimately, by addressing the nuanced challenges faced by older adults in navigating real-world urban environments, this project aims to foster safer and more inclusive communities, while advancing our understanding of human behaviours across the lifespan.

Quantum Machine Learning for Adversarially Robust Cybersecurity: Network and Phishing Attack Detection

This project aims to enhance the detection of modern cyberattacks by investigating quantum machine learning (QML) and quantum neural network (QNN) approaches alongside classical machine learning and deep learning models. As cyber threats evolve, attackers increasingly employ stealthy strategies, such as low-and-slow attacks and phishing/spear-phishing, designed to evade traditional detection systems. Moreover, many existing models remain vulnerable to adversarial attacks, where subtle input manipulations can lead to incorrect predictions.

To address these challenges, the project will first establish strong classical baselines using traditional ML algorithms and deep neural networks (DNNs). Subsequently, quantum-based models, including QML and QNNs, will be implemented to assess their potential advantages in feature representation, learning efficiency, and robustness. A comprehensive comparative evaluation will be conducted using standard metrics such as accuracy, precision, recall, F1-score, evasion rate, and robustness under adversarial perturbations. The objective is to determine whether quantum approaches can improve the reliability and resilience of cybersecurity detection systems against advanced and evasive threats.

Causal Deep Learning for Cross-Country Generalization in Multimodal Remote Sensing

This project aims to develop an automated and robust satellite-based pipeline to map hedgerows, which are import for carbon storage, biodiversity and agricultural landscapes. Even though a large amount of satellite data is available, current methods often rely on manual labeling and are sensitive to changes in region or sensors. During this internship, a machine learning model will be trained on annotated data in France and evaluated on manually labeled data in Canada to study cross-country transfer. Causal machine learning methods will be added to the model to reduce sensitivity to spurious correlations and improve generalization. The expected final result is a semi-automated scalable and reliable pipeline for detecting hedgerows across region while reducing labeling effort using foundation models.

Conductive and printable hydrogel with wet-adhesive properties

Electrical stimulation therapy has shown significant potential for activating neural precursor cells, which can undergo activation, differentiation, proliferation, and cathodal migration under an applied electrical field. Currently, this relies on rigid metal electrodes or micro-fabricated soft electrodes, both of which require transportation to the treatment site and surgical implantation. These approaches can lead to complications such as inflammation and surgical errors, highlighting the need for advances in in situ additive manufacturing of conformable surface electrodes. This requires electrode materials that are electrically conductive, printable, mechanically compliant, and capable of strong adhesion to tissue surfaces under wet conditions due to the presence of cerebrospinal fluid (CSF). Hydrogels have emerged as well-suited materials, however, most reported formulations rely on external triggers such as temperature changes or UV light for crosslinking and typically satisfy only one or two of the three key functional requirements: printability, electrical conductivity, and wet adhesion. This project aims to address this gap by developing a biocompatible hydrogel that integrates all three functionalities for direct in situ deposition. The outcomes of this work will introduce a new material platform and manufacturing strategy to the host lab, while expanding hydrogel materials development and additive manufacturing innovation within the TSMART lab.

The effectiveness of stormwater ponds in water quality management

Stormwater ponds are widely used across urban areas to improve water quality, yet their actual long-term performance under different design standards remains poorly quantified. This project will evaluate how efficiently these ponds remove sediment and associated pollutants by applying a mass-balance approach and tracking deposition rates through time. Although the technology is assumed to work, the lack of empirical data on how quickly ponds fill, how their treatment capacity changes as they accumulate sediment, and whether different designs perform consistently leaves major gaps in stormwater planning. By comparing multiple ponds built to varying standards, the project will determine whether they function similarly or diverge in performance, and how these differences influence maintenance needs, lifespan, and water-quality outcomes. The results will provide municipalities with a clearer understanding of stormwater pond efficiency, sediment accumulation dynamics, and design-driven performance variability, supporting more informed decisions about future infrastructure and long-term watershed protection.

Leveraging Large Language Models to Investigate Machine Learning Specific Code Smells: An Empirical Study

This project seeks to improve the understanding of how Machine-Learning (ML)-specific code issues arise during software development by using Large Language Models to analyze developers’ code changes over time. An automated framework will be created to track code updates, detect ML-related issues, and identify the type of developer activity (feature development, bug fixing, enhancement, or refactoring) that introduced them. The project will strengthen collaboration between the institutions through shared supervision of the intern and interactions, and discussions of ongoing experiments and results. The results, datasets, and tools produced during the project will lay the foundation for long-term collaboration, student co-supervision, and future funding opportunities.

L2M – Smart Lung Physio: LTC Adoption Readiness

Smart Lung Physio is validating adoption readiness for a respiratory care solution in long-term care. The project will map decision-makers, identify adoption drivers and barriers, and define non-clinical pilot readiness criteria to support an evidence-based go/no-go decision for an LTC pilot.

L2M – A wearable home-use solution for hallux valgus management

This four-month project will support the early development of a home-use wearable approach for hallux valgus by reducing key market and validation uncertainties. The work will focus on structured discovery with clinicians to understand real-world needs, current practice patterns, safety expectations, and usability requirements for potential home use. In parallel, the project will review existing research and commercially available approaches in this area to clarify what is already known and where gaps remain. Findings will be summarized into a clear set of user and stakeholder requirements and an initial roadmap for next-step testing and evaluation. These outputs will help the partner organization make informed decisions on product direction, prioritization, and planning for future validation without disclosing proprietary implementation details.

L2M – Selective Lithium-Ion Capture Using Hybrid Ion-Imprinted Superabsorbent Polymers

This project will evaluate a new material designed to selectively capture lithium from battery recycling solutions and lithium extraction streams in a cleaner and more efficient way. Current recovery methods often rely on high chemical use and energy-intensive processes that increase cost and environmental impact while recovering lithium inefficiently. The project will work directly with industry stakeholders to understand their needs, test the material under realistic conditions, and assess how it could be integrated into existing recovery processes. The expected outcome is a practical pathway toward a lower-cost, lower-impact lithium recovery solution that improves resource efficiency, supports Canada’s battery supply chain, and enables more sustainable recycling and extraction operations.

Triboelectric nanogenerators-new materials development (KH)

Clean electricity is energy that is both sustainable and non-polluting. Such sources of energy are ideal for mitigating climate change. For the full benefit of clean energy generation to be realized, sustainable practices must be integrated into the development of clean technologies. This requires using materials that are both renewable and recyclable. These can indeed offset the otherwise large carbon footprint of clean energy generating technologies. The project will endeavor to integrate sustainable materials that are also compatible with electricity generation by mechanical actuation. Elastomers that are ideally suited for such actuation will be developed exclusively from renewable materials with the advantage of being compostable. The project will evaluate how renewable materials can be compounded to maximize electricity generation without sacrificing sought after elasticity. The capacity of the elastomers to generate electricity by mechanical actuation will also be evaluated. A major benefit of the project will be zero polluting technology for generating electricity that can be recycled by composting.

LoC-SensAI: Low-Cost IoT RF Sensing with AI for Real-Time Precision Irrigation Decision

The LoC-SensAl project develops an affordable system to help farmers manage irrigation in real-time using low-cost sensors and Artificial Intelligence (AI). One intern will focus on designing the physical sensors to measure soil moisture, while the other will create AI models to turn that data into precise watering schedules. This technology will help farmers save money and water while improving crop yields. Ultimately, the project promotes sustainable agriculture and protects natural resources in both Canada and Tunisia by making “smart farming” tools more accessible.