Projets novateurs réalisés

L2M_Business Strategy for CELLECT’s Research and Commercialization

Despite advancements in women’s health diagnostics, cervical cancer screening rates remain critically low due to the invasive nature of traditional collection methods like Pap smears and self-swabs. CELLECT is pioneering a non-invasive, nanotechnology-enabled collection device—the CELLECTPad—that passively captures high-quality cervical and epithelial cells during menstruation. By removing the discomfort and accessibility barriers of existing methods, CELLECT aims to transform the landscape of gynecological screening and reproductive health research.

Agent conversationnel d’intelligence artificielle de sources de littérature scientifique en santé mentale

Le projet consiste à développer un agent conversationnel intelligent, c’est-à-dire un assistant virtuel capable de répondre aux questions en utilisant des informations scientifiques fiables. Contrairement aux outils classiques de recherche sur Internet, cet agent est conçu pour fournir des réponses basées uniquement sur des données validées par des experts et provenant de sources reconnues en santé mentale. Ce projet vise à résoudre un problème majeur : la surcharge d’informations souvent contradictoires ou non vérifiées disponibles en ligne sur la santé mentale. En filtrant les données pour ne conserver que les plus pertinentes et crédibles, l’agent aide les utilisateurs à prendre des décisions éclairées, qu’il s’agisse de comprendre un diagnostic, de choisir une stratégie de traitement ou de mener une recherche académique.

Safety constrained learning for industrial manipulators

1)
Ocado group builds the Ocado Smart Platform, an end-to-end ecommerce, fulfilment, and logistics solution for smart online grocery businesses. This team develops cutting edge technologies across robotics, artificial intelligence, machine learning, and data science to support various stages of warehouse automation and logistics. Each application requires robots with specific capabilities tailored to different tasks.
2)
Behaviour cloning, imitation learning and reinforcement learning are growing in popularity for learning behaviours that generalize across tasks, but deploying these methods on industrial robots presents unique safety challenges. This project aims to identify, discover and develop robust methods for training that ensure safe operation of industrial robots performing pick and place tasks.
3)
For Ocado, this means increased efficiency and safety in their warehouses. This in turn improves the speed, efficiency, and reliability of the grocery distribution industry. Ocado will provide the robots, simulation software, task definition, and guidance to support this project.

Building a Scalable Data Platform Infrastructure for Smart Fleet and Urban Analytics

Geotab creates hardware that it installs into customer vehicles. Telematic and video data from this hardware is collected and aggregated by its data platform where it can be used for analysis and machine learning endeavours to provide insights to
customers. Large volumes of data are generated and must be ingested efficiently and accurately as well streamed to customers in real time. The data, in particular video data, must also be stored in an efficient manner. Lastly the platform needs to be scalable as the quantity of data continues to increase. The data platform currently has methods to handle these needs, but Geotab is a rapidly growing company. New techniques are needed to ensure the data platform remains performant and reliable.
Successful research will benefit Geotab by allowing it to further expand its operations while ensuring low latency, low cost, and reliability. Society as whole will benefit due to Geotab’s software allowing its customers to manage their fleets of vehicles more
efficiently, safely, and sustainably.

Inventing the Future of AI Applications

AXL is a venture studio focused on developing cutting-edge AI-powered applications. Their mission is to create the next generation of human-augmenting AI technologies by identifying real-world challenges and exploring novel technologydriven
solutions. AXL conducts applied research in AI and Human-Computer Interaction, with a focus on product development and prototyping to fuel innovative startups. This project directly aligns with AXL’s mission by addressing a key challenge: how organizations can effectively leverage advanced machine learning models, particularly large language models (LLMs), to build novel interactive systems. AXL faces the challenge of leveraging rapidly evolving advanced machine learning models to create powerful, accessible, and user-friendly applications. While technologies like Large Language Models have proven useful in tasks including summarization, question answering, and decision making, user trust remains a significant barrier. This research aims to uncover necessary knowledge to implement user interfaces that enhance user accessibility and build user trust in model outputs, paving the way for development of impactful, interactive AI products in high-stakes sectors like finance. his research will contribute to advancing the AI sector by improving user trust in model outputs and enhancing human interactions with AI applications across industries. The project’s success will have wide-reaching social benefits, particularly by enhancing financial literacy and accessibility to trustworthy financial advice. AXL will benefit from the development of novel interactive systems emerging from this research, potentially leading to spin-off companies and partnerships.

Les anthroponymes chez les Anicinabek et les cérémonies d’attribution des noms : regards sur la littérature et les archives

Cette recherche vise à explorer l’importance des noms et des cérémonies d’attribution de noms pour les peuples autochtones, particulièrement les Anicinabek. Le projet se concentre sur les noms comme éléments essentiels de l’identité et comme indicateurs de la vitalité de la langue anicinabe.
L’étude cherche à approfondir notre compréhension du système de dénomination traditionnel anicinabe et des cérémonies qui lui sont associées. Elle examinera également comment les survivantes et survivants des pensionnats autochtones dont les noms ont été changés durant leur enfance entreprennent des démarches pour se réapproprier et obtenir la reconnaissance légale de leurs noms d’origine.
Comme il existe peu de documentation scientifique sur ces sujets, ce projet exploratoire s’appuiera principalement sur des revues de littérature grise (documents non-académiques), d’articles scientifiques, de presse et de recherches en archives. Cette phase préparatoire servira de base à des recherches qualitatives futures.
Les résultats de cette recherche serviront à plusieurs fins concrètes, notamment à documenter le système dénominatif anicinabe et à contribuer à un documentaire produit par Minwashin, un organisme à but non lucratif anicinabe partenaire du projet. Ce documentaire suivra le parcours de survivantes et survivants des pensionnats autochtones qui cherchent à se réapproprier et à obtenir la reconnaissance légale de leurs noms d’origine. Un rapport de recherche complet sera également produit.

