Innovative Projects Realized

Augmented Virtual Reality Interactive Training Program, with focus on older adults for Improving their Cognitive Function

During the current Covid-19 pandemic, more than ever, our seniors and those with dementia are isolated and at risk of a faster cognitive and mental health decline. This research offers a novel approach through an augmented virtual interactive social environment for cognitive training to prevent dementia and cognitive impairment. Built upon our successful previous research, our proposed program is an innovative virtual social environment with sessions of “brain exercises”. This allows older adults to attend a session through high-speed Internet using a locked-up PC laptop that automatically connects the user to the program upon pressing a “Start” button. This is particularly beneficial for the elderly who have difficulty commuting, adding to their depression and loneliness. Our proposed program will not only improve the participant’s cognitive abilities but also has great potential to improve their mental well-being.

Few-Shot Object Segmentation

Computer vision researchers have been moving beyond simple image classification and tackling more complex tasks such as object localization, detection and semantic segmentation. However, many of the proposed methods require large amounts of annotated data such as segmentation masks, which are expensive and time-consuming to acquire. Moreover, those methods cannot segment new object categories which were not present in the training set.

Few-shot segmentation alleviates both those problems by learning end-to-end to segment new object categories from few examples. In particular, weakly-supervised few-shot object segmentation only requires weak supervision such as sparse pixel annotations, bounding boxes and scribbles, which is substantially easier to gather than dense pixel annotations.

In this project, we propose a few-shot segmentation approach to alleviate the requirement of large strongly supervised datasets. Specifically, we propose a model which can learn how to segment new object categories using only a few annotated examples.

Developing an Efficient Ensemble Machine Learning Model for Evaluating Construction Project Bidding Quality and Optimal Winning Strategies

PledgX is interested in building a solution that aims to optimize the bidding process to maximize key performance indicators for contactors and vendors. For bidding optimization, several strategies and methods have been proposed; however, with the massive amount of available bidding datasets, the quality and performance of such methods are questionable. Machine learning introduces intelligent solutions to optimize the bidding decision, however these solutions are applicable to a range of prediction or classification tasks. Thus, ensemble modelling is introduced for efficient performance and to overcome drawbacks for individual modeling. In this project, we propose a novel data-driven bidding model based on ensemble predictive learning, which extracts sophisticated features and learns to bid automatically using the collected data. The model is composed of sub-models aggregated to form a more robust global model. The proposed ensemble learning model enables PledgX to learn complex rules of bidding with optimized overall bidding performance.

Using causal probabilistic fuzzy logic (PFL) rules integrated with Deep learning algorithms (DLs) to analyze Electroencephalography (EEGs)

Major Depression Disorder (MDD) is a big problem in our society. About 8% of Canadians may suffer from depressions in their life. Major depression can cause suicide and take families apart. Canadian governments spend more than $51 billion a year in the mental health sector. When treatment with medications fail, mental healthcare professionals, use Electroconvulsive Therapy (ECT) to treat patients with Major Depression Disorders (MDD). During an ECT session, electroencephalogram (EEG) signals let the mental healthcare professionals record patients’ brain activities which are helpful to decide whether the treatment was successful. However, there is no standard way to know how and with what intensity a healthcare professional needs to apply electroshock to treat patients with MDD. In this work, we will use non-classical logics such as probabilistic fuzzy logic and deep learning algorithms in order to find the ECT features resulting in successful ECTs.

Development of a noncontact PPG (ncPPG) system for oxygen saturation clinical analysis to increase the data reliability of SpO2 detection through consumer-level cameras

Sterasure Inc. in its current mission to provide biomedical tools to reinvent the clinical decision support, partners with the University of Waterloo to work towards the development of cutting-edge contactless vital sign monitoring systems. The technology studied in this research will allow the advance in the detection of oxygen saturation levels through cost-efficient systems. Although existing non-contact low-cost vital sign detection sensors offer great advantages for clinical environments, there is lack of studies showing medical reliability. Sterasure and the academic partner will prototype and design a reliable SpO2 contactless sensor which will be integrated in Sterasure’s biomedical device for future clinical studies.

Evaluating green roof performance and design in the Toronto area

Green roofs minimize stormwater runoff, building cooling costs, and provide other social, economic, and environmental benefits. Green roofs are also highly-proprietary, with the industry having many components to suite different applications, all influencing green roof survival and performance. With a green roof by-law and construction standard in Toronto, green roof coverage is consistently among the top cities in North America annually, however there exists no published data on green roof performance for the region, nor on the health and status of the City’s existing green roofs. The objectives of this project are to quantify green roof performance at the Green Roof Innovation Testing (GRIT) lab, and to survey and document Toronto’s existing green roofs, critically examining them using Toronto’s Best Practices Guidelines. The collaborative project will result in improved design and planning of green roofs by STLAi, creation of interactive material for referencing Toronto’s green roofs online, and peer-reviewed publication.

