Innovative Projects Realized

Efficient algorithms and software for eye tracking on an embedded platform such as a smartphone

The goal of this project can be divided into three subobjectives. First, we need to propose, implement and train an accurate eye tracking model on the server, then migrate it to an embedded platform with a simple application that can run the model. Finally, we need to experiment different pruning methods for the network and possibly explore new approaches in order to improve energy efficiency while preserving other performance metrics of the model such as frame per second and accuracy. The focus of the project will be the third subobjective. Throughout the project, various network pruning methods will be explored and incorporated into the model. Some existing approaches are found in literature. This includes energy-aware pruning and layer-by-layer pruning. TO BE CONT’D

Optimizing Gastric Banding Surgery Outcomes in Obese Patients

The negative effects of obesity on quality of life, daily functioning and overall health are well documented. Specifically, obesity is associated with increased morbidity and mortality. Despite this knowledge, the incidence of obesity continues to increase exponentially leading some, generally extreme cases, to seek out laparascopic adjustable gastric banding (LAGB), as a means to control life-long obesity. Although LAGB can be successful, its long-term success is dependent on collaborative, individualized behavioural interventions. This research, a unique collaboration between academics and industry, seeks to develop a multi-factorial assessment/screening tool that may better predict and optimize success of the LAGB procedure.

Tightly-coupled Visual-Inertial-LiDAR SLAM

Since Amazon robotics expanded the use of drones to package deliveries to customers, drone applications have been expanded to many industries along with its ability to perform various tasks autonomously. The fundamental technology of drones’ autonomy comes from perceiving its surrounding, creating its own map based on onboard sensors and estimate its location within the map. This technology, also known as Simultaneous Localization and Mapping (SLAM), has been on the rise especially in mining and construction industries for surveying and mapping the site more efficiently; thus, many research works have been performed to improve robot’s SLAM technology. Although various sensor suites have been researched to improve SLAM performance, this project focuses on the novel contribution of developing a robust and accurate 3D SLAM by jointly optimizing stereo cameras, IMU and LiDAR measurements. TO BE CONT’D

Privacy Guarantees and Risk Identification: Statistical Framework and Methodology

A risk-based approach to anonymization includes an assessment of the risk that an attack to reveal or uncover personal information will be realized, known as threat modelling, against the risk that an attack on the data will be successful (e.g., a re-identification). We wish to incorporate the provable guarantees of differential privacy into this assessment of risk, to produce safe data in context of the environment in which it will be used. We also need adapt the methods of statistical disclosure control to such an updated approach.

Design of the next-generation of content-based, context-aware product recommender systems

We are in the process of creating and growing a team of researchers expert in the field of machine learning and data-mining. Ultimately, our aim is to create solutions to eliminate the need to manually define personalization strategies. We are working with more than 1000 retail locations across North America and collecting large-scale datasets of customer behaviour. Through a data-sharing/consulting partnership we plan to perform research on the design of recommender systems and predictive models customized for the datasets available to retailers. These methods can be used in their physical and online marketing programs as well as in their dynamical promotions/pricing strategies.

Investigation of the effects of whole hemp seed dietary supplementation on the microbiome-endocannabinoidome axis and its implications in diet induced obesity

Nowadays, the balance of omega-3 and omega-6 fatty acids (FAs) has shifted in favour of the latter and fiber consumption has decreased, both of which are associated with poor cardiometabolic health. Omega 3 FAs and fiber may impart their health benefits by modulating the endocannabinoid system (ECS) and the gut microbiome, both of which are key regulators of cardiometabolic health and obesity. Whole hemp seeds possess excellent nutritional value; rich in omega-3 FAs, fibers, proteins and vitamins and minerals. Therefore, dietary hempseed may be able to improve cardio-metabolic health by modifying the gut microbiome and ECS. Nature’s Decision is a Canadian hemp producer that pays particular attention to the quality of their hempseeds, and are keen on understanding on their potential cardiometabolic health benefits. TO BE CONT’D

Advanced Analytics in Multiple Sclerosis Research

The multiple sclerosis (MS) clinic at St. Michael’s Hospital (SMH) is among the largest in the world. While considerable data is collected from the MS clinic in both structured and unstructured form, the ability to glean this information to assess quality of care and conduct advanced analytics such as predictive modeling is limited. In this project, a quality improvement dashboard will be developed based on automation of clinical information extraction process. Predictive models will be used on existing clinical data to optimize treatment strategies and predict patient outcomes such as relapse rates, disability progression, and treatment failure. These models could then be used in clinical practice to identify high risk patients in a timely manner for appropriate follow-up and treatment optimization.