Linear array CMUTs for medical imaging applications

We propose the construction of novel ultrasound transducer structures based on existing MEMS technology that has been in development through collaboration between Micralyne and Prof. Roger Zemp. We will be exploring linear array transducers intended for medical imaging, and individual transducers for automotive use for ultrasound range finding. We will undertake the design, modelling, fabrication, packaging, and testing of the devices. The end goal is to produce commercially-viable transducers that may be used as replacements for the current generation of piezoelectric-based transducers. Micralyne would benefit by having a commercial offering at the end of the project. The Zemp lab will demonstrate array imaging with a novel architecture, and the work done will serve as a platform for further development including the use of novel materials and new array imaging schemes.

Mapping the space of motion diffusion models to optimize performance

This research project focuses on optimizing the efficiency and performance of motion diffusion models for real-time applications in video games. Diffusion models have shown great potential in producing high-quality and diverse human motion animations, but are often limited by their computational demands. This project will explore different model architectures and optimization techniques to reduce their resource consumption, making them faster and more memory-efficient without sacrificing animation quality. The project will aid CD Projekt Red in deploying optimized motion diffusion models for games in production, affirming its position as a technical innovator in the gaming industry while providing material benefits through the development of an evaluation pipeline for future model assessments.

Quantitative Investing Intern

Vestcor, Atlantic Canada’s largest in-house investment manager with $21 billion in assets, specializes in quantitative investing and risk management to optimize portfolio performance. One of the key challenges Vestcor faces is the continuous need for innovation and improvement in security selection and risk modeling. As financial markets become increasingly complex and data-driven, Vestcor must refine and enhance its quantitative models, statistical approaches, and risk assessment techniques to maintain a competitive edge. The company seeks to integrate new data sources, machine learning techniques, and advanced financial modeling methods to improve portfolio management strategies. To remain at the forefront of investment innovation, the Quantitative Investing Team requires empirical research and development of new techniques to improve decision-making, optimize risk-adjusted returns, and ensure long-term financial sustainability for its clients.

Agentic AI for Automated Essay Scoring

This research project aims to develop an AI-powered essay grading system that is both cost-effective and highly accurate. The project will explore how a dual-agent AI system, one that optimizes the cost and another that performs prompt refinement, can improve automated essay scoring at scale. By using advanced techniques such as Retrieval-Augmented Generation (RAG) and error-based prompt refinement, the system will ensure more precise and consistent grading. The partner organization, a leading educational service provider, will benefit from operational improvements, faster feedback for students, and improved accuracy in assessments. Additionally, the system’s open-source, on-premise design ensures student data security. Overall, this project will not only enhance the partner organization’s competitiveness but also strengthen its reputation as a leader in AI-driven education. By improving grading efficiency and educational outcomes, the project will contribute to the broader goal of making AI a valuable tool in modern education.

Predicting Engine Failure from Vehicle Telematics

(1) Main activities of the partner
Geotab is a global leader in telematics specializing in fleet management solutions to enhance operational efficiency, safety, and sustainability. For this project Geotab will be providing their vast collection of data from over 80,000 customers. Additionally, Geotab will provide the intern with support and structure within their data science and maintenance/ safety team.

(2) Challenges
Diagnostic trouble codes (DTC) and the corresponding warning lights are often the first indicator of a severe mechanical problem with a vehicle. Geotab’s rich data set of ongoing vehicle metrics may allow for the detection of these issues before they become severe enough
to trigger a DTC. However historical breakdowns are recorded in a raw and unstructured timeseries dataset. Drawing insights from this will require extensive data cleaning and modeling to successfully predict severe maintenance issues before their occurrence.

(3) Social or economic benefits
A successful implementation of a predictive model will enable Geotab to empower their customers to achieve even more efficient fleet operations through:
? Enhanced Fleet Efficiency: Increased proactive maintenance will reduce vehicle downtime and repair costs for fleet operators.
? Cost Savings: Improved predictive analytics can help fleet managers optimize maintenance schedules, leading to lower operational expenses.
? Environmental Benefits: Reduction in emissions through better engine performance monitoring and operations.

Research and Implementation of LLM based Autonomous Agent Based on IoT Big Data Environment

Geotab will provide the foundational platform and infrastructure, data, and expertise for the project. This includes access to their IoT big data environment on Google Cloud Platform, containing telematics data from 4.5M+ connected vehicles globally. They will offer mentorship and support through their AI Platform team, share historical data and documentation for training the large language model agent, and enable the intern to utilize various tools and resources to conduct analysis and construct solutions.
Geotab possesses a wealth of telematics data gathered from over 4.5 million devices globally, which presents untapped potential for AI-driven improvements in safety, sustainability, and operational efficiency for its customers. However, manual identification of critical events—such as safety incidents, customer dissatisfaction signals, or computational inefficiencies—is time-consuming, reactive, and prone to delays. The reliance on human intervention for analysis compromises real-time responsiveness and scalability, with suboptimal resource usage and consumption.
This project addresses the challenge of developing an autonomous Large Language Model (LLM) agent to provide actionable insights to both Geotab developers and customers, enabling rapid responses and proactive problem-solving.
The project solution enhances Geotab’s product quality, improves driver safety, reduces computation cost, and strengthens customer relationships.