Projets novateurs réalisés

Optimizing Deep Learning Models for Edge Devices in Threat Detection for Computer Vision Applications in Smart Cities and Retail

During the internship, the selected candidate will focus on developing edge computing solutions that can recognize and alert the relevant personnel in real-time in case of potential security threats (e.g. theft, robbery) and safety issues (e.g. employee accidental falls). This would help retailers to prevent or respond quickly to incidents, reducing losses and improving safety for customers and employees. The internship will have a set of milestones which will involve developing the key building blocks for such an application, including exploring algorithms and techniques for object detection, motion detection, face recognition, and activity recognition. The models will be trained on high-performance computing resources and tested on edge compute devices. The intern will also be responsible for real-time testing of the developed edge computing solution, executing proof of concepts, and deploying the AI models on the edge computing hardware platform such as cameras.
The internship will provide J-Squared Technologies with valuable research that can be used to continue building optimized edge-computing solutions. Additionally, the project will help J-Squared Technologies establish an AI/ML research team that can work on the latest state-of-the-art artificial intelligence techniques to create solutions that benefit society.

Supporting Virica’s Inorganic Growth Objectives

“Supporting Virica’s Inorganic Growth Objectives” is primarily a strategy project which will lead to actionable insights to promote Virica’s goal of expanding our VSE library through inorganic methods. Initially, the project will entail research into the current viral sensitizer space, with the goal being to identify Virica’s competitors and other companies that hold IP on viral sensitizers with applications in the manufacture of viral medicines. The intern will develop an excel database containing relevant information regarding specific molecules and programs which may complement Virica’s current platform, with special attention towards molecules with mechanisms of action unexplored by Virica’s current platform, or mechanisms of action which will complement Virica’s current molecules. Once the database is complete, the intern will evaluate each molecule in terms of how well it will fit into Virica’s current library, and develop a triaging system to identify the top candidates for in-licensing. The intern will then develop a strategy regarding the structure of deals regarding the purchase and/or in-licensing of top candidates and will begin outreach and negotiations with the IP holders of candidates with the goal of adding such molecules to Virica’s library.

Strategic Marketing and Volunteer Onboarding Strategy Yorkton Brick Flour Mill Historical Society

An extension of an earlier Mitacs project, this project will complete some uncompleted objectives in the last internship and complete some new objectives including a strategic communications plan for the Historic Brick Flour Mill in Yorkton.

Representation Learning with Time Series Data

The proposed research aims at learning better representations for multivariate time series (MTS) data, which can be applied to various important real-life applications such as weather, traffic, and electricity forecasting. Better forecasting accuracies for these tasks could help with efficient risk aversion and decision making, and save costs for decision makers. The proposed research will look into current techniques for MTS data modeling and will be focusing on filling gaps in existing research to further improve forecasting results. The proposed research will explore novel modeling approaches that can effectively capture the important and distinct characteristics of MTS data, and will develop approaches that can transfer between MTS data in different domains.

The Fertility Partners 2023

The Fertility Partners is the business partner of choice for distinguished IVF and prenatal care providers, working together to identify and institute best practices so they can deliver fulfilling outcomes and exceptional experiences to patients and their families. Since the platform started less than 3 years ago projects for this Mitacs internship are assisting in building and improving processes so the corporate team can offer better services to the clinics who can then offer better care to its patients.

Item Identification for Robotic Pick and Place Applications

This research project aims to develop a robot pick and place model that can be used in Kindred AI’s robotic arms to improve efficiency and reduce production costs. The intern will work closely with the partner organization’s experts in computer vision and MLOp to design and build new models, modify existing ones, and experiment with them on robotic arms. The project will benefit Kindred AI by enhancing its product offerings and increasing their competitiveness in the market. Additionally, this research will contribute to the advancement of robotics technology in Canada, which can potentially have significant implications for various industries.

Research and Implementation of Streaming Data Analytics Based on IoT Big Data Environment for Safer Fleets and Smart Cities

Road safety affects everyone, not just Geotab customers. With several years of driving and environmental data from over 2 million connected vehicles we have an opportunity to make our customers safer, as well as our communities and cities.
In an effort to reduce accidents, we need to understand both the driving behavioural patterns that are predictive of accidents, and the environmental factors involved. To achieve this, the data infrastructure should be capable of processing a series type of real-time video and telematics data, as well as time-series historical records generated from existing machine learning models to respond to real world incidents within a short period of latency.

