## Through our members and academic partners, our community brings together more than 1400 scientists working in the field of artificial intelligence.

For the duration of the Canada First Research Excellence funding, we support the hiring of a minimum of 25 professors divided between the Université de Montréal, HEC Montréal and Polytechnique Montréal.

### Aishwarya Agrawal

Assistant Professor

Université de Montréal

### David Ardia

Associate Professor

HEC Montréal

### Okan Arslan

Assistant Professor

HEC Montréal

### Pierre-Luc Bacon

Assistant Professor

Université de Montréal

### Elie Bou Assi

Assistant Professor

Université de Montréal

### Julie Carreau

Assistant Professor

Polytechnique Montréal

### Margarida Carvalho

Assistant Professor

Université de Montréal

### Karim Chalak

Associate Professor

Université de Montréal

### Dr Michaël Chassé

Clinical Assistant Professor

Université de Montréal

### Moncef Chioua

Assistant Professor

Polytechnique Montréal

### Francesco Ciari

Assistant Professor

Polytechnique Montréal

### Martin Cousineau

Assistant Professor

HEC Montréal

### Philippe Doyon-Poulin

Assistant Professor

Polytechnique Montréal

### Guillaume Dumas

Assistant Professor

Université de Montréal

### Sarah Gagliano Taliun

Assistant Professor

Université de Montréal

### Gauthier Gidel

Assistant Professor

Université de Montréal

### Julie Hussin

Assistant Professor

Université de Montréal

### Güneş Karabulut-Kurt

Associate Professor

Polytechnique Montréal

### François Leduc-Primeau

Assistant Professor

Polytechnique Montréal

### Antoine Legrain

Assistant Professor

Polytechnique Montréal

### Sébastien Lemieux

Assistant Professor

Université de Montréal

### Antoine Lesage-Landry

Assistant Professor

Polytechnique Montréal

HEC Montréal

### Jorge Mendoza

Associate Professor

HEC Montréal

### Ioannis Mitliagkas

Assistant Professor

Université de Montréal

### Sean Molesky

Assistant Professor

Polytechnique Montréal

### Oury Monchi

Full Professor

Université de Montréal

### Eilif Benjamin Muller

Assistant Professor

Université de Montréal

### Tomas Paus

Professor

Université de Montréal

### Jean Provost

Associate Professor

Polytechnique Montréal

### Juliana Schulz

Assistant Professor

HEC Montréal

### Zohreh Sharafi Tafreshi

Assistant Professor

Polytechnique Montréal

### Jian Tang

Assistant Professor

HEC Montréal

### Pablo Valdes-Donoso

Assistant Professor

Université de Montréal

### Wietske Van Osch

Associate Professor

HEC Montréal

### Guy Wolf

Associate Professor

Université de Montréal

### Amal Zouaq

Associate Professor

Polytechnique Montréal

More than 90 researchers associated with our academic members and/or having received CFREF funding hold a research chair, of which 40 have been obtained since our creation. These three FRQ-IVADO chairs promote EDI in digital intelligence.

### Margarida Carvalho

Assistant Professor

Université de Montréal

### Foutse Khomh

Full professor

Polytechnique Montréal

Full professor

HEC Montréal

## Student Intersectoral Committee

### Dorothy Armand-Lima

MSc student in E-commerce

### Antoine Boudreau LeBlanc

PhD student in Bioethics

### Ève Campeau-Poirier

MSc student in Astrophysics

### Simon Faghel-Soubeyrand

PhD student in Cognitive neurosciences

### Patricia Gautrin

PhD student in Philosophy

### Junior Momo Ziazet

PhD student in Computer science

### Sacha Morin

PhD student in Machine learning

### Brice Rauby

PhD student in biomedical engineering

### Camille Rondeau Saint-Jean

MSc student in biological sciences

## Laureates of our different programs

• 2022
• 2021
• 2020
• 2019
• 2017
• 2018

### Strategic Research Funding Program

Topic 1 – Integrated Machine Learning and Optimization for Decision Making under Uncertainty: Towards Robust and Sustainable Supply Chains

Nearly all decision problems involve some form of uncertainty. This is especially true in supply chains where, e.g., demand, cost, capacity, and travel time’s high variability considerably complicate the planning of procurement, production, distribution, and service activities. Due to constantly evolving environments and the high frequency of data acquisition, classical decision-making that is based on training models, validating them, to finally optimize decisions does not suffice anymore. This research program aims at developing new methods for making the most effective and adaptive use of data in decision-making. It is founded on modern optimization and machine learning perspectives that encompasses developments in deep reinforcement/end-to-end learning, risk averse decision theory, and contextual/distributionally robust optimization. Its mission is three-fold: (i) develop the next generation of methods to deal with uncertainty in data-driven risk-aware optimization models by integrating machine learning; (ii) identify scientifically challenging and high-impact opportunities for improving robustness in supply chains; and finally (iii) stimulate the integration of stochastic optimization models among our partners while defining use cases that will guide future methodological advances. Overall, this program envisions a virtuous cycle of scientific discoveries that are both fueled by and transformative for an important sector of the Canadian economy.

Topic 2 – AI, Biodiversity and Climate Change

Lead researchers: Etienne Laliberté (Université de Montréal, IRBV), Christopher Pal (Polytechnique Montréal, Mila), David Rolnick (Université McGill, Mila), Oliver Sonnentag (Université de Montréal), Anne Bruneau (Université de Montréal, IRBV)

Climate change is altering plant biodiversity, with potentially catastrophic consequences on the resilience and functioning of terrestrial ecosystems. A major source of uncertainty in the global terrestrial carbon budget, and thus for future climate projections, is how plant species differ in their phenological responses to seasonal climate fluctuations. In addition, climate change reshuffles plant species distributions across entire landscapes, but we are unable to keep track of those changes in biodiversity using classical field-based sampling. Remote sensing technologies such as phenocams or drones offer potential to study plant phenology and biodiversity in great detail across spatial scales. These new approaches could revolutionise biodiversity science and conservation, and help guide the design of nature-based solutions essential to mitigate the effects of  climate change. New AI algorithms are needed to unlock the full potential of this transformative technology and its links to more traditional data streams and products. This program will develop these new algorithms, building on the most recent developments in computer vision and meta-learning to map plant species and their phenological signatures. Algorithms will be put directly into the hands of scientific and non-scientific end-users via the development of an active learning platform. This AI research will empower researchers and practitioners to turn imagery into actionable data about plant biodiversity and phenology, providing them with tools to help fight biodiversity loss and the effects of climate change.

Topic 3 – Human health and secondary use of data

Lead researchers: Michaël Chassé (Université de Montréal, CRCHUM), Nadia Lahrichi (Polytechnique Montréal, CIRRELT), An Tang (Université de Montréal, CRCHUM)

Artificial intelligence (AI) technologies hold the potential to transform healthcare. These technologies are emergent in logistics and imaging, and hundreds of algorithms are now being developed to help support care delivery. Many challenges remain, however, when it comes to scale-up for use in the field. One such challenge is ensuring the generalizability of such algorithms. How can we guarantee the effectiveness of one model on a data set with characteristics that differ from the one the algorithm learned with? For example, an algorithm trained using data from a specific population may not perform as well when applied to a different population.

This program therefore aims to study new methods for improving generalization, and pursues four objectives. First, set up a research environment enabling the study of methods likely to improve generalization in real-world contexts. Second, optimize data flows obtained in real-world healthcare settings to serve algorithm research. Next, investigate specific issues related to algorithm generalization and secondary use of medical data. Lastly, create an open data set that can be used to build upon the research program findings.

Topic 4 – AI for the discovery of materials and molecules

Lead researchers: Yoshua Bengio (Université de Montréal, Mila), Michael Tyers (Université de Montréal, IRIC), Mickaël Dollé (Université de Montréal), Lena Simine (Université McGill)

Designing molecules with desired properties is a fundamental problem in drug, vaccine, and material discovery. Traditional approaches to designing a new drug can take over 10 years and a billion US dollars. Materials have been developed solely based on their performance characteristics leading to materials composed of rare, often toxic elements, which can inflict significant environmental damage. Artificial intelligence (AI) has the potential to revolutionize drug and material discovery by analyzing evidence from large amounts of data accumulated and learning how to search in the compositional space of molecules, and hence significantly accelerate and improve the process.

This program aims to build an efficient and effective machine learning framework for searching molecules with designed properties. It will be crucial to build upon, and extend, ongoing collaborations (i) between Mila and IRIC, aimed at optimizing the algorithms to discover new antibiotics and (ii) between Mila and materials experts at McGill and Université de Montréal, on the development of materials with environmental applications like fighting climate change. This multidisciplinary project also raises exciting fundamental challenges in AI regarding learning to search, modeling and sampling complex data structures like graphs, and may have applications to scientific discovery more broadly.

Topic 5 – Human-centered AI: From Responsible Algorithm Development to Human Adoption of AI

Lead researchers: Pierre-Majorique Léger (HEC Montréal, Tech3lab), Sylvain Sénécal (HEC Montréal, Tech3lab)

Human-AI interactions are common nowadays. We interact with artificial intelligence daily in performing many professional and personal tasks. Humans’ adoption of AI, however, is far from automatic, successful or satisfactory. Whether we are citizens, employees or consumers, issues such as bias, lack of trust and even low user satisfaction affect our likelihood of adopting AI in various contexts. To foster adoption, a holistic approach to AI is therefore needed. This multidisciplinary research program is investigating the full cycle of responsible AI development, from inception to adoption by users, putting people at the heart of the process. The goal is to map out guidelines for human-centred AI design using an iterative, multimethod methodological approach led by a multidisciplinary research team.

### Berk Bozkurt

McGill University

Model-Based Reinforcement Learning for Constrained Markov Decision Processes

“Despite the significant amount of research being conducted in the literature regarding the changes that need to be made to ensure safe exploration for the model-based reinforcement learning methods, there are research gaps that arise from the underlying assumptions and poor performance measures of the methods that are being proposed. As these certain gaps are preventing these proposed methods to be actually implemented as real life engineering solutions, they should be thoroughly investigated.

Our project aims to closely study these research gaps in order to obtain methods that asymptotically converge to the true model while ensuring safe exploration and to characterize the regret analysis of these methods, ultimately coming up with models that are closer to being actually implemented for real life applications.”

### Simon Chamorro

Supervised by: Christopher Pal

Polytechnique Montréal

Sim-to-real transfer for robotic control learning algorithms

Machine learning is a growing field that is very well suited for the development of robotic control algorithms. More precisely, reinforcement learning is especially relevant for this type of problem, where the robot must perform actions and interact with its environment. However, deep learning algorithms require phenomenal amounts of data as well as a lot of computational resources. These requirements make their training on robotic platforms very difficult. For this reason, the use of simulation platforms for the development and training of neural networks is very important, especially in the field of robotics. Therefore, it becomes essential that the learning performed in simulation be transferable to reality. This represents a great generalization challenge for the algorithms used, and it is an active research topic. The subject of the proposed research is to explore different generalization methods in order to develop algorithms to transfer the knowledge acquired in simulation to a real context. This could have a big impact and push the limits of the intelligence of today’s robots. Robotics can be applied to almost any industry and can automate, accelerate, and democratize services as well as help alleviate labor shortages. Possible applications are autonomous vehicles, assistance robots for the elderly or in hospitals, delivery drones, and many more.

### David Chemaly

Supervised by: Julie Hlavacek-Larrondo

Université de Montréal

Mesurer la masse de trous noirs lointains à l’aide de lentilles gravitationnelles et de l’apprentissage automatique

Dû au temps que prend la lumière à se déplacer, observer un objet distant signifie regarder dans le passé, soit quand la cible est beaucoup plus instable et énergétique. Ainsi, étudier un trou noir lointain nous permettrait d’en apprendre énormément sur les débuts de notre univers. L’objectif du projet est donc de développer, à l’aide de l’apprentissage automatique par réseau de neurones, un algorithme capable de rapidement et aisément résoudre la cinématique de trous noirs supermassifs lointains observés au travers de lentilles gravitationnelles. La résolution supérieure de ces données mènera à l’approfondissement de notre compréhension de jeunes trous noirs. Ceci permettra alors, pour la première fois, de mesurer la masse des trous noirs supermassifs lointains, un défi monumental qui s’avère impossible avec les méthodes traditionnelles d’astronomie et qui requière les techniques innovatrices des lentilles gravitationnelles, combinées aux outils d’apprentissage automatique. Suite à cela, les données obtenues seront sujettes à une étude physique approfondie lors d’un stage d’été à l’université de Cambridge au sein de l’équipe de Prof. Roberto Maiolino, directeur du Kavli Institute for Cosmology. Considérant leur résolution révolutionnaire, celles-ci mèneront certainement à de nouvelles découvertes. Étant donné le grand nombre de lentilles gravitationnelles connues, ce projet permettra de distinguer les structures jamais vues auparavant d’une quantité importante de trous noirs, tout en ayant le potentiel d’enrichir le savoir de la communauté scientifique sur l’état primaire de notre univers. Ce projet, co-supervisé par Prof. Hezavehe et Prof. Hlavacek-Larrondo, deux détenteurs des Chaires de recherche du Canada se terminera par la publication d’un article en tant que premier auteur.

### Olivier Denis

Supervised by: Jean-François Arguin

Université de Montréal

Identification des électrons du LHC à l’aide de réseaux neuronaux convolutifs

“La recherche en physique des particules a recours à des expériences qui tentent de recréer les conditions propices à la création des constituants fondamentaux de notre univers. L’expérience qui nous permet présentement d’accomplir cet exploit est un accélérateur de particule situé près de Genève en Suisse : le Grand collisionneur de hadrons (LHC). Celui-ci est un immense anneau souterrain de près de 27 km de circonférence, dans lequel deux faisceaux de protons sont accélérés en direction opposée à des vitesses très près de la vitesse de la lumière pour ensuite entrer en collision et ainsi produire les particules que l’on désire étudier.

Notre équipe se sert des données provenant du détecteur ATLAS (A Toroidal LHC ApparatuS), qui est l’une des 4 principales expériences au LHC. Celui-ci est composé d’une multitude de couches d’appareils servant à mesurer les trajectoires ainsi que les énergies des différentes particules produites à chaque collision.

Lorsqu’il fonctionne, le LHC génère une quantité faramineuse de données à analyser, parmi lesquelles l’identification des électrons joue un rôle clé pour de nombreuses analyses et a, entre autres, permis la découverte du fameux boson de Higgs.

Dans cette optique, notre groupe de recherche travaille donc à développer un classificateur qui peut identifier quelles particules sont de vrais électrons et quelles autres ne sont que du bruit de fond. Le classificateur est réseau de neurones qui est typiquement utilisé pour reconnaître des images. On fournit à ce réseau les images correspondant aux trajectoires et aux dépôts d’énergie des particules détectées dans ATLAS et le réseau peut alors identifier les électrons avec une efficacité 10 fois meilleure que celle de l’algorithme actuellement utilisé.”

### Clara El Khantour

Supervised by: Karim Jerbi

Université de Montréal

Étude de la cognition sociale dans la perception du rire : comparaison des modulations de l’activité cérébrale en MEG chez des sujets neurotypiques et sur le spectre de l’autisme par des méthodes d’apprentissage machine

“Le rire est une forme de communication non-verbale que nous utilisons souvent dans nos interactions au quotidien, mais qui a longtemps été négligée par la littérature neuroscientifique. Pourtant une meilleure compréhension de notre capacité à décoder le rire d’un interlocuteur aurait des implications importantes dans l’étude des interactions sociales et de certains troubles neuro-développementaux associés à des déficits de la cognition sociale, comme c’est le cas des personnes présentant un trouble du spectre de l’autisme (TSA).