Integrated Logistics Network Design at Hydro-Québec

This partnership project with Hydro-Québec focuses on optimizing the integrated logistics network planning. The Hydro-Québec existing network has a building stock of average 37 years old, and is mainly based on the geographic and demographic needs of the 1980s. This project will adapt the network to the modern needs and prepare it for the future. The problem we consider includes the location of service points, determination of electric vehicle sizes and routes, demand forecasting, and inventory management. The project offers strong practical value for Hydro-Québec and a strong academic potential. We leverage the abundant data from Hydro-Québec to study the uncertainty profiles of customer demand and perform data aggregation to scale down the problem size by considering geographic and demographic information of customer regions. We use prescriptive and predictive analytics methods that span advanced optimization and data science techniques.

Improving the collection performance and reducing the complexity of cyclone systems using a rotary classifier in a dynamic cyclone particle separator

Global pollution emissions contribute to climate change and are damaging to health. In many industrial applications that produce particulate matter, devices such as cyclones are used to separate and capture the particles from the exhaust gas. However, these do not capture the very small, but hazardous, particles and so expensive and energy-intensive secondary systems have to be added to the process. The industrial partner is developing a novel dynamic cyclone separator that has rotating vanes which improve the particle separation efficiency and allow capture of the fine particles. This project will develop a computer model of the flow and particle motions in the cyclone that can then be used to design these dynamic cyclones for different industrial applications and operating conditions, thus leading to a more efficient reduction of particulate pollution.

Data analytics on city 311 information requests

In the current era of big data, huge volumes of a wide variety of data are generated and collected at a rapid rate. Embedded in these big data is implicit, previously unknown and potentially useful knowledge and information. This calls for data science—which use techniques like data mining, machine learning, etc.—for social good. With popularity of the initiates of open data, more data are made openly accessible to citizens. An example of these open big data is data collected at the 311 contract centre in the City of Winnipeg for the 311 information requests. In this research, we analyze and mine this dataset to find characteristics associated with the callers and the information requests. Knowledge on these characteristics helps users (e.g., decision makers at the City) to get a better understanding of the requests (e.g., why residents request information through 311 instead of other online options). In a longer term, the discovered knowledge and the understanding of the data helps improve the 311 and other online services. Along this direction, this research will add to the growing body of knowledge for all Canadian cities and communities.

Targeting granzyme B with a novel inhibitor for the treatment of radiodermatitis

Radiodermatitis is a group of skin reactions that occur as a result of radiation therapy. It is a significant health challenge as approximately 70% of all cancer patients receive radiation therapy and approximately 95% of them experience radiodermatitis. Patients with radiodermatitis experience redness, itchiness, pain, scaling, and weeping or crusted wounds. Importantly, radiodermatitis can impede cancer treatments. Current treatments for radiodermatitis have shown limited efficacy; thus, improving our understanding of radiodermatitis and developing novel therapies are urgent needs. In our preliminary studies, we have found that the protein Granzyme B is present at very high levels in radiodermatitis skin. We also showed that Granzyme B damages components of the skin and promotes inflammation in radiodermatitis. Our proposed project will test if Granzyme B induces the symptoms of radiodermatitis and if our newly developed medication that stops Granzyme B activity can alleviate these symptoms. Findings from this study will provide further rationale to pursue inhibitors of Granzyme B as a novel treatment option for radiodermatitis.

Governing innovation and knowledge sharing to increase adaptive capacity of forest industry in Quesnel, British Columbia

In Quesnel, British Columbia, efforts to stimulate innovation and diversify actors of local forest industry can be challenging for emerging locally-driven small forest enterprises that have limited capacity and resources. Building an inclusive forest industry that constitutes existing large forest enterprises and locally-controlled small forest enterprises in the landscapes therefore involves strengthening networks of knowledge and innovation. This research seeks to gain a deep understanding of Quesnel forest industries and to identify conditions that foster learning and innovation among locally-driven forest enterprises. To meet the objective, we will apply a participatory approach to learn about the underpinning social network and points of leverage to achieve resilient forest industry. Community-driven study using participatory approach will help the City of Quesnel to establish a long-term collaboration with private institutions and local communities to improve the effectiveness of Forestry Initiatives Program

Label-free Multiphoton Microscopy Imaging for Guiding the Surgery of Skin Basal Cell Carcinomas

Basal cell carcinoma (BCC) is the most common cancer type. Although it can be surgically removed, to confirm the clean removal by histology is time-consuming, which complicates the treatment and results in many incomplete removals. We propose to develop a special microscopy imaging platform that can image the skin tissue directly without sectioning and staining. This will enable detection of residual tumor cells by examining the excised fresh tissue samples on site during the surgery, providing immediate guidance for improving the treatment procedures. It will make the surgery fast, reduce patient pain and anxiety, and also help reduce cancer recurrence. These will enable the hospital to provide improved care for skin cancer patients. And it will also bring significant cost savings to the hospital and the Canadian healthcare system.