Impact Report 2021-22

In the summer of 2022, Alloprof and Mila collaborated together, harnessing modern deep learning techniques, to improve Alloprof’s Help Zone, an online platform where students can ask a question about their school work and get an answer from another Alloprof user or moderator.

To keep students from getting stuck on a question and giving up on their homework, it’s important to give them help right away when they come to a stumbling block. Thanks to AI, students who ask a question on Alloprof now receive instant individualized support while they wait to receive assistance from a peer or Alloprof’s team.

In summer 2022, Mila’s AI Activation team has done a demonstration for Dubois Agrinovation to show the potential of optimizing their operations in selling parts for farm irrigation systems. Mila developed a computer-vision-based system that lets Dubois Agrinovation’s employees, or their clients directly, quickly identify the spare part needed by their customers using only a photograph of it.

Handling request for parts without this system had been a painstaking process for the company’s employees and their clients. This streamlined process, made possible by Mila's system, can be accomplished by less experienced employees, and even allows Dubois Agrinovation's customers to identify parts on their own, thus enhancing the company's customer experience.

The World Anti-Doping Agency (WADA) is joining forces with Mila researchers to explore AI-based solutions for improving athlete anti-doping testing strategies. The goal is to optimize testing resource allocation by better estimating an athlete’s doping risk. The resulting solution will be combined with other tools and strategies used by anti-doping organizations to prioritize testing. This collaboration can potentially have a meaningful impact in fighting the use of doping in sports.

With the help of Mila, LUCID built an affective music recommendation system (AMRS) that leverages patients’ self-assessed responses to music. AMRS proposes audio experiences (in the form of track sequences) for patients using a reinforcement learning agent and a model that predicts the average affective response of a generalized anxiety patient to music. The agent selects the music from a list of available tracks based on the listener’s current and final targeted moods. The first version of this technology led to a successful commercial API product for therapeutic platforms now integrated into a globally distributed and clinically validated research iOS/Android app.

This 5-year project, funded under the Natural Sciences and Engineering Research Council's Discovery Horizons pilot program, supports interdisciplinary research that spans health, the natural sciences, engineering and the social sciences. Yoshua Bengio and UdeM’s Michael Tyers are joining their efforts to identify new antimicrobial peptides using machine learning. The World Health Organization (WHO) considers antibiotic resistance one of the most serious threats to global health. This project, which seeks solutions to a worldwide problem, fosters dialogue between AI and health researchers in order to incorporate AI into drug discovery.

This IVADO PRF3-funded 30-month research framework program is led by Yoshua Bengio with three other principal investigators from the fields of health and materials chemistry. Nine Mila faculty members (Sarath Chandar Anbil Parthipan, Doina Precup, David Rolnick, Pierre-Luc Bacon, Jian Tang, Simon Lacoste-Julien, Irina Rish, Guy Wolf and Mathieu Blanchette) and a number of academic researchers from different universities are involved in the program’s interdisciplinary projects. These projects aim to design algorithms through a series of iterative rounds of data gathering (from laboratory experiments and data simulators) and algorithm development for applications such as drug and material discovery.

Under the bilateral Bavaria-Quebec Artificial Intelligence in Health collaborative research program, Professor Julien-Cohen Adad of Mila and Polytechnique Montreal and his collaborator Mark Mühlau at the University of Munich have received 3-year funding for a project on multiple sclerosis diagnosis using MRI quantification of brain and spinal cord lesions. While medications are available to control the disease, each person responds differently and an individualized approach would be much more effective. This would require effective analysis of MRI images. The research team hopes to solve the problem by means of AI algorithms. This approach has the potential to help determine treatment efficacy and to predict the clinical course of the disease, all while strengthening Quebec-Bavaria relations.

Following the November 2021 symposium hosted by Mila, IVADO and CIFAR, four projects were selected to receive nearly $400,000 in funding to explore AI methods to generate synthetic imaging data while ensuring fidelity and patient privacy.

Among the selected projects, Mila researcher Christian Gagné, Full Professor at Université Laval and Canada CIFAR AI Chair, will be leading a project in partnership with the Régie de l’assurance maladie du Québec (RAMQ) on the generation of confidentiality-preserving synthetic data from administrative databases on prescription drug consumption for the analysis of drug use in the Quebec population.

Professor David Rolnick is one of the lead authors of the report, “Climate Change and AI: Recommendations for Government Action,” commissioned by the GPAI and presented in November 2021 at COP26—the United Nations Climate Change Conference. The report includes 48 concrete recommendations for governments to address climate change, including a wide range of use cases.

As part of IVADO’s Strategic Research Funding Program, Mila professors Christopher Pal and David Rolnick, along with three other researchers from the Institut de recherche en biologie végétale and Université de Montréal, are leading a biodiversity and climate change research framework to provide researchers and practitioners with tools to combat biodiversity loss and the effects of climate change. Remote sensing technologies, such as phenocams or drones, offer the potential to study plant phenology and biodiversity in great detail, generating enormous amounts of data. The resulting data will enable the development of new algorithms grounded in the latest advances in computer vision and meta-learning to map plant species and their phenological signatures.