Ce projet vise à déterminer les corrélats neuronaux et les processus de la cognition sociale sous-tendant la reconnaissance du rire social et spontané en magnétoencéphalographie (MEG) chez des individus neurotypiques et présentant un TSA en utilisant des méthodes d’apprentissage machine supervisée.

Nous appliquerons sur les données cérébrales MEG des méthodes statistiques en apprentissage machine supervisé, afin de mettre en évidence les paramètres discriminants entre les individus neurotypiques et les individus présentant un TSA lors de la perception et la reconnaissance du rire social et spontané.

À plus long terme, ce projet permettra de fournir de nouveaux outils pour le dépistage des maladies mentales, telles que l’autisme, dont les capacités sociales sont affectées.”

### Jean-Simon Fortin

Supervised by: Sébastien Hétu

Université de Montréal

L’habénula en contexte de gambling: une étude en imagerie par résonance magnétique fonctionnelle

Le jeu pathologique est un trouble mental caractérisé par un patron de jeu continu malgré des conséquences physiques, psychologiques et sociales négatives. L’habénula, une petite structure du cerveau situé dans le mésencéphale, semble impliquée dans cette pathologie. Il a été proposé qu’une dysfonction dans le traitement de l’information liée au feedback négatif empêcherait l’habénula de jouer son rôle dans l’extinction des comportements et dans l’utilisation du feedback négatif par le cerveau pour adapter sa stratégie d’un essai à l’autre. Cela serait à la source de l’incapacité des joueurs pathologiques à se servir du feedback négatif pour adapter leurs réponses et, ultimement, arrêter de parier (Zack et al., 2020). En raison de son possible rôle dans le jeu pathologique, il est essentiel de mieux comprendre le rôle du système de l’habénula dans une tâche de prise de décision similaire à celles auxquelles prennent part les joueurs. Toutefois, le système de l’habénula demeure méconnu (Yoshino et al., 2020) dû à la difficulté d’acquisition de données d’IRMf des régions du mésencéphale, entre autres à cause de leur petite taille et de leur proximité avec des artères importants du cerveau (Lawson et al., 2013). La présente étude vise, en employant une tâche de gambling, à vérifier si l’habénula fait partie d’un réseau distribué (Hétu et al., 2016) dont l’activité permet de prédire si le participant a pris une bonne décision et obtenu une récompense vs une mauvaise décision et subi une perte. Pour ce faire, nous utiliserons des “multi-voxel pattern analysis” (MVPA). Les MVPA prennent en compte des patrons d’activité s’étendant sur des ensembles de voxels afin de décoder l’information qu’ils contiennent collectivement (Cohen et al., 2017). L’une des manières d’utiliser les MVPA consiste à utiliser des classificateurs issus de l’apprentissage machine. Les classificateurs apprennent à pondérer l’activité de chaque voxel afin d’identifier une frontière de décision pour classifier Condition X vs Condition Y. Notre hypothèse est que l’activité cérébrale du système de l’habénula permettra de classifier si le participant a gagné ou perdu, suggérant ainsi un rôle de l’habénula dans le traitement de l’information relative à l’issue d’une décision (gain vs perte) en contexte de gambling. En mettant à profit des analyses basées sur l’apprentissage automatique, ce projet fournira de nouvelles connaissances sur le rôle du système de l’habénula dans un contexte de gambling, ce qui pourrait, à terme, améliorer notre compréhension du jeu pathologique et mener à des traitements.

### Oumayma Gharbi

Supervised by: Elie Bou Assi

Université de Montréal

AI and data mining EEG signals to classify depression and anxiety

According to the mental health commission of Canada, mental illness impacts 1 in every 5 Canadians each year. Unfortunately, the screening of patients with psychiatric conditions remains descriptive and is based on subjective questionnaires. The aim of this research project is to develop quantitative methods based on artificial intelligence capable of automatically screening patients with psychiatric conditions based on electroencephalography (EEG) recordings (brain’s electrical activity). Our specific objective is to use computational methods to implement probabilistic classification algorithms able to provide anxiety and depression scores based on EEG recordings. This research will be deployed at the Centre Hospitalier de l’Université de Montréal (CHUM) where more than 1200 patients are admitted for a routine EEG per year. The ability to identify anxiety and depression precursors from routine EEG recordings would significantly improve patient care and accelerate the referral of patients before their condition worsens.

### Rose Guay Hottin

Supervised by: Marco Bonizzato

Polytechnique Montréal

Clinical transfer learning pour l’intégration des connaissances d’experts dans l’optimisation automatisée des neurostimulations.

“La neuromodulation, ou stimulation électrique du système nerveux, est de plus en plus utilisée pour plusieurs indications cliniques. Par exemple, la stimulation cérébrale profonde peut être utilisée pour réduire les effets moteurs de la maladie de Parkinson tandis que la stimulation spinale épidurale peut être indiquée pour traiter la douleur chronique. L’efficacité de ces techniques dépend grandement des paramètres de stimulation (cible spatiale, intensité, forme des impulsions). Ainsi, les cliniciens sont confrontés au processus très coûteux de programmation de neurostimulations optimales dû au nombre énorme des combinaisons possibles.

Nous avons développé des systèmes automatisés, basés sur l’IA, pour pallier ce problème. Nos algorithmes surpassent la programmation humaine en trouvant rapidement la combinaison optimale dans plusieurs contextes de neuromodulation en temps réel. Il est important de noter que ce résultat est possible même en ayant fourni aucune connaissance préalable sur l’allure des données. Or, les chercheurs et cliniciens ont souvent des connaissances préalables ou des hypothèses quant à l’efficacité de certains paramètres. Ces connaissances proviennent de sources variées, telles que des données de groupe, de l’expérience du clinicien ou d’évidence antérieure spécifique au patient. Le transfert de ces connaissances au système algorithmique est d’une grande pertinence pour son utilisation en clinique.

Nous proposons un système de clinical transfer learning (CTL) permettant l’intégration flexible d’évidences cliniques préalables pour la programmation des neurostimulations. Un tel système devrait pouvoir tirer avantage des connaissances apportées en convergeant plus rapidement sur une neurostimulation optimale. Il devrait mettre à jour la connaissance existante en identifiant les différences avec les données récoltées. Le système avec CTL devrait également être assez robuste pour que sa performance ne soit pas affectée négativement si l’information proposée est inexacte. Notre nouveau système algorithmique avec CTL sera développé et évalué à l’aide de bases de données de stimulation chez l’animal déjà disponibles, ainsi que des données cliniques obtenues chez des patients traités pour la maladie de Parkinson et le Tremblement Essentiel avec la stimulation cérébrale profonde.

En ligne avec l’objectif d’IVADO de transformer les soins de santé grâce à l’intelligence numérique, ce projet vise à engendrer un changement de paradigme en programmation clinique des neurostimulations en développant un système d’optimisation pouvant tirer avantage conjointement de l’IA et du savoir clinique.”

### Nanda Harishankar Krishna

Supervised by: Guillaume Lajoie

Université de Montréal

Probing learning dynamics in recurrent neural networks and the brain

“Artificial Intelligence powered by neural networks is the driving force behind numerous applications in today’s world. While neural network models are capable of surprisingly good performance on many tasks, they are brittle and susceptible to adversarial attacks. Such challenges raise questions on their reliability and applicability to safety-critical domains, and thus motivate the study of mechanisms to improve their robustness.

Studying the mechanisms that underpin learning in the brain, which is capable of robust generalisation, could provide us insight on making neural networks more robust. Key to this study is understanding the dynamics of learning in the brain. We propose to develop techniques for the analysis of learning dynamics from activations observed during training, based on experiments with RNNs. We then aim to utilise these tools to analyse learning dynamics in a proxy parameter space extracted from neural activity, for a brain-machine interface task.

This would help us gain valuable insight on the nature of parameter updates in the brain, and also the nature of objectives that drive learning. Furthermore, it would spur the development of superior biologically-inspired algorithms for deep learning models, with the goal of improving their reliability.”

### Nizar Islah

Supervised by: Eilif Muller

Université de Montréal

Neocortical Continual Learning

Long-term activity dependent changes in synaptic strength are thought to underlie learning in the brain, but the mechanisms by which these synaptic changes are coordinated to achieve human-level learning remain a mystery. While deep learning in artificial neural networks provides a concrete example for how learning might be coordinated, contrasting and understanding the key differences to learning in brains could provide both a better understanding of the learning mechanisms in the brain, and inspire new deep learning approaches. The ability to learn new things without forgetting previously what was previously learned, known as Continual Learning, is one property of the human neocortex that deep learning doesn’t reproduce. This project aims to understand the mechanisms of how this emerges using modeling approaches unifying deep learning with architecture and constraints from the neocortex.

### Katia Jodogne-del Litto

Supervised by: Guillaume-Alexandre Bilodeau

Polytechnique Montréal

Détection et segmentation d’instances d’objets en temps réel par approximation de masque

“Le développement des véhicules intelligent est une avancée technologique majeure, qui se repose en partie sur des techniques de vision par ordinateur pour observer l’environnement, que ce soit dans les cas de conduite autonome ou d’assistance à la conduite. Il est nécessaire de pouvoir détecter en temps réel et de manière précise les usagers de la route dans une scène dense. L’utilisation d’approximation permet d’augmenter fortement la vitesse de détection sans pour autant perdre beaucoup d’informations.

En vision par ordinateur, une des briques essentielles à de nombreuses tâches est la détection d’objets dans une image. Il s’agit pour le système d’être capable de produire un rectangle autour d’une instance de l’objet, tout en l’identifiant. On peut ensuite indiquer pour chaque pixel s’il fait partie ou non de l’objet détecté, c’est la segmentation d’instances. Celle-ci produit un masque pour chaque instance détectée. Il s’agit d’une méthode beaucoup plus précise, mais également beaucoup plus coûteuse en temps de calcul. L’approche intermédiaire proposée dans ce projet est un compromis entre le simple rectangle et le masque. Pour chaque instance, c’est un polygone qui sera tracé autour de l’objet, détecté à partir de son centre. La méthode s’applique principalement aux usagers de la route dans des cadres urbains où la densité est forte, avec les ensembles de données Cityscapes, KITTI, IDD. Il s’agit dans ce projet de s’inspirer de la méthode de détection CenterPoly pour proposer une approche plus robuste, en s’intéressant tout particulièrement à l’impact de la fonction d’erreur et du système de représentation des polygones d’approximation.”

### Ilitea Kina

Supervised by: Louis-Martin Rousseau

Polytechnique Montréal

Utilisation de l’intelligence artificielle pour prédire les délais logistiques menant à la congestion de l’urgence

“La congestion dans les urgences est un véritable fléau au Québec, avec un impact sur la qualité de soins aux patients et le bien-être du personnel soignant. Malgré que de nombreuses études ont démontré une augmentation de la mortalité et de la morbidité chez les patients qui subissent la congestion hospitalière, celle-ci demeure difficile à prédire avec les outils et mesures actuels. Dans un contexte de pénurie de personnel, ce problème est d’autant plus important.

La congestion des urgences découle des délais logistiques associés aux examens de radiologie et de biochimie, aux délais de consultations et aux délais de transfert intra-hospitalier des patients. En analysant les données de ces différents délais dans la trajectoire du patient, nous voulons prédire, avec un outil d’intelligence artificielle, des seuils de délais qui sont à risque de faire congestionner l’urgence. Ces délais seront dynamiques; ils varieront selon l’état de l’urgence en temps réel. En ayant des seuils qui prédisent la congestion à l’urgence en temps réel, ou bien une combinaison de plusieurs seuils, les urgences auront la capacité de rapidement détecter pour quels départements des solutions devraient être entreprises. Une prise en charge plus rapide de ces délais pourrait permettre d’éviter la congestion à l’urgence.

Notre analyse se fera à partir de l’urgence de l’hôpital de la Cité-de-la-Santé de Laval, une des urgences les plus achalandées au Québec, et inclura les données des trois dernières années. Nous créerons des algorithmes d’apprentissage automatique, en se concentrant sur les optimal classification trees, afin de prédire la congestion de l’urgence. Nous évaluerons leur performance prédictive pour déterminer les techniques les plus adaptées à l’urgence de Cité-de-la-Santé.”

### Élodie Labrecque Langlais

Supervised by: Frédéric Lesage

Polytechnique Montréal

Développement d’un algorithme de prédiction automatisée du succès de la procédure chirurgicale Transcatheter Aortic Valve Implantation (TAVI) à partir d’images CT

Les valvulopathies cardiaques touchent de 8 à 13 % de la population âgée de plus de 65 ans dans le monde entier et ce fardeau s’alourdit avec le vieillissement de la population. Longtemps traitées par des interventions chirurgicales à cœur ouvert, le remplacement valvulaire aortique percutané (Trans Aortic Valvular Implantation, TAVI) a révolutionné le traitement de la sténose aortique. Il consiste à remplacer la valve aortique malade en utilisant un cathéter introduit par une ponction de l’artère fémorale. Initialement adopté seulement pour les patients inéligibles à la chirurgie ou à haut risque chirurgical, il est maintenant validé pour les patients avec tous les profils de risque. L’expansion accrue du TAVI reste cependant dépendante de son succès clinique, soit le succès procédural (décès, migration de la valve) et de la sécurité précoce (la réduction au minimum des complications telles que les fuites autour de la valve et les troubles de conduction qui affectent la survie à long terme). Le succès clinique du TAVI repose lourdement sur la caractérisation de la valve aortique et des structures anatomiques adjacentes observés sur des CT scan cardiaque. Aucun score clinique de prédiction du succès du TAVI n’existe actuellement. Seules des avancées technologiques comme l’intelligence artificielle et plus précisément les réseaux neuronaux convolutifs peuvent prédire le succès du TAVI à partir des CT scans cardiaques. L’objectif principal du projet est alors de développer un algorithme qui permet de prédire automatiquement le succès d’une procédure TAVI à partir de CT scans du cœur.

### Myriam Lizotte

Supervised by: Guy Wolf

Université de Montréal

Diffusion Geometry & Topology Approach to Data Fusion and Mitigating Batch Effects

“It is becoming crucial to combine datasets collected in different circumstances, but this task is challenging due to various inconsistencies introduced by data collection artifacts and inherent biases. This gives rise to a set of challenges often referred to as data fusion or batch effect removal, which can be divided into two main tasks that we aim to tackle in this MSc research. The first is combining data of the same system from different sensors, each with its own calibration, scale, and level of noise. The second is combining datasets that measure the same variables but in different conditions (e.g., subjects, locations, or time of day). This creates a problem called batch effects, where confounding variables hide the effect of true variables of interest. Even small batch effects, or their incomplete removal, can significantly bias statistical conclusions and detract from the ability to provide reliable insights from biomedical data.

Significant efforts have been recently invested in unsupervised data fusion and mitigation of batch effects. Several diffusion-geometry approaches have been developed, including integrated diffusion and harmonic alignment. The former aims to combine multimodal data while denoising and adjusting for discrepancies in data capture resolutions. The latter aims to rigidly align or fit datasets together so that their information is comparable, in other words alleviating or removing batch effects. In this project, we will build upon and combine ideas from previous diffusion-based approaches, while aiming to relax the required assumptions. We will leverage the recent diffusion condensation framework, which captures data representations at different scales or resolutions, to identify which local regions or resolutions are most appropriate for aligning data, based on their intrinsic topological “shape”. Another goal is to pinpoint meaningful local differences between datasets, as opposed to global deformations due to technical artifacts.

