Transforming the reality of COVID-19 pandemic into photorealistic virtual reality for immersive training of health-care professionals

In the context of pandemics like COVID-19, the value of multidisciplinary skills for health-care professionals becomes more evident. Training these professionals to take on new roles and responsibilities can be time-consuming, costly, resource-demanding, and most importantly, risky. Photorealistic transformation of training environments into virtual reality (VR) can be performed using the advanced techniques of computer vision, photogrammetry, and computer graphics.

Advancing traceability in informal supply chains through applied AI and ML

PemPem develops tools to ensure product traceability in informal supply chains using AI. These informal supply chains employ over 60% of the working population worldwide. They typically have highly inefficient operations due to very limited access to information and a reliance on opaque word-of-mouth coordination. While PemPem has started solving the problem of collecting data on commercial transactions and activity in these informal supply chains and of improving their efficiency, it now has to convert this data into actionable traceability insights.

Koala Pro – Dossier intelligent

Conception et développement de 3 plateformes intégrées pour commercialisation dans l’intérêt d’augmenter la productivité des nutritionnistes au travers d’une prise de données accélérée, une analyse de données améliorant la précision de la note au dossier, une aide au diagnostic alimentaire pour améliorer l’intervention client-patient, et la création de statistiques professionnelles liées aux résultats des interventions.
Le développement d’une première plateforme mobile permet la récolte de donnés patient-client hors consultation, une deuxième plateforme de note au dossier permet l'agrégation d

Contextual portrait detection

A frequently occurring problem in face verification in Jumio is that stock face detectors find multiple faces in the input image. The decision which one should be selected for the face verification step is not clear. Common reasons being, users submitting a single image with both selfie and document id. There can be other people, paintings, posters or television screens in the background.
Users may upload the selfie and the document picture in reverse order, etc.

Leveraging Deep Learning in Asset Pricing in a Multi-Factor Modelling Framework

Providing relevant quantitative trading strategies requires obtaining financial data from multiple sources to obtain market information and then use this data to model outcome. One difficulty in this process is that data entry is done by financial analysts who spend a large portion of their activities entering data from PDF to an application. This project seek to improve data collection in Canada by automating the process and focus analysts on their core competencies.

Deep learning approaches for semantic textual similarity on low-resource languages and specialized domains

The aim of this research is to investigate from traditional methods to deep learning methods, how to measure the meaning relationship between two sentences, by combining the local context, at word-level, and the global context at the sentence-level, and their ability to model informativeness and diversity of meanings expressed in natural language, i.e. in English or in French.
Moreover, as we are interested in Information Extraction of entities, concepts, triplet and semantic relation in unstructured text, we will adapt the BERT model for low resource domains and languages.

Visual-haptic Representation for Zero-shot Learning

Humans recognise objects in the world leveraging multi-modal sensory inputs beyond visual aspects (images and videos). Touch based information (Haptics) possesses rich information about structure, shape and other objetness properties. In this work, we will study and learn cross-modal representations between vision and touch. To connect vision and touch, we plan to introduce a zero shot classification task of recognising unseen object categories from shapenet dataset using haptics signals.

Monitoring of turbine runner blade strains from indirect measurements using AI

Hydro-Québec has data acquisition systems for a multitude of sensors, some of which have been installed since almost 20 years in its electrical generation equipment (turbine-generator units - TGU). The collected data is primarily used to ensure that the information is adequate in the event of an equipment breakdown or for specific behavioral studies.

Short term Electrical Load Forecasting

Load forecasting is an essential activity for a company like Hydro-Québec. It is necessary for objectives as varied as the management of production or the management and maintenance of the electricity network. Any significant forecasting error can result in reliability issues, loss of opportunity, or additional costs to the business. On the other hand, a good prediction would allow Hydro-Québec to generate additional sales in neighbouring markets. With the deployment of its Advanced Measurement Infrastructure (AMI), Hydro-Québec now has a significant amount of new consumption data.

Solar Radiation Forecasting

The main duty of Hydro-Quebec is to respond efficiently to the energy demands of customers, in a safe way while remaining competitive in the markets as well. In a changing energy context, the production of solar photovoltaic energy represents a new challenge for Hydro-Quebec, which will have to integrate and to balance this intermittent resource to guarantee the reliability of the electricity grid.