Interactive Reinforcement Learning Speedup with Confidence-based Transfer Learning

Reinforcement learning (RL) is a type of machine learning that focuses on allowing a physical or virtual agent to complete sequential decision-making tasks, such as video games. It has had many successes, but can be slow in practice, requiring large amounts of data. This project aims to speed up such learning problems by leveraging information from an existing agent. This existing agent need not be perfect – the algorithm developed will leverage information from the existing agent whenever possible and learn to outperform it where it is suboptimal.

Efficient face recognition for wearable camera devices

Titan Sécurité Inc. has deployed wearable video camera devices for security and surveillance applications, and seeks to accurately detect and recognize objects appearing in captured videos. This project focuses on video-based face recognition (FR), where facial trajectories captured with video cameras are compare against one (or few) reference stills for each individual of interest. The performance of these FR systems is typically poor due to complex and changing video surveillance environments, e.g., variations of facial appearance due to pose, illumination, blur, etc.

Efficient design and implementation of concatenated error-correction coding for high-throughput fiber-optic links

The project targets design and implementation of error-correction codes for high-throughput fiber-optic communication links. We focus on the error-correction encoding at the transmitter side as well as decoding at the receiver side considering the simplicity of implementation and low power consumption at both transmitter and receiver.

Policy Optimization in Parameter Space

Model-free Reinforcement Learning (RL) has recently demonstrated its great potential in solving difficult intelligent tasks. However, developing a successful RL model requires an extensive model tuning and tremendous training samples. Theoretical analysis of these RL methods, more specifically policy optimization methods, only stay in a simple setting where the learning happens in the policy space. This project attempts to advance the analysis of the policy optimization methods to a more realistic setting in the parameter space.

Data Driven Assessment for Self-Determined Employment

MatchWork enables non-profit employment support organizations to support marginalized people to find meaningful employment opportunities. This includes people with physical and mental challenges, veterans, new immigrants and refugees.

Experimental and numerical investigation on force-based and performance-based seismic design of bridges

Bridge infrastructure constitutes a substantial portion of national wealth of Canada, whose performance during earthquake events has a significant impact on the public safety. This study focuses on investigating the force-based and performance-based seismic design of bridges specified in the latest version of Canadian Highway Bridge Design Code 2014. Both experimental and numerical studies will be conducted, and design guidelines will be recommended.

Design Analytics

Architectural design data have mostly been limited to visual representations, specifications and contractual documents. Today, design firms generate vastly more and diverse data, but lack adequate access to tools to gain insight from such data. Yet the field of “visual analytics” provides concepts and systems exactly for working with such data. The proposed research aims at the visualization and analytics of design data, that is, collections of designs, their alternatives, project documentation, and other data collected from buildings and their settings.

Generalization in Deep Learning

In recent years, deep learning has led to unprecedented advances in a wide range of applications including natural language processing, reinforcement learning, and speech recognition. Despite the abundance of empirical evidence highlighting the success of neural networks, the theoretical properties of deep learning remain poorly understood and have been a subject of active investigation. One foundational aspect of deep learning that has garnered great intrigue in recent years is the generalization behavior of neural networks, that is, the ability of a neural network to perform on unseen data.

Investigating Driving Condition Impact on Millimeter-wave (77GHz) Automotive Radar for Autonomous Vehicles

To enable the development of self-driving vehicles, an accurate characterization of automotive radar modules under various road or weather conditions is required to ensure reliability is maintained under all circumstances. With this fundamental building-block established, ACAMP will be able to support Canadian technology companies in the development of autonomous vehicles.

Ahead of Time Compiled Code Generation

Compilers are large software projects consisting of many separate but common components like code generators, garbage collectors, and runtime diagnostic tools, to name but a few. Historically compiler developers have had to write each of these components from scratch. The Eclipse OMR project was created to provide generic components for use in new compilers and language runtime environments. OMR has enough flexibility to accommodate a wide range of programming languages without sacrificing performance, portability, or robustness.