On a fundamental level, this research will further explore the intersection between diffusion geometry and topological data analysis, merging the two dominant approaches to manifold learning in data exploration. On the application side, it will address critical challenges in biomedical data exploration, especially encountered in high-throughput multi-modal data from multi-sample cohorts, to enable new research avenues and significantly advance the frontiers of AI and health.”

### Aristides Milios

Supervised by: Siva Reddy

McGill University

Language Models Enhanced with Hyperlink Graphs

### Camille Rondeau Saint-Jean

Supervised by: Timothée Poisot

Université de Montréal

Reconnaissance de microdialectes et de chants individuels de Bruants des prés avec un réseau neural profond

“Mon projet consiste à explorer les capacités de l’apprentissage profond pour distinguer le chant d’individus ou de différents groupes sociaux au sein d’une même espèce d’oiseaux. Je dispose d’un répertoire de plusieurs centaines d’heures d’enregistrements de Bruants des prés issus de la population de l’île Kent, au Nouveau-Brunswick. On y entend des chants appartenant à six microdialectes, qui s’apparentent à des accents locaux qui distinguent les oiseaux vivant dans différentes parties de l’île.

Je prévois entraîner un réseau neural profond à distinguer ces microdialectes et à identifier acoustiquement les mâles que les chercheurs ont observés visuellement sur le terrain. Il est difficile pour un observateur humain de distinguer différents Bruants des prés à l’oreille, car leur chant consiste en une succession très rapide de syllabes brèves et aigües. En analysant les caractéristiques du chant sur lesquelles se basera le réseau neural pour sa classification, nous comprendrons mieux comment ils communiquent entre eux leur identité et leur appartenance à un groupe local.

Le développement d’outils automatiques facilitant la reconnaissance d’individus ou de dialectes aidera à accélérer les recherches sur le comportement et sur la structure des populations. Quand ils seront mis au point, il sera possible d’exploiter avec un tout autre niveau de détail et à très grande échelle les enregistrements de sons d’oiseaux qui sont désormais faciles et peu coûteux à obtenir sur le terrain. Comme la diversité culturelle et génétique au sein même d’une même espèce est importante pour la survie des populations, les informations que nous pourrons alors acquérir serviront à mieux orienter la conservation dans le but de freiner la crise de la biodiversité.”

### Myriam Sahraoui

Supervised by: Bruno Gauthier

Université de Montréal

Étude de l’évaluation du profil cognitif de l’enfant en combinant EEG haute densité, EEG mobile et intelligence artificielle

L’évaluation des capacités cognitives chez les enfants se fait essentiellement à l’aide de tests neuropsychologiques. À titre d’exemple, les tests de fluidité graphique mettent en valeur diverses aptitudes comme la flexibilité cognitive, la planification et la créativité. À ce jour, les processus cérébraux associés à ces capacités chez l’enfant sont encore mal compris. Cette limitation est en partie due aux contraintes associées au contexte d’examen nécessitant un enregistrement d’électroencéphalographie (EEG) en laboratoire. Depuis une quinzaine d’années, des versions portatives à faible coût de ces appareils ont été développées pour des usages commerciaux, permettant par exemple de faire de la méditation de façon plus efficace. Ces appareils sont de plus en plus utilisés en recherche, et ouvrent la voie vers une toute nouvelle façon d’étudier la cognition de l’enfant, particulièrement dans des milieux naturels, hors laboratoire. Toutefois, à notre connaissance, aucune étude ne s’est penchée sur leur utilisation pour l’étude de la cognition normale chez l’enfant. Par ailleurs, le développement d’algorithmes en intelligence artificielle permet aujourd’hui une meilleure exploration des données en neuroimagerie. L’objectif de ce projet est, dans un premier temps, de collecter des données cérébrales chez des enfants pendant l’exécution d’un test neuropsychologique, puis d’utiliser des algorithmes d’apprentissage pour comparer les résultats d’un EEG mobile avec un EEG de laboratoire. Tout d’abord, 40 participants de 6 à 12 ans seront recrutés et assignés à une condition EEG mobile ou de laboratoire. Ensuite, un enregistrement EEG sera effectué pendant la passation d’une version numérique du Five-Point Test, un test mesurant la fluidité graphique. Ces groupes seront ensuite comparés sur la qualité des enregistrements cérébraux. Finalement, l’apprentissage machine sera utilisé afin d’identifier des sous-groupes de participants avec des profils cognitifs distincts. L’utilisation de l’apprentissage non supervisé permettra une exploration des données sans apriori en neuropsychologie de l’enfant, un milieu où ce type d’approche est encore sous-exploité. Dans un second temps, ce projet consistera à développer des modèles de prédiction permettant d’améliorer la résolution spatiale de l’EEG mobile pour l’approcher de la résolution d’un EEG de laboratoire. Ces modèles prédictifs seront développés à partir d’un ensemble de données de plus grande envergure en utilisant des algorithmes basés sur les réseaux de neurones artificielles. Au-delà de l’intérêt d’utiliser des EEG sur le terrain, fournir des preuves de l’utilité de L’EEG mobile pourrait favoriser une plus grande participation des populations vulnérables et éloignées pour la recherche en neurosciences cognitives.

### Rebecca Salganik

HEC Montréal

Exposure Fairness in Music Recommendation

Given the expansive growth of musical databases, streaming companies have employed the use of recommendation systems to streamline the process of discovering new music. Unfortunately, recent years have come with the discovery of a wide range of biases that can seep from within these models and into the lives of users. One such discovery has been that of popularity bias. Popularity bias is exemplified when algorithmic reliance on pre-existing data causes a feedback loop in which previously popular items (with many ratings) are recommended instead of new, or less well-known items with less ratings. Such a loop can force the algorithm into avoiding new or niche items when it guides a listener’s discovery process. This is deeply concerning as such a bias can have very serious consequences on the financial prospects of artists, the musical experiences of listeners, and, on a broader scale, our cultural values related to art. As such, our goal takes on the task of mitigating popularity bias, or, equivalently, promoting exposure fairness in music recommendation. In our work, we develop a novel, multimodal musical dataset containing social, cultural, and musical information necessary. We use this dataset to train a state-of-the-art graph neural network based recommender system, PinSage, to develop robust representations for musical items. Finally, we use in-processing methods to develop a fairness loss which, when combined with training loss, enables PinSage to perform both fair and relevant recommendations.

### Hugo Schérer

Supervised by: Yashar Hezaveh

Université de Montréal

Characterization of the distribution of dark matter on small scales using strong gravitational lensing

“Astrophysical observations indicate that there exists an unknown, exotic type of matter called dark matter, which actually constitutes the vast majority of matter in the Universe. Discovering dark matter and its properties is one of the most important priorities in modern physics. One powerful way to characterize dark matter it through a phenomenon called strong gravitational lensing. This phenomenon happens when there are two galaxies, one directly behind the other from the Earth’s perspective. The closer one is the lensing galaxy, and the one further away is the source galaxy. Because massive objects curve spacetime, the lensing galaxy distorts the image of the source galaxy, very much like an actual lens distorts images. Studying these distortions make it possible to learn about the distribution of dark matter inside the lensing galaxy, and provide invaluable information about this unknown form of matter.

My research will contribute to this study of dark matter using data from strong gravitational lenses. One major challenge is the analysis of this data which is difficult and takes a lot of computing time with traditional analysis methods. Furthermore, very large amounts of data will become available in the coming years, which makes this challenge even more important. Neural networks have been shown to be extremely promising to dramatically reduce the computing time in analyzing such data. Neural networks are computing algorithms that fall under the category of machine learning, also sometimes referred to as artificial intelligence. My research will be to take part in developing machine learning analysis methods to analyze data from strong gravitational lenses. I also plan to apply the methods that I develop to real data, which will further advance our understanding of dark matter in particular and of the Universe in general.”

Supervised by: Laurence Perreault-Levasseur

Université de Montréal

Reconstructing the Initial Conditions of the Universe Using Deep Learning

“We, humans, have had the old question “where do we stand in the universe?” for centuries. We are curious to understand our origin, our place in the cosmos, and how our universe evolves. Nowadays, these questions are investigated in modern cosmology by analyzing astronomical observations through physical models. An essential step in this direction is reconstructing the universe’s initial conditions (i.e., its contents, their initial spatial distribution, and a handful of parameters underlying their evolution through time), contributing to our understanding of the universe and constraining fundamental physics models.

The data coming from modern telescopes such as the James Webb Space Telescope, Euclid, and Vera Rubin Observatory significantly enhance the volume and the resolution of available cosmological data. This wealth of data provides remarkable opportunities for discoveries; however, it requires innovative inference techniques. This project will utilize the advances in deep learning and data science to develop novel tools for high-dimensional inference in cosmology. These tools will not only pave the way for cosmologists to discover the initial conditions of the universe but also hold the promise of bringing new ideas and methods that can be fruitful for inference in other realms of science.”

### Clémence Delfils

Université de Montréal

Stage chez La Presse

### Ève Ménard

Université de Montréal

Stage chez Le Devoir

### Baptiste Pauletto

Polytechnique Montréal

Stage chez Le Devoir

### Sahar Ramazan Ali

Université de Montréal

Stage chez CIRANO

### Thi Sopha Son

Université du Québec en Outaouais

Stage chez CIRANO

Université du Québec à Trois-Rivières

Stage chez La Presse

### Postdoctoral research funding

Supervised by: Charles Audet

Polytechnique Montréal

Topology Optimization for Conformal Cooling in Molds and Dies

“The manufacturing of aluminum parts is plagued by problems related to the optimization of manufacturing molds. The quality, reliability, and production efficiency of these parts are directly related to the ability of the molds to efficiently transfer heat to facilitate cooling. The industrial technique traditionally used to dissipate heat is the drilling of holes in the mold die. These holes are generally not optimally placed, resulting in thermal stresses and microstructural variations in the manufactured parts. In this context, the availability of molds with complex cooling channels could remedy this limitation.

In this project, we propose to design an open-source topological optimization platform based on the tools of Finite Element Modeling (FEM), Artificial Intelligence (AI), and Black Box Optimization (BBO) for thermofluidic problems to specify the optimal location and shape of cooling channels. The project will leverage the strengths of the fellow, host supervisors (Professors Charles Audet and Bruno Blais) at Polytechnique Montreal, and R&D collaborators at the National Research Council (NRC), Canada which allows for a significant synergy between the proposed research and institutional strategic priorities. Furthermore, the project will contribute towards strengthening the development of energy-efficient manufacturing expertise of the local Quebec Aluminum manufacturing industry.”

### Yacine Bareche

Supervised by: John Stagg

Université de Montréal

DeepPredictIO: A Pan-Cancer Deep Learning Framework to Predict Response to Immune Checkpoint Inhibitors