Intelligent Character Recognition (ICR), Optical Character Recognition (OCR) and machine learning based corrections of data transcription from scanned business documents

SS&C processes more than 80% of financial scanned and faxed documents in the US and requires large amount of manual labor in order to map information from a document into another form. Advances in neural networks applied to computer vision have produced text detection and recognition that nears human performance. This project will be leveraging these approaches to address the main challenge of applying image segmentation and character recognition techniques to large volumes of documents, namely the sensitivity of the process to phenomena like the variability of text, document formats and imaging conditions. The expected benefits of the project to the industrial partner are (i) reduction of human error in the document workflow, and (ii) faster turn-around times for customers of SS&C.

Detecting Company-Specific Purchase Evidence from Twitter Posts

Delphia’s business model revolves around generating insights for investing firms that allow them to make better trading decisions. It has been shown that detecting when Twitter users post about recent or future purchases has the potential to increase the accuracy of company sales forecasts, which in turn can inform stock trading strategies. This internship project aims to develop automated means to detect and quantify purchase related posts on Twitter. The intern will conduct a machine learning project which will involve creating a dataset of purchase related tweets and using it to train a purchase tweet detector. Success in this project could lead to new data products for Delphia to sell to its financial investor clients.

Scaling simulations in population health via machine learning

Computer simulations provide a safe alternative to taking a trial-and-error approach in the real-world. If a simulating intervention is found to be inefficient or even harmful, then it can be canceled without causing harm to real individuals. Consequently, simulations are increasingly sought after for complex social problems such as homelessness and the spread of the Human Immunodeficiency Virus (HIV). However, such complex problems can be tackled using many different interventions (e.g., increasing shelters for homelessness), each being defined by several parameters (e.g., number of beds). Simulating all possible interventions and their parameter values is prohibitive; there are too many combinations to simulate, and each simulation can take a long time to complete when the computer model is highly detailed. TO BE CONT’D

Enhancing Security and Quality of Service in NFC-based Smartphone Applications

The primary objective of this MITACS Cluster project is to investigate, design and prototype novel

techniques for the integration of security and quality of service in near field communication (NFC)-

based smartphone applications. Universal NFC Cloud Connect Inc., a new start-up company based in

Halifax, Nova Scotia, that ties mobile devices to location-specific events via the Cloud, thus enabling

anyone with an NFC-based smartphone to be identified at a specific location at a specific time.

However, two important challenges that need to be addressed before the product can be successfully

deployed are security and quality of service (QoS). The project will explore techniques that will

integrate existing security and QoS protocols with novel identity-based encryption and QoS

enhancement algorithms on smartphone applications. The Cluster project will be a crucial part in the

partner organization’s road to commercialization of the product. This can open up enormous

application possibilities in a variety of consumer application areas, such as the hospitality industry…………..TBC

Visualizing Unspoken Words

In order to a facilitate a productive dialogue about contemporary Canadian race, ethnicity and religious rights, my research project entails the following three components:
1- an interactive artistic installation;
2- data collection from the target audience in response to the installation using sensor-based software application;
3- public visualisation of such data in the form of computer-assisted art in order to present the results back to the community for further discussion.
This three-tier project further develops the existing interactive art models as it creates a clear one-on-one experience, proposes an anonymous and seamless method of data collection and utilizes artistic data visualization methodology.
The initial artwork consists of a mirror reflecting visual and textual elements concerning race, ethnicity and religious discrimination which then the audience respond to using the interactive controls that appear on the mirror. TO BE CONT’D