Towards making graphics accessible to blind people

There has been a lot of effort in making printed media accessible to low vision or blind individuals. Braille has been extensively utilized to make text accessible to the blind. Software that automatically converts text to speech has also been employed for this. However, the existing solutions are not adequate for conveying graphical information to low vision or blind individuals. Common practice is based on manually converting images to tactile. A tactile is a representation of an image that is accessible by touch. Manual conversion by tactile designers is a tedious and expensive process. This project aims to develop Artificial Intelligence models for converting graphics to tactile format. More specifically, this project aims to develop methods that can generalize to new categories of images.

Optimization of Interventions to Reduce Combined Sewer Overflows

The dissertation title is “Optimization of Interventions to Reduce Combined Sewer Overflows (CSOs) (Quebec City as a Case Study)”. For cities where combined sewer systems are designed to transport both wastewater and stormwater flows in a single pipes network, during wet weather conditions, the volume of runoff may exceed the system capacity which results in the discharge of untreated or only partially treated water to the nearest outfall, this is known as combined sewer overflows (CSOs).The goal is to propose a practical solution to mitigate stormwater overflows and CSOs cities which are currently critical concerns for the existing infrastructures. To achieve this aim, integration of traditional and most recent techniques solutions will be assessed to better cope with CSOs’ quality and quantity adverse impacts on the environment at a minimum cost. The outcome will make a noticeable contribution in providing a framework for resiliency and infrastructure sustainability in big cities.

Multi-material 3D printing with FFF and SLA

We propose a solution to achieve a strong interfacial connection between FFF-printed materials and SLA-printed materials by utilizing an ester exchange reaction. Common FFF materials such as PETG and PLA contain an ester structure that can undergo ester exchange reactions with other alcohols in the presence of a strong base such as Triazabicyclodecene (TBD) and alter the original cross-linked structure. We can leverage this property by incorporating monomers containing hydroxyl groups into the resin of SLA and adding a certain concentration of alkali.

Our goal is to develop a multi-material printing technology that combines FFF and SLA methods, which we anticipate will be useful for the production of multi-material soft robots, as well as other applications, including the manufacturing of microfluidic chips using engineering materials. By achieving a strong interfacial bond between FFF-printed and SLA-printed materials, we aim to enable the production of multi-material parts with varying mechanical and physical properties. This technology has the potential to revolutionize the field of soft robotics, where the ability to combine materials with different elasticity and hardness can lead to the creation of complex and functional robotic systems.

Object Tracking for High-Speed Pick-and-Place Robot

The demand for eCommerce and online orders has risen rapidly in recent years, this drives the need for highly efficient and automated item sortation systems. Kindred AI is a technology company with the objective to bring artificial intelligence and robotic technologies into the workforce of eCommerce, parcel and order fulfillment. As a part of the effort to achieve accurate and robust robot grasping and placing tasks in such workforces, this research project aims to leverage modern machine learning and computer vision algorithms to enable safe, accurate, and fast item handling operations. In particular, the proposed research project will focus on designing and prototyping an intelligent and efficient item tracking system capable of predicting an item’s location in the near future in a robotic manipulation environment. Successes in this research will bring advancement in the partner organization’s robotic product and improve the safety and productivity in the human-robot workforce.

Machine learning-driven molecular classification of pediatric brain tumors

Brain tumors are the leading cause of cancer-related death in childhood and are generally categorized as low-grade or high-grade. More granularly however, there can be over 100 types of brain tumors which can vary widely in both prognosis and treatment. Machine learning has increasingly been applied to classify brain tumors and other cancer types but despite the success of these algorithms in classifying adult brain tumors, performance for pediatric tumors has been suboptimal in part due to the lack of sufficient training data. With larger sample sizes now available and extensive archives of in-house pediatric molecular and clinical data, the aim is to develop methods to better classify and diagnose these pediatric brain tumors, utilizing new computational techniques to analyze and integrate diverse biological datasets. By having a better understanding of what drives these tumors and more refined classification frameworks, improvements can be made to the process of matching patients with the most effective therapeutic options.