Cancer is the second deadliest disease worldwide. In the last decade, several agents called immune checkpoint blockade (CPI), aimed at boosting patients’ immune system to fight the tumor and led to exciting results with a long-lasting response for some patients. However, CPI remains extremely costly for patients and public health care (>10,000 per patient per month), yet a large portion of CPI-treated patients (60-80%) do not derive benefit from this therapy. Thus, the identification of a robust biomarker of response to treatment and easily applicable in clinical routine practice is currently one of the most important active fields in immuno-oncology. In a previous work, we developed a robust and powerful biomarker of response to CPI therapy based on RNA-sequencing, called PredictIO. Yet, despite the continuous effort in decreasing cost and processing time, RNA-sequencing remains not suited for clinical routine practice. Hematoxylin & Eosin (HE) stained Whole-Slide Images (WSIs) is the current gold standard for solid tumor diagnostic, with cheap and fast protocols used worldwide. We hypothesize that building a model able to efficiently predict CPI-response from HE WSI, would greatly improve clinical decisions, with more robust patient stratification through the use of a tool directly applicable in clinical routine practice. ### Edoardo Maria Ponti Supervised by: Siva Reddy McGill University Skill Discovery in Language Models One of the key goals of natural language processing is devising models that can use language creatively and in unforeseen circumstances. In humans, this is possible by virtue of the fact that each linguistic “task” (e.g., answering a question) results from the combination of different sets of skills, which are autonomous and reusable facets of knowledge. The goal of my project is to integrate such a modular design into neural architectures. In particular, I will focus on language understanding grounded in a simulated environment. Upon receiving a new instruction, an agent will execute a sequence of actions conditioned on a specific subset of skills it has learned (navigating a room, picking up objects, opening doors, etc,). Similarly, latent skills can be discovered for language generation, thus controlling the text a model outputs and making it more diverse. Thus, my project holds promise to better align machine learning with human creativity in language usage. ### Denahin Hinnoutondji Toffa Supervised by: Dang Khoa Nguyen Université de Montréal Adaptation d’un classificateur multimodal au diagnostic électroencéphalographique de l’épilepsie L’épilepsie est une maladie cérébrale qui affecte environ 140000 Canadiens et 50-60 millions de personnes au monde. En 2020, le diagnostic d’épilepsie dépend encore de l’identification de détails visuels caractéristiques sur l’électroencéphalogramme (EEG) : les pointes épileptiformes. Toutefois, ces pointes sont absentes ou de signification incertaine dans environ 30-70% des cas, conduisant ainsi à des excès ou à des retards diagnostiques parfois graves. Pour améliorer les chances d’un diagnostic précoce même en l’absence de pointes, nous proposons un algorithme pour une analyse multimodale automatisée ciblant à la fois les pointes et des détails EEG infravisuels. De même, là où un neurologue tablerait subjectivement sur son expérience de 2-30 ans, notre algorithme d’IA qui pourra objectivement résoudre les situations diagnostiques en se basant sur un recul équivalent à des centaines d’années de carrière. Au total, plus de 25000 échantillons EEG anonymes avec les diagnostics corrélés seront sélectionnés au CHUM pour entrainer la version de base de l’algorithme à reconnaître les EEG qui sont corrélés à l’épilepsie. Cet outil sera une solution innovante dont l’optimisation en contexte clinique améliorera significativement le seuil diagnostique de l’EEG. Il pourrait aussi radicalement changer la stratégie prise en charge de l’épilepsie, voire celle d’autres affections neurologiques. ### Ankur Mali Supervised by: Eilif Muller Université de Montréal Surprising you learn, not surprising you don’t: A model of neocortical perception and learning based on prediction Here we aim to develop a deep learning network architecture that combines the strengths of predictive coding and self-supervised learning approaches and is constrained by the neocortical architecture, to account for the role of integrated contextual priors and surprise in both inference and learning. We will leverage an initial prototype bio-inspired deep learning model of a mechanism for combining contextual priors with sensory inference at one cortical region that was recently developed in the lab of Dr. Eilif Muller. We will augment this prototype, and 1) incorporate a model of surprise accounting for known neocortical perceptual dynamics under violation of contextual priors, and 2) develop a local learning algorithm consistent with synaptic plasticity dynamics between pyramidal integration zones for sensory input and contextual priors (basal and apical dendritic compartments, respectively). The latter will be performed in close collaboration with computational neuroscientists working in the lab and other experimental groups studying such plasticity dynamics. We will develop the network architecture to scale to learning in a deep hierarchy on tasks and datasets consisting of natural images. ### Aaron Berk Supervised by: Tim Hoheisel McGill University Realistic sampling strategies for deep generative inverse problems in medical imaging Generative neural networks (GNNs) have shown impressive performance in capturing intrinsic low-dimensional structure in natural images. For instance, some produce realistic-looking images of human faces. This makes them promising candidates for modelling complex structured data. Recently, GNNs are being developed as structural proxies for inverse problems in medical imaging, such as magnetic resonance imaging and computed tomography. Robustness and interpretability are mandates in medical imaging. However, both are major open challenges for neural network-based reconstruction methods, and a solid theoretical understanding is required to address them. In this research program, we propose a new theoretical analysis aimed at improving the reliability of GNNs in crucial applications. In particular, we will investigate the use of GNNs as structural proxies for inverse problems by elucidating optimal sampling strategies for realistic measurement processes. This work will develop new probabilistic machinery, which we expect to be useful for analyzing other open questions about neural networks. ### Véronique Brousseau-Couture Supervised by: Normand Mousseau Université de Montréal Optimisation de l’ensemble d’entraînement pour la modélisation atomistique de batteries solides Face à la menace des changements climatiques, le besoin de développer des méthodes de stockage d’énergie efficaces, sécuritaires et peu coûteuses est considérable, car de tels dispositifs sont nécessaires pour favoriser la migration vers des énergies propres et renouvelables. Les batteries solides comptent parmi les avenues les plus intéressantes. Pour identifier des matériaux prometteurs, il faut toutefois investiguer un très grand nombre de candidats, ce qui est beaucoup trop coûteux numériquement avec les méthodes de pointe actuelles. L’utilisation de l’apprentissage automatique pour développer les potentiels interatomiques qui nous permettront d’effectuer les calculs à moindre coût est donc une avenue de prédilection. Or, les calculs nécessaires pour construire un ensemble de données d’entraînement efficace demeurent particulièrement lourds, puisque l’on doit simuler des milliers d’atomes. Ce projet propose donc d’utiliser le concept d’espace latent développé dans les réseaux de type encodeur-décodeur pour identifier un ensemble de configurations atomique plus simples, comprenant moins d’atomes, qui donneraient néanmoins au modèle d’apprentissage machine une information équivalente à des configurations très complexes. En optimisant ainsi la construction de notre ensemble de données d’entraînement, on pourra donc améliorer les capacités de prédiction de nos modèles d’apprentissage machine tout en réduisant considérablement le coût numérique des calculs. ### Arthur Chatton Supervised by: Mireille Schnitzer Université de Montréal Clustered super learning for optimal decision rules Super learner is a powerful supervised ensemble learning method optimizing the predictions by combining several machine learning approaches. However, several contexts in health or social sciences involve repeated measures for each unit, requiring specific loss functions and cross-validation schemes due to autocorrelation. Our main goal is to extend the super learner to such clustered data to predict the most at-risk statistical units of poor outcomes. We plan to develop a dynamic weighted outcome linear regression estimator for the clustered data in a second step. Then, we will use this estimator to identify effect-modifying variables to build rules for optimally setting the exposures to improve patient outcomes. We will apply these methodological developments in nephrology to optimize online dialysis sessions according to the patient’s and session’s characteristics. ### Eduard Gorbunov Supervised by: Gauthier Gidel Université de Montréal Improving the Theory of Numerical Methods for Solving Variational Inequalities and Distributed and Stochastic Optimization Problems The goal of the project is to improve theoretical understanding of existing methods for solving variational inequalities, distributed and stochastic optimization problems, and to design new efficient methods with better convergence guarantees in comparison to existing ones. In particular, we plan to push further the theory of stochastic methods for variational inequalities and min-max problems motivated by machine learning applications, obtain new theoretical results on stochastic optimization with heavy-tailed noise in the gradients, develop new efficient distributed methods robust to Byzantine attacks, and new communication-efficient distributed methods with compression. Theoretical challenges we consider are motivated by the various applications including training of GANs, federated learning, collaborative learning, and training complicated deep learning models on NLP tasks. ### Hélène Verhaeghe Supervised by: Gilles Pesant Polytechnique Montréal Improving the solving of scheduling problems using machine learning “The goal of this project is to use machine learning algorithms to help improve the solving of scheduling problems such as the Resource-Constrained Project Scheduling Problem (RCPSP). Constraint programming (CP) has been proven effective to solve such scheduling problems. However, instances with more than 500 tasks are still hard to solve by any method. The target of this project is to be able to solve instances with up to 2000 tasks. Combining ML and CP has already been proven successful in multiple occasions. Here, ML algorithms would principally be used to find clusters of hard-to-schedule tasks in order to schedule them first, as a sub-problem. “ ### Hao-Ting Wang Supervised by: Pierre Bellec Université de Montréal Impact of age and sex on transdiagnostic brain biomarkers amongst neurodegenerative conditions “With the ageing of the Canadian population, neurodegenerative diseases are reaching epidemic levels. At present, it is not possible for clinicians to predict accurately if and when a patient with no symptoms or mild cognitive impairment will start experiencing debilitating symptoms of dementia given basic information such as age and sex. An early, reliable diagnosis could however dramatically increase the effectiveness of current and future interventions. Magnetic resonance imaging is a promising technology to assist clinicians in making a precise diagnosis by providing a non-invasive window in the structure and function of the brain. With the innovation of AI combined with availability of multisite datasets, we can apply the rich information found in these brain images to clinical practice. In order to understand what is the impact of age and sex on diagnostic brain markers, we need to see how brain markers vary in a large number of contexts and individuals. “ ### Marta Zagorowska Supervised by: Moncef Chioua Polytechnique Montréal Robust and data-Efficient Learning for Industrial Control “Increasing energy and resource efficiency in industrial systems is key to decrease harmful emissions by 90% by 2050. Reaching the environmental targets requires a holistic approach to how resources and energy are delivered to the industry by means of distribution networks, such as heat networks, electricity networks, or gas transport networks. I will devise new control strategies that ensure robust operation of distribution networks while ensuring safety and satisfaction of environmental objectives. The environmental performance of the whole system hinges on the performance of distribution networks. Optimal control of such networks is complex due to timescales, from milliseconds to ensure safe operation of pumps or generators, to days or months to include environmental goals, spatial complexity, uncertainty related to varying operating conditions, incomplete information available, and limited computational power. Existing control frameworks are usually application specific and have limited use in large-scale systems. In the project, I will use advanced theory in data analytics and optimisation and build on my industrial experience to develop operating strategies for distribution networks that will enable safe implementation and reaching the environmental targets.” ### Wentao Zhang Supervised by: Jian Tang HEC Montréal Graph Data Mining “I will continue to focus on graph data, graph models, and graph systems. Concretely, my future research works will include: 1. AutoML on Graph: model and system Recently, the combination of AutoML and graph data mining has aroused lots of concern. I will focus the researches on auto-knowledge graph, auto-network embedding and graph neural architecture search, and then build a system to better use them. 2. Data-centric Graph Mining Many state-of-art graph models are data-driven. Building data-driven graph model (e.g., robust model under data noise, weakly-supervised learning, self-supervised learning and federated learning under limited graph data) is my other future work. 3. Machine Learning for Graph Data Data is playing an increasingly central role in creating ML solutions. So, I hope to build a system that can deal with the data challenges (e.g., data annotation, data cleaning, and data augmentation) in graph. “ ### Undergraduate research initiation grants ### Anna Andrienko Supervisé.e par : Margarida Carvalho Université de Montréal A binary decision diagram-based approach for interdiction games: Critical Node Problem ### Vanessa Bellegarde Supervisé.e par : Julie Hussin Institut de cardiologie de Montréal (ICM) Prédire la proportion d’ethnicité à l’aide du Diet Network ### Geneviève Bistodeau-Gagnon Supervisé.e par : Guy Wolf Université de Montréal Integreted data-driven approaches for understanding immunological data ### Marise Bonenfant Supervisé.e par : Lubna Daraz Université de Montréal Ontology Assessing the Reliability of Mental Health Tools and Information on the Internet ### Mégan Brien Supervisé.e par : Frédéric Gosselin Université de Montréal Decoding real-world visual recognition abilities in the human brain ### Ariane Brucher Supervisé.e par : Phaedra Royle Analyses statistiques de potentiels évoqués lors du traitement linguistique chez les adolescents francophones neurotypiques ### Sara-Ivana Calce Supervisé.e par : An Tang Centre hospitalier de l’Université de Montréal (CHUM) Application de techniques d’apprentissage profond pour la classification de maladies diffuses du foie par imagerie ultrasonore ### Marianne Chevalier Supervisé.e par : Éric Lacourse Université de Montréal Exploration d’approches computationnelles pour analyser et comparer des données d’enquête ### Jonathan Couture Supervisé.e par : Alexandre Dumais Centre intégré universitaire de santé et de services sociaux de l’Est-de-l’Île-de-Montréal (CIUSSS - EMTL) Hôpital Maisonneuve-Rosemont et Institut universitaire mentale de Montréal) Thérapie Avatar pour traiter la schizophrénie résistante : Modélisation du processus thérapeutique à l’aide du Traitement Automatique du Langage humain ### Simon Del Testa Supervisé.e par : Vincent-Philippe Lavallée Centre hospitalier universitaire Mère-Enfant (CHU Sainte-Justine) Analyse génomique comparative de leucémies au diagnostic et à la rechute ### Guillaume Dubé Supervisé.e par : Éric Lacourse Université de Montréal Démocratisation du concept de régularisation pour la recherche en sciences sociales ### Clara El Khantour Supervisé.e par : Karim Jerbi Université de Montréal Analyse de données cérébrales MEG combinant analyses spectrales et apprentissage machine ### Gaspar Faure Supervisé.e par : Guillaume-Alexandre Bilodeau Détection et suivi simultané d’objet ### Louis-Simon Guité Supervisé.e par : Julie Hlavacek-Larrondo Université de Montréal Evidence of Massive Runaway Gas Cooling in High-Redshift Clusters of Galaxies ### Armelle Jézéquel Supervisé.e par : Michel Gagnon Polytechnique Montréal Grattage web des événements culturels du Québec ### Rose Jutras Supervisé.e par : Frédéric Gosselin Université de Montréal Individual differences in semantic and conscious processing of natural scenes. ### Jin Kwon Supervisé.e par : Aude Motulsky Centre hospitalier de l’Université de Montréal (CHUM) Développement d’outils pour un recueil de données standardisées du prescriptome des patients ### Yassine Lamrani Supervisé.e par : Elie Bou Assi Centre hospitalier de l’Université de Montréal (CHUM) Objets connectés et IA : Détection des crises d’épilepsie à partir de la respiration ### Audrey Lamy-Proulx Supervisé.e par : Sébastien Hétu Université de Montréal Le traitement de la hiérarchie sociale et son influence sur la prise de décisions ### Julie Lanthier Supervisé.e par : Jean Provost Implémentation d’un logiciel permettant de réaliser l’imagerie fonctionnelle cérébrale ultrasonore dans plusieurs modèles animaux ### Justine Le Blanc-Brillon Supervisé.e par : Sébastien Hétu Université de Montréal Studying social influence using a neuroimaging and data science approach ### Dragos Cristian Manta Supervisé.e par : Adam Oberman Generalization for Semi-Supervised Learning ### Nadine Mohamed Supervisé.e par : Roberto Araya Centre hospitalier universitaire Mère-Enfant (CHU Sainte-Justine) DIGITAL MORPHOLOGICAL ANALYSIS AND MODELING OF NEURONAL DENDRITES AND DENDRITIC SPINES IN NEURODEGENERATION ### Manuel Pellerin Supervisé.e par : Michel Desmarais Polytechnique Montréal Analytique de l’apprentissage dans Moodle en vue de la réalisation de tableaux de bord d’engagement et d’apprentissage ### Félix Pellerin Supervisé.e par : Antoine Lesage-Landry Polytechnique Montréal Optimal policies to mitigate power system-ignited wildfires and peak demand ### Sophie Rodrigues-Coutlée Supervisé.e par : Alexandre Dumais Centre intégré universitaire de santé et de services sociaux de l’Est-de-l’Île-de-Montréal (CIUSSS - EMTL) Hôpital Maisonneuve-Rosemont et Institut universitaire mentale de Montréal) Innovative Virtual Reality and Artificial Intelligence based psychotherapy to reduce cannabis use in patients with psychotic disorders ### Emily Tam Supervisé.e par : Blake Richards Exploring the role of Dale’s Law in Artificial Neural Networks ### Nicole Tebchrany Supervisé.e par : Jean Provost Senseur optique d’ultrasons ### Jacqueline Nguyen Phuong Trieu Supervisé.e par : Anne Gallagher Centre hospitalier universitaire Mère-Enfant (CHU Sainte-Justine) Optimization of data analysis strategies for the ELAN Project: a multimodal approach ### Konstantinos Tsiolis Supervisé.e par : Adam Oberman McGill University Statistical Learning Theory Applied to Word Embeddings ### Élise Vaillancourt Supervisé.e par : Louis Doray Université de Montréal Calcul de rentes et primes d’assurance-vie individuelles avec la Covid-19 ### Michael Vladovsky Supervisé.e par : Vincent-Philippe Lavallée Centre hospitalier universitaire Mère-Enfant (CHU Sainte-Justine) Analyse comparative du séquençage du transcriptome et de l’exome pour le diagnostic des leucémies myéloïdes aiguës. ### Anton Volniansky Supervisé.e par : Alexandre Dumais Centre intégré universitaire de santé et de services sociaux de l’Est-de-l’Île-de-Montréal (CIUSSS - EMTL) Hôpital Maisonneuve-Rosemont et Institut universitaire mentale de Montréal) Le Traitement Automatique du Langage humain comme outil de prédiction de la réponse à la Thérapie Avatar pour traiter les hallucinations de la schizophrénie résistante ### Masters excellence scholarships ### Benjamin Akera Supervisé.e par : David Rolnick McGill University Learning Global Embeddings for social good – A case study of Remote Sensing and Citizen Science “Despite the growth of initiatives for monitoring life on earth, there continue to be vast gaps in data and knowledge across modalities. To bridge this gap, we learn a global ecosystem embedding to approximate terrestrial biodiversity at different regions on a geographical scale. This embedding will utilize data from different sources, including remote sensing and citizen science observations of animals and plants. The combination of unlabelled satellite imagery and sparse information on biodiversity will allow us to establish a framework for learning global embeddings in ecology which can be extended to other societal challenges such as climate change and food security in regions where little or no data might exist.” ### Ève Campeau-Poirier Supervisé.e par : Laurence Perreault Levasseur Université de Montréal Modélisation de lentilles gravitationnelles à l’aide d’une machine à inférences récurrentes (RIM)) “Les deux méthodes employées jusqu’à maintenant pour déterminer le taux d’expansion de l’univers génèrent des résultats différents. L’une a recours au décalage de la lumière des étoiles locales, c’est-à-dire à l’état récent de l’univers. L’autre emploie les assertions du modèle standard de cosmologie ainsi que les variations de température du fond diffus cosmologique, soit un rayonnement électromagnétique issu de l’univers primordial. Il est important de comprendre l’origine de ce désaccord, entre autres, pour s’assurer de la validité de nos présentes connaissances sur l’évolution et sur la structure de l’univers. Une troisième méthode de mesure indépendante pourrait permettre de confirmer une des valeurs, et donc, de nous renseigner sur l’erreur qui cause l’écart. Il en existe une qui requiert l’observation d’un vaste échantillon de lentilles gravitationnelles. Celles-­ci sont des systèmes formés de deux galaxies dont l’une est appelée source, et l’autre, lentille. La galaxie lentille se trouve entre la Terre et la galaxie source. Dû à des effets gravitationnels, la lentille dévie la lumière émise par la source, déformant l’image de la source que reçoit la Terre. Pour évaluer le taux d’expansion de l’univers avec ces systèmes, il faut modéliser la distribution de masse de la lentille. Or, il n’existe pas de relation qui procure la distribution de masse de la lentille à partir de l’image déformée de la source. La procédure actuelle consiste à simuler des images déformées à partir de plusieurs distributions de masse différentes et à retenir celle qui engendre l’image simulée ressemblant le plus à l’image réelle. Ce procédé est lent et coûteux en ressources de calculs. Notre projet vise à découvrir la relation de la distribution de masse en fonction de l’image déformée. Pour ce faire, nous utiliserons une machine à inférences récurrentes (RIM), soit une méthode d’apprentissage automatique qui produit un algorithme d’inférence. En entrainant un réseau de neurones avec cette méthode, il apprendra le processus d’extraction de la distribution de masse à partir de l’image déformée. Ce processus pourra alors être appliqué à tout nouveau système de lentille de façon rapide et efficace. Étant donné l’important volume d’images attendu de la part de la nouvelle génération de télescopes, cela permettra d’évaluer le taux d’expansion de l’univers dans un délai raisonnable. De plus, les RIM n’ont jamais été testées sur des problèmes non-linéaires comme celui-ci. Ce projet pourrait donc ouvrir un éventail de possibilités pour leurs applications futures.” ### Clémentine Courdi Supervisé.e par : Éric Lacourse Université de Montréal Prédiction des profils de santé physique et mentales des mères canadiennes à partir de leur profil de facteurs de vulnérabilité par l’analyse de classes latentes “L’objectif de la recherche est de déterminer si des profils de facteurs de risque, incluant par exemple l’âge, le niveau d’éducation, le revenu et le statut d’immigration, affectent le profil de santé des mères canadiennes, incluant la santé physique et mentale. La santé maternelle est au cœur de ce projet, définie comme incluant tous les aspects (autant physiques que mentaux) de la santé de la femme, avant et pendant la grossesse, jusqu’à l’accouchement et la période post­partum dans les mois suivants. En investiguant quels sont les facteurs et plus précisément les profils de risques associés à des problèmes de santé physique et mentale chez les mères, on espère trouver des stratégies de santé publique permettant d’encourager une meilleure santé chez les mères, et par conséquent chez les enfants. Pour étudier le lien entre les profils de risques et les profils de santé des mères, des techniques d’analyse inspirée de l’intelligence artificielle vont être utilisées. Plusieurs recherches ont étudié auparavant des facteurs de risque précis, comme le fait d’être une mère adolescente, et leurs conséquences sur la santé maternelle. Ce projet se distingue par le fait qu’une grande quantité de facteurs de risque seront pris en compte simultanément pour vérifier comment ils agissent en synergie pour affecter la santé maternelle. Dans le même ordre d’idées, aucun problème de santé en particulier n’est étudié; c’est l’état général de santé physique et mentale avant, pendant et après la grossesse qui sera analysé. Cette analyse sera conduite sur les données issues de l’Enquête sur la santé dans les collectivités canadiennes (ESCC), précisément celles des années 2010, 2014 et 2018. Le projet comprend donc également un volet comparatif visant à évaluer l’évolution de la situation.” ### Ali Fakhri Supervisé.e par : James Goulet Polytechnique Montréal Bayesian Dynamic Linear Models for Structural Health Monitoring “Roads, bridges, and dams are critical components of the modern infrastructure system. Their deterioration from ageing, usage, and environmental exposure may require costly repairs or even result in catastrophic failure. This can negatively impact the economy or even lead to loss of life. A large portion of Canadian infrastructure is old or in bad condition. Subsequently, this poses risks to the Canadian economy and public safety. Research in Structural Health Monitoring (SHM) offers a promising path towards addressing this issue. Namely, we can detect changes in structural behaviour using the data captured by sensors placed on the structure. This will allow us to monitor the structure’s condition in real-time, and any unusual behaviour can trigger preventative actions (e.g., field inspections and repairs). There have been considerable advances in SHM methods in recent years. In particular, Bayesian Dynamic Linear Models (BDLMs) were shown to be very effective in detecting anomalies in the context of SHM. Nonetheless, much remains to be done before a BDLM framework can be deployed to a large population of different structures. Specifically, the existing methods rely on expert judgement and prior data analysis to prepare the model. Thus, suboptimal selection of parameters can greatly hinder the performance. Alternative parameter selection methods must be investigated to improve the performance and make the models applicable to a wide range of structures. The existing model capabilities also need to be extended to include and interpret multiple data sources and account for non-periodic patterns in data. This project aims to investigate the presented research gaps and further advance the BDLMs for SHM, getting us closer towards managing the ageing civil infrastructure via a large-scale deployment of a generic low-cost SHM system.” ### Alexandre Hudon Supervisé.e par : Alexandre Dumais Centre intégré universitaire de santé et de services sociaux de l’Est-de-l’Île-de-Montréal (CIUSSS - EMTL) Hôpital Maisonneuve-Rosemont et Institut universitaire mentale de Montréal) Traitement Automatique du Langage humain comme outil pour soutenir le thérapeute dans l’évolution clinique d’un patient dans le cadre d’une Thérapie Avatar pour soigner les hallucinations résiduelles de la schizophrénie résistante “En cette ère de la médecine de précision, l’un des défis fondamentaux de la recherche en psychiatrie est d’améliorer nos modèles de prédiction de l’évolution clinique des patients. Dans le présent projet, nous souhaitons mettre sur pied une procédure automatisée d’analyse du discours des patients atteints d’un trouble mental afin de prédire la réponse au traitement à la suite d’une séance de thérapie et d’adapter conséquemment la séance suivante. Le projet vise à développer une procédure automatisée d’analyse du discours (Traitement Automatisé du Langage-TAL) des patients engagés dans une Thérapie Avatar (thérapie dialogique), qui, à l’aide de la réalité virtuelle, permet au patient d’entrer en dialogue avec une hallucination négative représentée par un avatar. Sur le plan opérationnel, la procédure automatisée permettra également de transformer les enregistrements audios (des séances de thérapie) en texte écrit, de coder automatiquement les unités du discours à l’aide de l’algorithme, puis d’appliquer des modèles statistiques tirés de l’apprentissage machine afin d’identifier des features discriminant les bons et les mauvais répondeurs, et ainsi adapter la thérapie pour suivre la trajectoire des bons répondeurs. La pertinence de ce projet, par son côté novateur et avant-gardiste, permet d’ouvrir la porte sur la possibilité de prédire adéquatement la réponse clinique d’un patient à une thérapie basé sur un champ lexical, c’est-à-dire sur les dialogues avec son thérapeute. Ce concept pourrait être utilisé non seulement pour la Thérapie Avatar, mais être extrapolé pour une panoplie de psychothérapies.” ### Pascal Laferriere-Langlois Supervisé.e par : Nadia Lahrichi Polytechnique Montréal Prédiction des variations de tension artérielle chez le patient subissant une chirurgie “Ce projet a comme objectif d’améliorer les outils de monitoring dont dispose le médecin lorsqu’il administre une anesthésie à un patient subissant une chirurgie. Nous souhaitons intégrer l’ensemble des informations disponibles pour prédire l’évolution de la tension artérielle du patient opéré au cours des prochaines minutes. Nous savons déjà que de brèves périodes de tension artérielle trop élevée (hypertension) ou trop basse (hypotension) est délétère pour le patient et augmente son risque de complications post-opératoires. En prédisant à l’avance ces épisodes d’hypotension ou d’hypertension, nous aiderons le clinicien à maintenir la tension artérielle dans un intervalle sécuritaire. Pour y parvenir, nous développerons des algorithmes bâtis sur d’anciens patients ayant subis des variations de tension artérielle. En collaborant avec l’Université de Californie à Los Angeles (UCLA), nous aurons accès à des banques de méga-données de patients opérés et nous pourrons analyser les dossiers médicaux pour analyser et identifier quels antécédents médicaux influencent la dynamique de la tension artérielle. Nous pourrons également analyser l’évolution dans le temps des signes vitaux et des courbes physiologiques de ces patients opéré, afin d’identifier quelles caractéristiques peuvent nous aider à prédire les variations de tension artérielle. En utilisant une séquence d’analyse par apprentissage supervisé et non supervisé, nous pourrons trouvé quels paramètres sont importants et ainsi bâtir nos algorithmes, pour ensuite tester la qualité de leur prédiction sur des banques de patients. En intégrant ces algorithmes au monitoring du patient opéré, nous pourrons faciliter le maintien de la tension artérielle et possiblement réduire les complications du patient.” ### Robin Legault Supervisé.e par : Emma Frejinger Université de Montréal Modèles de capture de flot et simulation stochastique pour réseaux congestionnés avec usagers hétérogènes “La transition à la voiture électrique est un axe phare des plans de lutte aux changements climatiques mis de l’avant par les États du monde entier. Sa réalisabilité à large échelle est cependant conditionnelle à une accessibilité accrue aux installations de recharge. Cette considération peut être formalisée au moyen de modèles de capture de flot, une famille de problèmes d’optimisation qui consiste à maximiser la couverture, par les installations d’un décideur, des entités se déplaçant dans un système donné. Le problème de capture de flot tel qu’étudié dans la littérature repose toutefois sur des hypothèses simplificatrices empêchant son application adéquate aux problèmes routiers de grande taille, soit la fluidité du réseau et le comportement uniforme des usagers. Des travaux récents ont permis de formaliser l’hypothèse plus réaliste selon laquelle les choix des voyageurs sont dictés par des préférences variées et aléatoires, mais ces modèles requièrent la simulation d’un nombre important de scénarios, une tâche computationnellement exigeante limitant la taille des problèmes pouvant être considérés. Le projet consiste ainsi à développer un modèle de capture de flot réaliste, efficacement résolvable et applicable au problème de la localisation des bornes électriques sur des réseaux de taille réelle. Trois axes seront étudiés pour concrétiser cet objectif, soit les techniques de réduction de variance permettant l’identification d’une solution optimale en présence d’usagers hétérogènes par la simulation d’un nombre minimal de scénarios, la combinaison des approches de résolution principales explorées séparément dans la littérature du problème de capture de flot, puis l’application de ces méthodes au cadre plus général des réseaux de transport congestionnés. Le modèle développé sera appliqué à des données réelles décrivant les réseaux routiers de grandes villes nord-américaines.” ### Xing Han Lu Supervisé.e par : Siva Reddy McGill University Explainable and Faithful Models for Question Answering In the past few years, various deep learning architectures have shown a great ability in automatically answering free-form questions. In fact, they are able to match human-level performance when we evaluate them using popular benchmark datasets. Unfortunately, those datasets take a lot of time and are very expensive to collect, since they require a team of annotators to manually write a question and find the correct answer. Furthermore, deep learning models that are trained on such datasets cannot be improved once they are online. Additionally, the models can find an answer that’s likely correct, but they cannot explain why they chose it. To overcome those issues, we propose a framework that lets those models continuously improve their ability to answer questions even after they are online, and can learn from user feedback to explain why a certain answer is relevant when you give them a question. The proposed framework will be useful in emergency situations (such as the COVID-19 pandemic), since we can quickly build a question answering system that is capable of improving itself and learn to explain why its answers are correct, both of which would not be possible using traditional approaches. ### Sarthak Mittal Supervisé.e par : Guillaume Lajoie Université de Montréal Multiple Faces of Modularity There has been a lot of research that incorporate different flavors of modular­ity and sparsity in typical Neural Network models to endow them with inductive biases that are inspired from human cognition. We analyze one such type of model, Recurrent Independent Mechanisms (RIMs), that aims to do both semantic as well as episodic factorization. We investigate the key properties utilized by these modules in an effort to understand exactly what properties are essential for their generalization capacity. In particular, we perform ablations relating to the following properties – independence, de-centralized organization, communication and try to understand to what extent do the different modules in RIMs specialize based on semantic information. We believe that understanding the limitations of these models would pave the way to future research on improving and scaling them up. ### Sacha Morin Supervisé.e par : Guy Wolf Université de Montréal Geometry preserving deep networks Artificial intelligence models learn by minimizing an objective function on a given data set. Objective functions are tailored to specific tasks, such as classifying samples, generating new images or reducing the dimensionality of the data. Their design is critical to model performance, especially when considering how well a model generalizes to previously unseen data points that were not used for training. This project concerns the design of objective functions for deep learning algorithms. Specifically, we aim to study geometric objective functions for deep learning, i.e. learning objectives that consider some measure of distance between data points to preserve the intrinsic geometry of the data. Deep learning is known for dramatically changing the structure of the data when learning new representations of it. For example, data points forming a circle may be embedded as a closed curve with many self intersections. Previous work has shown that encouraging deep neural networks to preserve data geometry could be beneficial for the task of dimensionality reduction and, to some degree, classification. We aim to show similar benefits for new tasks, such as the generation of synthetic data points, and further explore the mathematical theory that could explain the superior performance of deep neural networks with geometry preserving properties. ### Justine Pepin Supervisé.e par : Margarida Carvalho Université de Montréal Jeux de programmation en nombres entiers : approches pour la sélection des équilibres corrélés “Afin d’optimiser leur bénéfice individuel, plusieurs joueurs d’un jeu de programmation en nombres entiers tentent de jouer la combinaison de stratégies qui les avantagera le plus en anticipant le comportement des autres. Conséquemment, la solution du jeu se trouvera en un point d’équilibre à partir duquel aucun joueur ne gagnerait à dévier unilatéralement de la combinaison de stratégies qu’il joue actuellement. Un équilibre peut être corrélé, c’est-à-dire qu’un coordinateur oriente le choix de la combinaison de stratégies des joueurs en s’assurant que chaque joueur aura intérêt à suivre ses recommandations. Dans de nombreux jeux d’intérêt pratique, de multiples équilibres peuvent exister. Il en découle qu’une telle coordination peut être cruciale afin que les joueurs s’entendent sur un équilibre efficace, par exemple sur celui qui optimise le bien-être social. Cependant, comment s’assurer qu’un équilibre corrélé optimise l’objectif choisi? Pour cela, il faudrait disposer d’un algorithme certifiant que nous avons le meilleur point d’équilibre. Nous proposerons une approche algorithmique inspirée des méthodologies classiques de décomposition en optimisation.” ### Emmanuelle Richer Supervisé.e par : Farida Cheriet Polytechnique Montréal Segmentation d’images volumiques de la rétine par apprentissage profond “Les maladies oculaires telles que la dégénérescence maculaire liée à l’âge, le glaucome et la rétinopathie diabétique sont la cause principale de la cécité chez la population active. Avec le vieillissement de la population et la prévalence croissante du diabète, on prévoit qu’en 2025, 333 millions de patients atteints de diabète à travers le monde vont avoir besoin d’un examen ophtalmologique chaque année. L’apprentissage profond a déjà fait ses preuves dans le diagnostic automatique de pathologies à partir d’images médicales. Des architectures de réseaux de neurones convolutionnels ont déjà été proposés pour, par exemple, la segmentation de vaisseaux rétiniens et du disque optique ainsi que la détection de ces trois maladies, à partir d’une banque d’images du fond d’œil. Même si l’image de fond d’œil représente la modalité la plus utilisée en clinique, les images volumiques de la rétine acquises à l’aide de la tomographie par cohérence optique (OCT) sont souvent requises afin de confirmer un diagnostic. Un des inconvénients de l’utilisation de réseaux de neurones pour la prédiction et diagnostic de maladies est l’interprétabilité des résultats par les cliniciens. La segmentation préalable de biomarqueurs est donc un atout lors de l’entraînement de tels réseaux. Dans le cadre de ce projet, nous allons développer un modèle d’apprentissage profond afin de segmenter automatiquement les biomarqueurs associés aux différentes maladies oculaires à partir des images OCT. Ce modèle sera contraint par un modèle préalablement entraîné sur des images du fond d’œil. Ainsi, la segmentation des images OCT sera guidée par la segmentation des lésions sur des images de fond d’œil obtenue par apprentissage profond. Les deux cartes de segmentation obtenues à partir des deux modalités seront par la suite soumises à un classifieur afin d’établir le diagnostic final. Le projet proposé permet d’exploiter les avancées récentes en apprentissage profond pour une analyse automatique des images volumiques de la rétine. Une détection automatique de ces trois maladies permettra de prioriser les patients à haut risque afin d’éviter l’évolution vers un stade de pathologie irréversible. La prise en charge de ces patients à temps entrainera des bénéfices à notre système de santé en augmentant l’efficacité des protocoles préventifs et en réduisant le coût des traitements.” ### Maria Sadikov Supervisé.e par : Julie Hlavacek-Larrondo Université de Montréal Comprendre les plus gros trous noirs de l’univers en utilisant des techniques innovatrices d’apprentissage automatique “Les avancées des récentes années dans le domaine de l’apprentissage automatique révolutionnent la recherche dans une multitude de domaines, que ce soit en physique, en biologie ou en économie. Ces algorithmes permettent d’analyser et d’interpréter de très larges ensembles de données de manière plus rapide et plus efficace qu’auparavant, nous donnant accès à une nouvelle compréhension du monde. Notre objectif est d’utiliser des méthodes innovatrices d’apprentissage automatique afin de pousser plus loin notre compréhension des objets les plus extrêmes de l’univers : les plus gros trous noirs supermassifs situés au centre d’amas de galaxies. Ces amas de galaxies contiennent un gaz intra-amas, qui émet d’énormes quantités d’énergie sous forme de rayons-X. On s’attendrait à ce que cette perte d’énergie résulte en un refroidissement rapide de l’amas, mais ce processus est contrebalancé par les jets relativistes éjectés par le trou noir supermassif, qui causent des perturbations dans le milieu intra-amas. Pour pouvoir mieux caractériser ces trous noirs supermassifs, il est impératif d’avoir une bonne compréhension des propriétés du milieu intra-amas ainsi que de l’évolution de ces systèmes. Ces propriétés ont déjà été étudiées pour des amas de galaxies proches avec des méthodes traditionnelles. Cependant, le développement de nouveaux instruments d’observation extrêmement puissants nous permet d’obtenir pour la première fois des larges échantillons d’amas lointains. La taille de ces échantillons, ainsi que le grand nombre de paramètres à considérer, rend nécessaire l’utilisation de méthodes avancées d’apprentissage automatique. L’objectif du projet est donc de développer des algorithmes d’apprentissage automatique permettant l’étude de ces amas de galaxies. Les modèles seront entraînés puis validés sur des ensembles d’images obtenues à l’aide de simulations cosmologiques, avant d’être appliqués sur un échantillon d’amas de galaxies ayant déjà été analysé avec des méthodes traditionnelles. Le but est de reproduire et d’améliorer ces résultats en employant des algorithmes de classification d’images qui repéreront des structures additionnelles. Nous cherchons ainsi à découvrir l’information supplémentaire obtenue à l’aide des méthodes d’apprentissage automatique, nous permettant notamment de trouver des indicateurs de l’impact du trou noir supermassif sur le gaz environnant. “ ### Jérôme St-Jean Supervisé.e par : Dang Khoa Nguyen Centre hospitalier de l’Université de Montréal (CHUM) Vêtement intelligent Hexoskin et intelligence artificielle : Détection des crises d’épilepsie Le but de ce projet de recherche est de développer des méthodes de détection de crises d’épilepsie basées sur des signaux multimodaux enregistrés avec des chandails intelligents fournis par notre partenaire industriel Hexoskin. Les objectifs spécifiques sont : 1) Caractériser les signaux physiologiques en comparant les périodes ictales (en état de crise) et les périodes interictales (entre les crises). 2) Développer un algorithme de détection des périodes ictales avec des techniques d’intelligence artificielle (IA). 3) Développer un algorithme permettant la détection en temps réel d’une crise d’épilepsie. ### Yuanyuan Tao Supervisé.e par : Derek Nowrouzezahrai McGill University Physics-aware deep learning for cellular dynamics “A physics-aware deep learning system is developed to almost instantaneously infer cellular dynamics with minimum experimental procedure. People are paying heavy prices for diseases that would otherwise be diagnosed and treated much more easily and cheaply by employing cell mechanics. Cellular forces dictate cellular processes and the onset and progression of diseases such as cancer and asthma. The use of cellular dynamics as biomarkers and modulators for cell behaviour indicates the potential of cell mechanics in diagnosis, treatment, drug development, and the study of disease mechanisms. However, the application of cell mechanics is hindered by the costly, time-consuming, and complex procedures to measure cellular forces. Current methods examining cellular forces measure the deformation directly resulted from the forces and then calculate the forces back from the deformation. However, those methods completely depend on in-vivo experiments, so cannot go beyond the parameter space in the experiments, and often require complex and tedious procedures. Sometimes, the accuracy is limited by solving inverse problems. This project develops a physics-aware deep learning system to apply cell mechanics to diagnosis, treatment, and drug development with no additional procedures, time, and cost. To integrate digital intelligence into biology and medicine, this project searches for the best way to model the physical interaction between the cell and environment with deep learning. To resolve the limits of the current methods and to revolutionize the methodology in the field of cell mechanics, a comprehensive and generalizable physics-aware deep learning system with adjustable parameters is developed to accurately and almost instantaneously infer and simulate the dynamics in cells and tissues under different conditions from merely the time series of cell morphology. ConvLSTMs and transformers can be used to capture the features and spatiotemporal relations in the morphology. Physical equations, for example, in solid mechanics, and biological parameters are embedded into neural networks optimized for the physical laws and biological models. Our approach is also much more sustainable as minimum lab waste is produced. This system provides revolutionary insights and utility in drug development, diagnosis, treatment, and researches. Moreover, the application of this project exceeds cellular dynamics as the fundament of the project is modelling physical interactions and problems with deep learning.” ### Doctoral excellence scholarships ### Nicolas Cabrera Malik Supervisé.e par : Mendoza-Gimenez Jorge HEC Montréal Multi-modal vehicle routing problems Avec les progrès de l’informatique et la complexification croissante des modèles utilisés dans l’industrie, de nombreux problèmes d’ingénierie ne peuvent plus être approchés à l’aide des méthodes de l’optimisation classique. Les fonctions définissant ces types de problème sont des boîtes noires, c’est-à-dire des simulations numériques ou des codes informatiques comportant des entrées réglables par l’utilisateur et retournant une ou plusieurs sorties. L’évaluation de ces fonctions pour des paramètres d’entrée donnés peut être coûteuse, voire approximative, et les dérivées ne sont pas disponibles. Les techniques d’optimisation classique (par exemple celles s’appuyant sur les gradients) ne peuvent donc s’appliquer. De nombreux logiciels pour résoudre ce type de problèmes ont été développés. Parmi eux, le logiciel NOMAD implémente un algorithme à l’état de l’art, MADS. Cette méthode est efficace et possède de bonnes propriétés de convergence. L’objectif général de ce projet consiste à développer de nouveaux algorithmes en optimisation multiobjectif, où plusieurs objectifs contradictoires doivent être pris en compte dans la modélisation d’un problème donné, basés sur la méthode MADS possédant des propriétés de convergence similaires, avec un temps de calcul pour un problème donné plus court que les méthodes à l’état de l’art existantes. Les méthodes développées seront intégrées au logiciel open source NOMAD et testées sur des modèles de simulation concrets (problèmes de dimensionnement de moteur). Ces travaux de recherche ont des retombées importantes dans les domaines du génie ; en chimie (conception de réseaux ou l’optimisation de réactions chimiques) ; l’apprentissage machine est également concerné avec des problèmes de classification et de partitionnement de données. ### Veronica Chelu Supervisé.e par : Doina Precup McGill University Retrospective Causal Models for Credit Assignment Learning fast from a few samples of interaction is a fundamental skill for AI systems that capable of making robust good decisions. This ability is a core component of human intelligence and of autonomous agents adaptable to change. In many societal applications in areas such as healthcare and education, AI can help guide interventions by quickly searching through the space of effective strategies for the best decision policy to support human health, learning or performance. To this effect, the paradigm of reinforcement learning proposes a general solution framework by which agents learn through experience to make good choices. These methods are however limited in how quickly they can learn, typically requiring millions of trials of experience for learning to make adequate decisions. One core aspect of cognition is using knowledge of how the world works to explain observations, to imagine what could have happened that did not, or what could be true that is not. Humans are remarkably apt at inferring (or fabricating) causes for events, and of retrospectively updating their beliefs about the world in hindsight of experience. These mechanisms stand at the basis of counterfactual reasoning and are crucial components in tackling real-world problems with long-delayed feedback. Many questions in everyday life and scientific inquiry are causal in nature, particularly in areas such as healthcare — e.g. “What if I had administered the patient a different drug?”. In this project, we address the problem of efficient learning by interaction with an environment through trial and error. To tackle this problem we propose learning algorithms that build causal models to represent the fundamental workings of the world and apply said models to retrospectively infer the potential causes of events so that they can efficiently re-evaluate their beliefs. ### Pierluca D’Oro Supervisé.e par : Pierre-Luc Bacon Université de Montréal Sample-Efficient Reinforcement Learning via Metacognition “Global transformations such as new emerging epidemics and climate change are generating unprecedented challenges for humanity, making a clear call for effective artificial intelligence and control methods. Recent progress in reinforcement learning, the study of how an agent can learn to maximize a utility function while interacting with a system, makes it particularly promising, but with some caveats: while most previous successes needed the collection of considerable amounts of experience to solve a task, the nature of these new challenges requires very sample-efficient algorithms, able to rapidly learn from few interactions with the world. One of the most effective tools employed by humans during learning is metacognition, or the ability to think about thinking. The goal of the project is to create a new generation of reinforcement learning algorithms that, by injecting into the agent the ability to reason in a deeper way about its own training process, are able to increase their efficiency. This can happen by arming the agent both with curiosity towards the experiences that are maximally useful for improving its performance and with an understanding of how to employ the collected experience in a way that is efficient according to its learning aptitudes.” ### Dirk Douwes-Schultz Supervisé.e par : Alexandra Schmidt McGill University Coupled Markov Switching Count Models for Spatio-temporal Infectious Disease Counts “Accurate epidemic forecasting of infectious diseases remains a major challenge. A state-of-the-art class of statistical models known as “Markov Switching Models” have shown promise in this area. This approach breaks up epidemic forecasting into two components. Firstly, there is a component to predict epidemic occurrence, i.e. when an epidemic will begin in an area. Since an infectious disease will often sit at low levels of incidence, or be completely absent, for extended periods, being able to predict when an epidemic will begin is the first vital step to epidemic forecasting. The second component of the Markov switching model forecasts the resulting cases in the epidemic period. This component can predict when the epidemic will peak, how many cases are expected, how long it will last and other important metrics for the epidemic period. To summarize, Markov switching models predict when an epidemic will occur and then forecast the resulting trajectory of the cases. We offer some novel extensions to these models in order to improve forecasting performance. Firstly, we will allow seasonal, meteorological, socioeconomic and other factors to impact our predictions of epidemic occurrence. In contrast, previous approaches assumed a constant probability of epidemic occurrence. This assumption is not very appropriate as epidemics of an infectious disease are known to be highly seasonal and influenced by a wide range of factors. For example, an epidemic of dengue fever will almost never occur in the winter when temperature is too low for prolonged mosquito survival. A model not accounting for temperature in this case would give poor predictions of epidemic occurrence. Incorporating space-time varying predictions of epidemic occurrence into our modeling framework should improve forecasting performance considerably. Secondly, we will incorporate realistic human movement into our forecasts, even using cell phone data where available. Essentially, our framework allows for epidemics in an area to spread into other areas connected by high movement flows. Human movement to and from epidemic areas has been shown to be a major risk factor for the development of epidemics and we view this as an essential component of our modeling strategy. Our extensions create statistical challenges compared to more tradition Markov switching models that have been used by others. To overcome this, we will develop new fast and efficient algorithms for fitting the model. All model fitting software will be made publicly available and we plan on writing subsequent less technical papers to introduce policy makers to our methods.” ### Andreas Enzenhoefer Supervisé.e par : James Richard Forbes McGill University Robot learning using accurate dynamics simulation with frictional contact The objective of the proposed research is to develop an accurate dynamics simulation framework that will be used to generate control policies for autonomous robots. Discrepancies between the virtual environment and the real world will be reduced through accurate simulations and domain randomization of material parameters and initial conditions. An analytically differentiable contact model will be developed to allow for sample efficient model-based reinforcement learning. This contact model will be extended to include anisotropic and asymmetric friction critical for the accurate and realistic simulation of many surfaces and material types. Novel generative models for syntheses of simulation scenarios for robotic training and learning will be created. This also includes a simulation plausibility verification to prevent unrealistic or unphysical effects in the training data. The proposed research will significantly decrease the computational cost of controller optimization and improve the transferability of learned control policies to the real robot (“simulation to reality”) through more accurate and validated simulations. This will contribute to accelerating the technological readiness of autonomous robots such as self-driving vehicles. ### El Mehdi ER RAQABI Supervisé.e par : Issmaïl EL HALLAOUI Polytechnique Montréal Decomposition Learning: An Intelligent Framework For Large Scale Optimization Problems “With the growing human population, an increasing demand is emerging for several needs such as food, transport, and healthcare. With such a trend, many public, private, profit, and non-profit organizations are investing energy and time to serve our population. Several of them are facing large-scale problems in operations scheduling, budget allocation, resources assignment, etc. Given their size, it is not possible to solve these problems manually. Furthermore, sometimes, even using computers, no feasible solutions are rapidly found. By developing algorithms, which decompose the large-scale problem into smaller problems and solve them repetitively, there is an opportunity to tackle it efficiently. With such observation, the proposed research project aims to develop more efficient mathematical optimization algorithms that smartly decompose the large-scale problem after automatically selecting a suitable decomposition. For this purpose, in a world-class lab where researchers/entrepreneurs are gathering, I plan to bring the fields of Mathematical Optimization (MO) and Artificial Intelligence (AI) together to develop systems able to solve large scale problems. The AI component will analyze the data of the problem as well as the solutions of similar instances before providing useful information permitting to speed-up the decision algorithms. This research project will foster Canada’s research and industrial clusters. Furthermore, I believe in the potential of commercializing the proposed solution worldwide through a spin-off that could create dozens of high-quality jobs in Canada. The framework can tackle various large-scale problems like industrial production, supply chain management, airline planning, public transportation, urban planning, retail businesses, agriculture planning, and healthcare management. By pushing boundaries in MO and AI, I seek the reinforcement of Canada’s leadership in these two fields while contributing to humans’ wellbeing anytime and anywhere on earth. “ ### Simon Faghel-Soubeyrand Supervisé.e par : Frédéric Gosselin Université de Montréal Décoder les variations d’habileté perceptive dans la population en utilisant l’imagerie cérébrale et l’apprentissage machine L’expertise avec laquelle l’humain extrait de l’information sur l’identité, l’état émotionnel et le genre des visages de ses pairs est cruciale pour tout individu vivant en société. Le cerveau humain typique effectue ces opérations apparemment sans effort, à l’échelle du dixième de seconde. Par contre, il s’avère que l’habileté en reconnaissance faciale n’est pas aussi homogène qu’on a pu le croire : la performance individuelle varie considérablement dans la population, avec certains individus que l’on nomme “prosopagnosiques développementaux” n’ayant aucune lésion cérébrale mais étant incapable de reconnaître leur collègues ou leurs proches. D’autres, nommés “super-recognisers”, sont au contraire capables de se souvenir d’un visage vu une seule fois des années plus tôt. Étonnamment, les mécanismes cérébraux qui sous-tendent ces variations d’habiletés ont été sous-explorés, et restent largement inconnus. Une compréhension des mécanismes qui causent ces variations est centrale pour implémenter des programmes d’entraînement pouvant améliorer l’habileté des individus avec des troubles perceptifs et sociaux (incluant ceux sur le spectre de l’autisme et de la schizophrénie). L’objectif de ce projet doctoral est d’utiliser l’apprentissage automatique et l’imagerie cérébrale afin de modéliser le cerveau d’individus qui sont extraordinairement habiles, moyens, et déficitaires en reconnaissance faciale. Nous révélerons d’abord comment (par.ex. la géométrie des représentations, les computations effectuées à différents moments) un cerveau perceptif optimal devrait se comporter pour reconnaître les visages des individus de son entourage. Nous développerons ensuite des modèles qui imiteront le comportement du système visuel optimal et déficitaire en reconnaissance faciale. Ces modèles pourront nous informer sur des mécanismes précis sur lesquels agir afin de réduire l’impact des troubles perceptifs d’individus prosopagnosiques (par. ex. renforcer l’activité cérébrale d’une région spécifique pour mieux reconnaître l’identité). Ultimement, nous souhaitons créer des modèles d’entraînement spécifiques à différents troubles perceptifs comme les troubles sur le spectre de la schizophrénie ou de l’autisme. Cette caractérisation formelle du code cérébral derrière l’habileté visuelle exceptionnelle pourra potentiellement inspirer de nouveaux modèles d’apprentissage profond, qui sont en ce moment performants en reconnaissance d’objets/visages mais non-robustes à des changements visuels minimes dans les images. ### Jose Gallego-Posada Supervisé.e par : Simon Lacoste-Julien Université de Montréal Towards a Geometric Theory of Information “Information theory is a highly developed and active research area and has been of paramount importance in the development of modern machine learning techniques. However, this field was originally developed in the framework of random variables taking values on mere discrete sets of symbols. This austerity results in a blindness to additional structure amongst the symbols, which limits the power and applicability of the theory. My long-term vision is to build a generalization of the theory of information developed by Shannon in a way that directly incorporates the geometric structure in the domains of the random variables. Tools from information theory have been used in the context of representation learning to understand the “surprising” generalization properties of deep learning systems. Expanding our understanding on why deep neural networks perform well on unseen examples, and the potential role that its learned representations play in this process, is a key step towards the deployment of deep learning-based systems in applications for which performance guarantees are critical. However, one of the challenges faced by these approaches arises from the invariance of the mutual information between two random variables with respect to smooth invertible transformations of their sample spaces. My proposal aims at providing machine learning researchers with theoretical tools to tackle these challenges. In previous work, we have imported a notion of similarity-sensitive entropy originally developed in theoretical ecology to the machine learning community. Based on this definition, we propose geometry-aware counterparts for several concepts and results in standard information theory, as well as a novel notion of divergence which incorporates the geometry of the space when comparing probability distributions while avoiding the computational challenges of optimal transport distances. In the future, my research will focus on the theoretical and practical implications of these ideas: 1) can we obtain an axiomatic characterization of geometry-sensitive entropy?; 2) are geometric mutual information objectives better behaved for representation learning?; 3) what are the connections between our proposed divergence and rate-distortion theory, in particular, regarding deep-learning based compression techniques; 4) can we improve the entropic regularization used in reinforcement learning to encourage exploration by considering similarities on the action space?” ### Dóra Jámbor Supervisé.e par : Siva Reddy McGill University Zero-shot Natural Language to SQL translation “Much of the world’s knowledge is stored in relational databases. To access this knowledge, however, users have to express their questions in Structured Query Language, i.e., SQL programs. This causes a severe bottleneck in efficiency across many organizations, as the vast majority of key decision-makers are not fluent in SQL. Natural language to SQL models (Text2SQL) have emerged to automatically translate questions expressed in plain English to SQL programs that can then be readily executed against a given database. Although there has been great progress in recent years with Transformer-based Text2SQL models, it is still challenging to generalize in the zero-shot setting where models have to generate SQL programs for previously unseen databases. In this work, we propose a novel procedure to fuse structural and semantic signals to find better alignment between a given question and database pair. Specifically, our hypothesis is that by explicitly modeling how information flows through a database graph, we can better capture the semantics of the database entities. We believe that these richer semantics can consequently improve how we map more complex lexical and structural references in a given question to databases. We believe that our structure-aware alignment can help models generate semantically valid SQL programs not only for known databases but also for previously unseen databases, thereby helping zero-shot generalization. “ ### Sékou-Oumar Kaba Supervisé.e par : Siamak Ravanbakhsh McGill University Equivariant Deep Models for Materials Property Prediction “Materials discovery is a key driver of technological innovation, especially now that environmental and sustainability constraints are core priorities. To answer this challenge, materials informatics is emerging as a new field making use of the increasing availability of experimental and computational data on materials. A fundamental problem in this area is to create algorithms that can be trained on a large number of already known materials to predict the properties of previously unseen materials. A trained algorithm would have the advantage of making estimates of desired properties much faster than the currently available methods based on physical simulation. The goal of the proposed project is to design an algorithm able to perform such predictions using deep learning. Although deep learning methods have demonstrated their power in multiple fields, they have seen less use in materials modeling. One crucial reason is that seen at the atomic level, a solid-state material is an extremely large structure that is difficult to input to a deep learning model. To tackle this problem, we will leverage the symmetry properties of crystals. These materials are composed of atoms arranged in ordered structures, a feature we can use to build efficient models. We will assess the performance of our architecture on the Materials Project dataset, a collection of more than one hundred thousand materials for which properties have been computed with quantum mechanical methods. If the model achieves satisfying results, it will be made available for practical applications.” ### Sanaz Kaviani Supervisé.e par : Jean-François Carrier Université de Montréal Enhancement of quantitative estimation of metabolism and vascularization with positron emission tomography (PET) and Ultrafast Ultrasound Localization Microscopy (UULM) Using Deep Learning “1 Problem and context Structural and functional imaging of tissue vasculature has been studied using various imaging modalities such as positron emission tomography (PET), magnetic resonance imaging (MRI), computed tomography (CT), and ultrasound. Among all the molecular imaging modalities, no single modality is perfect and sufficient to obtain all the necessary information for any questions of interest. A recent novel technique inspired by super-resolved fluorescence microscopy called ultrasound localization microscopy (ULM), improved spatial resolution of vascular images, from hundreds to a few microns in vivo via the detection at thousands of frames per second of millions of individual microbubbles injected in the bloodstream. Our group (led by my co-supervisor J. Provost, IVADO member) has recently shown the feasibility of extending ULM to dynamic acquisitions in three dimensions using novel imaging sequences and reconstruction algorithms. With the introduction of deep learning algorithms, research focusing on multimodality medical imaging has increased exponentially, such as image segmentation, denoising, and image reconstruction. In this work, we propose to combine ULM with ultra-resolution images and dynamic PET imaging to estimate parametric images of dynamic PET based on deep learning, using compartment pharmacokinetic modeling. Moreover, I aim to enhance the quality of dynamic PET images and extraction of perfusion and vasculature parameters to have a precise model of tissue behavior with denoising and precise segmentation. 2 Methodology In this project, we aim to combine the microvascular information from 3D ULM to the molecular information of dynamic PET imaging in order to enhance the resolution and quantification of PET dynamic acquisitions. The first step is an iterative parametric image reconstruction using a deep neural network. I will not use prior training pairs, but only the same ULM image. I will utilize the ULM images from the same image as anatomical prior (blood content in every voxel) to guiding the parametric image reconstruction through the neural network. The neural network will be inserted into the iterative parametric image reconstruction framework and pharmacokinetic modeling to achieve more precise kinetic parameters, rather than using it as a post-processing tool. The second step of the project will be on denoising of dynamic PET images because of the high-level noises of these images. Generally, deep learning with convolutional neural networks (CNN), requires the preparation of large training image datasets. This presents a challenge in a clinical setting because it would be very difficult to prepare large, high-quality datasets. Recently, the deep image prior (DIP) approach suggests that CNN structures have an intrinsic ability to solve inverse problems such as denoising without any pre-training. The DIP approach iterates learning using a pair of random noise and corrupted images and a denoised image is obtained by the network output with moderate iterations. The third step will be the segmentation of PET images, which will be organ detection based on unsupervised learning. The main idea for this is that better image representation gets better clustering, and better clustering results helps to get better image representation. 3 Results Modified network structures are developed based on the 3D U-net for each step which consists of an “encode” part and a “decode” part. The encode part of architecture consisting of the repetitive applications of 3D convolution layers, each followed by a batch normalization (BN) and a leaky rectified linear unit (LReLU), and convolutional layers for downsampling. The decoding part consists of a deconvolution layer, followed by the BN and the LReLU, transposed convolutional layers for up-sampling, skip connection with the corresponding linked feature map from the encoding part. In each step, the network structure and the loss function are modified to have the best performance in that step. We will extract perfusion and vasculature parameters to have a precise model of tissue behavior. The extracted parametrical maps non-invasively depict beneficiary information about tissue microvasculature and are used as input in the PET compartment pharmacokinetic model. The technique will be used on pre-clinical dynamic data for small animal microPET and on numerical phantoms for dynamic quantitative analysis, validation and sensitivity study. Applications for human PET in nuclear medicine – including tumor microenvironment parametrization – will be developed. “ ### Jordan Lei Supervisé.e par : Eilif Muller Unraveling the Mystery of Neocortical Learning with Deep Neural Networks The human brain manifests a remarkable capacity for learning across a broad range of modalities and contexts. This feat is made possible in large part by the neocortex, a brain structure composed of a repeating layered neuronal circuit motif shared across modalities, and even across mammalian species.The neocortex is capable of learning in ways that modern deep learning systems struggle with. For example, the neocortex can learn from only a few training examples, generalize to different task conditions, and learn representations and associations without explicit instruction, all of which are open problems in the field of machine learning and artificial intelligence. Deep neural networks, originally inspired by the architecture of the neocortex, provide a powerful framework for modeling neocortical learning algorithms. In this project, we will build on differential target propagation (DTP), a promising family of neocortical learning models, to develop refined learning paradigms that account for recent neuroscientific insights into neocortical anatomy, synaptic plasticity and perceptual phenomena.Our work aims to unravel the mystery of neocortical learning, and will have direct implications for the fields of neuroscience, artificial intelligence, data science, and mental health. ### Bruna Pascual Dias Supervisé.e par : Jean-François Arguin Université de Montréal Identification d’électrons à l’aide des réseaux antagonistes génératifs (GAN) dans l’expérience ATLAS Localisée autour de l’anneau du Grand collisionneur de hadrons (LHC) au CERN, l’expérience ATLAS est conçue pour enregistrer les signaux provenant d’un milliard des collisions entre protons par seconde. Les algorithmes utilisés pour l’identification des particules produites par ces collisions sont aussi impliqués dans les simulations du détecteur, responsables pour estimer son efficacité et comparer notre connaissance actuelle des lois physiques avec les mesures expérimentales mesurées. Par contre, des imperfections dans cette simulation introduisent de grandes incertitudes systématiques dans les mesures expérimentales et donc limitent la qualité des résultats obtenus par le détecteur. Dans ce contexte, ce projet vise à développer et mettre en application des réseaux antagonistes génératifs (GAN) pour créer un algorithme d’identification d’électrons insensible aux perturbations des paramètres qu’introduisent ces imperfections. Ceci nous permettra de minimiser les incertitudes systématiques introduites par la différence de performance de l’algorithme d’identification d’électrons dans les simulations et les mesures expérimentales, ce que promet impacter positivement l’avenir de l’expérience ATLAS. ### Kellin Pelrine Supervisé.e par : Reihaneh Rabbany McGill University Broad Data Systems for Society: Leveraging Heterogeneous Data for Social Good “Today’s world is increasingly technological and interconnected, and we are in the era of big data. But although big data has enabled many breakthroughs, there are also many challenges we face as a society that have proven too complex to be solved with large quantities of data alone. For example, to counter misinformation or reduce political polarization, a dataset of millions of tweets is not enough – we need to leverage the full spectrum of our social interactions and more. We need systems that use not just big data but broad data. This project aims to develop such systems and apply them for social good. In work so far, with J. Danovitch and R. Rabbany, I laid a thorough empirical foundation by showing that current benchmarking of misinformation detection algorithms is flawed, and leads to models that fail to extract the potential of the broad data they try to use. In current work, I am analyzing political polarization with the full breadth of social media interactions, to understand causes, changes over time, and how to unite people for common good. I am also working to develop large scale models that combine text, images, and social interactions to power a wide range of social network research and applications. In future work, I will write a dissertation on broad data in general, aiming to go beyond existing tools and individual applications and show how to use it to solve the challenges of tomorrow.” ### Brice Rauby Supervisé.e par : Jean Provost Polytechnique Montréal Angiographie Quantitative du Myocarde par Localisation Ultrasonore En 2015, 110 millions de personnes étaient atteintes d’insuffisance coronarienne (CAD) et près de 9 millions en mouraient ce qui en faisait la première cause de mort dans le monde. L’imagerie cardiaque est souvent la première étape dans le diagnostic et la planification du traitement de ce groupe de maladies. Cependant, les angiographies conventionnelles ne permettent pas de mesurer le débit sanguin d’une manière directe qui soit non-invasive et largement disponible. Par ailleurs, les méthodes de microscopie de localisation ultrasonore (ULM) se sont développées et rendent possible une cartographie à haute résolution et non-invasive du débit sanguin sur des organes statiques. Dans ce cadre, l’application de méthode d’apprentissage automatique a montré une amélioration significative des résolutions temporelle et spatiale. L’objectif de ce projet est de transférer les méthodes d’ULM des organes statiques vers le muscle cardiaque grâce à des méthodes de correction de mouvement ayant montré des résultats prometteurs et en adaptant les méthodes d’apprentissage automatique pour bénéficier d’améliorations en résolution. ### Sima Rishmawi Supervisé.e par : Frederick Gosselin Polytechnique Montréal Digital Twin of a Rotating Machine: Model Order Reduction and Artificial Intelligence for Hydroelectricity Production “NASA defines a Digital Twin as an integrated multiphysics, multiscale, probabilistic simulation of an as-built vehicle or system that uses the best available physical models, sensor updates, fleet history, etc., to mirror the life of its corresponding flying twin.” This project aims at implementing a digital twin of a Vertical Axis Rotating Machine (VARM) exhibiting some similar physical characteristics to a hydroelectric machine. Some of these characteristics are represented in the dynamic behavior of both machines such as vibrations. Also, similar parameters can be measured and monitored, and similar loading conditions can be applied to both machines. This means that studying the dynamic behavior of a VARM can simplify the process of understanding the dynamic behavior of an industrial hydroelectric unit. To create the digital twin, two numerical methods will be used; the first is model order reduction using Proper Generalized Decomposition (PGD), which can be thought of as a method that transforms a multi-dimensional complicated system into a group of simple one-dimensional systems that are easier to work with. The second method is to use Physics Informed Neural Networks (PINN) in order to predict the parameters needed to create the system model by Artificial Intelligence (AI). “ ### Camille Rochefort-Boulanger Supervisé.e par : Julie Hussin Apprentissage profond en génomique pour la prédiction de phénotypes complexes Plusieurs traits humains comme la taille, ainsi que diverses maladies telles que la schizophrénie et les maladies cardiovasculaires sont qualifiés de complexes: leur manifestation est modulée par l’interaction de facteurs environnementaux et de nombreux variants génétiques. Les récents progrès en génomique humaine ont généré l’espoir de pouvoir prédire ces traits complexes à partir des nombreuses données collectées. Des scores de risque génétique ont été développés pour prédire la prédisposition des individus à des traits complexes à partir de leurs variants génétiques. Toutefois, ces scores ont un pouvoir de prédiction limité, surtout lorsqu’ils sont calculés sur des individus provenant de populations différentes, car ils sont biaisés par l’ethnicité génétique des individus utilisés pour la construction initiale de ces scores. De par leur capacité à considérer les interactions entre de multiples facteurs, les méthodes d’apprentissage profond, une branche de l’intelligence artificielle, représentent un moyen prometteur d’accomplir des progrès dans ce domaine. Mon projet consiste à développer une nouvelle méthode d’apprentissage profond tirant profit des données génomiques pour prédire des traits complexes. Toutefois, la quantité volumineuse de variants génétiques identifiés chez l’humain, ainsi que la qualité variable des jeux de données génomiques posent plusieurs défis à l’application des méthodes d’apprentissage profond actuelles. Dans le cadre de ce projet, je développerai des approches adaptées aux particularités des données génomiques et robustes à l’ethnicité des individus afin de réduire le biais causé par la diversité ethnique sur les prédictions de traits complexes. Ce projet aura un impact important dans le domaine de l’apprentissage profond, car il vise à développer des algorithmes adaptés aux données génomiques, ainsi que dans le domaine de la génomique, puisqu’il vise à améliorer la prédiction de traits complexes en prenant en compte la diversité ethnique génétique. ### Julien Roy Supervisé.e par : Christopher Pal Polytechnique Montréal Weakly Supervised Behavioral Modeling for Sequential Decision Making Reinforcement Learning (RL) is a popular approach to sequential decision making problems, and is a core component of many recent successes in Artificial Intelligence ranging from predicting the 3D structure of proteins to mastering the game of Go. However, most modern approaches focus on learning a single optimal policy for a given task. This property makes applying RL in the industry cumbersome as any desired change in the agent’s behavior requires tuning the reward function and training the policy all over again. In this project, we present and study a novel framework to train multi-modal, conditional policies that allow to capture a variety of near-optimal behaviors. Crucially, our method would give practitioners the capability to tune and adjust the policy at test-time, without re-training the agent. We hope that this contribution will take us closer to the deployment of more practical RL algorithms to tackle the real life problems of today and to better support the fast changing requirements of the industry. ### Ludovic Salomon Supervisé.e par : Sébastien Le Digabel Polytechnique Montréal Optimisation multiobjectifs de boîtes noires sous contraintes générales “Avec les progrès de l’informatique et la complexification croissante des modèles utilisés dans l’industrie, de nombreux problèmes d’ingénierie ne peuvent plus être approchés à l’aide des méthodes de l’optimisation classique. Les fonctions définissant ces types de problème sont des boîtes noires, c’est-à-dire des simulations numériques ou des codes informatiques comportant des entrées réglables par l’utilisateur et retournant une ou plusieurs sorties. L’évaluation de ces fonctions pour des paramètres d’entrée donnés peut être coûteuse, voire approximative, et les dérivées ne sont pas disponibles. Les techniques d’optimisation classique (par exemple celles s’appuyant sur les gradients) ne peuvent donc s’appliquer. De nombreux logiciels pour résoudre ce type de problèmes ont été développés. Parmi eux, le logiciel NOMAD implémente un algorithme à l’état de l’art, MADS. Cette méthode est efficace et possède de bonnes propriétés de convergence. L’objectif général de ce projet consiste à développer de nouveaux algorithmes en optimisation multiobjectif, où plusieurs objectifs contradictoires doivent être pris en compte dans la modélisation d’un problème donné, basés sur la méthode MADS possédant des propriétés de convergence similaires, avec un temps de calcul pour un problème donné plus court que les méthodes à l’état de l’art existantes. Les méthodes développées seront intégrées au logiciel open source NOMAD et testées sur des modèles de simulation concrets (problèmes de dimensionnement de moteur). Ces travaux de recherche ont des retombées importantes dans les domaines du génie ; en chimie (conception de réseaux ou l’optimisation de réactions chimiques); l’apprentissage machine est également concerné avec des problèmes de classification et de partitionnement de données.” ### Harsh Satija Supervisé.e par : Joelle Pineau McGill University Batch Policy Improvement with Multiple Objectives and Safety Constraints “My current goal is to design RL algorithms that can improve the agent’s performance from a batch of data collected under some behaviour policy, while still ensuring the safety guarantees. I’m interested in the setting where there are multiple feedback signals (reward functions), and the algorithm’s user (ML Practitioner) can control the trade-offs between them. This setting is also reflective of the real-world scenarios where there is an abundance of recorded data, collected via experts or suboptimal agents, but training the agent directly via interactions and experimentation with the real environment is expensive and risky, such as health-care, hiring, or finance. The goal is to build algorithms that provide practical high-probability guarantees about the undesirable behaviour that might be caused by deploying the policy returned by the learning algorithm in the real-world. “ ### Beheshteh Tolouei Rakhshan Supervisé.e par : Guillaume Rabusseau Université de Montréal Randomized numerical linear algebra approaches with tensor methods My research is focused on algorithms, and complexity results for problems arising in machine learning and data science. This broadly includes contributions in tensor factorization, randomized matrix computations methods, and theoretical computer science. In past and ongoing research, I have been specifically interested in developing algorithms with provable guarantees, accurate and fast solutions to computationally expensive methods by leveraging dimensionality reduction and tensor decomposition techniques. This will be broadly applicable in machine learning and artificial intelligence problems in areas ranging from human diseases to climate change to recent technological developments. My goal as a researcher is to contribute to both the theoretical understanding of computationally challenging problems, as well as the design of efficient techniques for large-scale high-dimensional data. In doing so I plan to help bridge the gap between the best theoretical results and the most practical algorithms for complex problems in machine learning. The main point of my PhD project is to explore and apply both randomized algorithms and tensor decomposition techniques to large-scale data sets. Significant expected outcomes include computationally fast methods for big data problems that cannot be solved in real time. ### Baptiste Toussaint Supervisé.e par : Maxime Raison Polytechnique Montréal Développement d’un membre supérieur robotique autonome pour la collaboration en réadaptation “La robotique est un marché mondial d’environ 40 millions de dollars en 2020 en forte croissance. Couplée à l’intelligence artificielle et à l’internet des objets, la robotisation affectera tous les secteurs de l’économie. Le Technopole en réadaptation pédiatrique est l’un des leaders dans le domaine, avec un nouveau bâtiment (2019) avec infrastructure thérapeutique principalement composée de systèmes robotiques et impression 3D pour plusieurs applications dont l’interaction et le jeu intelligent (de réadaptation) avec le patient. Cependant, pour ce type applications, le constat du problème est similaire. Bien que l’utilité des bras robotiques dans la réadaptation ait déjà été démontrée, leur utilisation n’est pas démocratisée. En effet, les bras commercialisés sont chers et non-adaptables. De plus, ces robots manquent souvent de dextérité, rapidité et précision. L’objectif de ce projet est donc de développer un robot humanoïde du membre supérieur intelligent, autonome et interactif en réadaptation. Les données et l’intelligence artificielle auront un rôle principal pour les trois sous-objectifs : 1. Augmenter la dextérité du robot pour l’accomplissement de tâches du membre supérieur, à l’aide d’apprentissage machine sur des bases de données de mouvements de la balle et du robot ; 2. Augmenter les performances de rapidité et de précision du robot, à l’aide de notre contrôleur quadratique linéaire itératif (iLQR) renforcé par réseau neuronal (NNiLQR) ; 3. Valider les deux premiers objectifs au travers de démonstrations appliquées uniques au monde, tel qu’un record du monde d’échanges de ping-pong entre humain et robot, tâche qui combine la dextérité, la rapidité et la précision.” ### Internship grants: Data to tell ### Ali Akbar Sabzi Dizajyekan Polytechnique Montréal Stage chez Wikimédia Canada, spécialité science des données ### Katharine O’Brien Concordia Stage chez Synapse-C, spécialité communication ### Laurence Taschereau UQAM Stage chez Wikimédia Canada, spécialité communication ### Simon-Olivier Laperrière Université de Montréal Stage chez Le Devoir, spécialité science des données ### Khaoula Chehbouni HEC Montréal Stage chez La Presse, spécialité science des données ### Clara Gepner Concordia Stage chez La Presse, spécialité communication ### Paul Fontaine Université Laval Stage chez Le Devoir, spécialité communication ### Isabelle Bouchard Polytechnique Montréal Stage chez Radio-Canada, spécialité science des données ### Postdoc-entrepreneur program ### Ehsan Moradi Supervised by: Luis Miranda-Moreno McGill University Project: CarboRate Stepping beyond the conventional energy and emissions assessment tools, “CarboRate” takes a multi-modal agent-based approach for estimation, tracking, and evaluation of transportation energy, carbon, and air-pollution footprint while targeting both the commercial and non-commercial markets. ### Daniel Pereira Supervised by: professor Louis-Martin Rousseau Polytechnique Montréal Startup: Matrius Technologie Matrius has developed a revolutionary alternative to the current approach to scheduled infrastructure maintenance shutdowns: the first ultrasonic non-destructive testing probes that operate fully exposed at temperatures up to 600°C (1112°F) and enable continuous monitoring of active infrastructure. The funding will enable the development of AI-based software to analyze the data stream generated by the sensors. The goal is to create AI-based predictive models that identify accelerated corrosion, improve maintenance planning, and aid in long-term decision making. ### Postdoctoral research funding ### Homa Arab Supervised by: Steven Dufour Polytechnique Montréal A Deep-Learning Method for Arrhythmias Detection Using a Millimeter-Wave MIMO Wireless multiple-input multiple-output (MIMO) radar sensors are attracting increased attention because of their capability to measure the angle-of-arrival (AoA), they have a larger signal-to-noise ratio (SNR), and they allow lower minimum detectable speeds. They can be used in a wide range of applications found in day-to-day life, such as for contactless vital sign detection, sleep monitoring, human fall detection, and smart surveillance and security systems. These systems have the capability to receive long term data from various receivers to detect tiny movements of one or multiple objects. Due to the noise and interference found in RX signals, extracting the desired information from received signals from nonstationary targets, multiple targets, and various RX antennas signals, is an important and challenging task. The focus of this research is to remove noise and random body movements from measured heartbeat and respiration signals by using convolutional neural network (CNN) encoder and LSTM multilayer decoders. LSTM network will also be applied to detect and cut out interference in the spectrograms of the signals. Furthermore, a coarse-to-fine generative adversarial network (GAN) and a CNN-LSTM Encoder Decoder will be used to restore the part of the spectrogram that is affected by the interferences. Denoised signals will be fed into a simple CNN network to carry out the final classification. ### Natasha Clarke Supervised by: AmanPreet Badhwar Institut universitaire de gériatrie de Montréal (IUGM) Machine learning based insights into the relationship between cerebrovascular pathology and brain functional connectivity in Alzheimer’s disease Alzheimer’s disease (AD) is a devastating disorder of the brain. The hallmark pathology is a build-up of abnormal forms of two proteins, amyloid and tau. Despite efforts to develop drugs that clear these deposits from the brain, there is currently no treatment for AD. One reason for this is that most patients also have damage to other brain components, such as the blood vessels, and that this damage also contributes to cognitive impairment. Our project will explore the interplay between (1) damage to the brain vascular system, and (2) brain connectivity, a measure of how well different parts of the brain communicate. Both can be assessed using magnetic resonance imaging (MRI). We will use machine learning to analyse MRI scans from a large study (UK Biobank) to identify subgroups of people with similar connectivity patterns, and then determine which of these subgroups are associated with damage to the brain vascular system. Then, we will determine which of these vascular damage-associated subgroups are related to clinical features of AD, and whether these multi-pathology MRI markers better predict future decline. Our findings will enable more precise treatments for people with AD, and better patient selection for clinical trials. ### Jonathan Cornford Supervised by: Richards Blake McGill University Exploring learning in neural networks with brain-inspired geometries There are many differences between artificial and biological neural networks. However, based on the fundamental assumption that biological neural networks have been optimized by evolution, the two fields have long shared a synergistic relationship. A simple question we might therefore ask is: “How similar are the parameter update rules that govern learning in biological and artificial networks?”. In AI, neural networks are generally trained to minimize empirical risk via stochastic gradient descent (SGD). As such, network parameters are updated additively with the negative gradient of the loss at every training iteration. In contrast, recent biological experiments have shown that synaptic weight updates in the brain are predominantly multiplicative in nature. In this research proposal we consider how these two forms of update can arise from the choice of distance generating function in Mirror descent, and propose to leverage our recent work building networks with sign-constrained weights to explore the use of multiplicative updates and non-euclidean distance functions for training artificial neural networks. ### Istvan David Supervised by: Eugene Syriani Université de Montréal Inference of simulation models in Digital Twins by reinforcement learning Digital Twins are virtual representations of physical assets, providing a proxy towards applications needing to access data on the physical asset. Simulators are one of those applications, extensively employed in Digital Twins to support real-time decision making. Due to the enormous complexity of the systems subject to digital twinning, constructing simulators by hand is an error-prone and costly endeavor, that automation can significantly improve. This project provides a framework for the inference of simulation models in Digital Twins by reinforcement learning. Specifically, we aim at the inference of Discrete Event System Specification (DEVS) models. DEVS is a versatile simulation formalism that has been show to be the common denominator of other simulation formalisms. The algorithm starts from system-specific architectural templates of DEVS models, and learns their dynamic elements: state-transition and timing functions. The learning mechanism is supported by a-priori formalized feedback provided by the environment. We will use a data set of 130.000 real sensor records, comprising 7 distinct metrics, and validate the scalability of the developed technique by applying it to a larger dataset of 250M-1.5B records from 100 sensors. ### Thang Doan Supervised by: Joelle Pineau McGill University Enabling Zero-Shot transfer in Reinforcement Learning through dynamic augmentation Reinforcement Learning (RL) has made great strides in recent years, facilitating the creation of artificial agents that can learn to solve a wide array of complex tasks, including robotic manipulation and challenging games (Atari, Go), purely from pixel inputs. However RL agents are still fragile when deployed in environments they were not directly trained on: small changes to the observations or environment dynamics often result in dramatic drops in performance. This project aims to develop a new method that will create RL agents that are robust to novel environment dynamics. As a first step, we learn a latent space of environment dynamics. We then apply data augmentation in the latent subspace, which will train the network to extrapolate to nearby environments. In effect, we train the agent on a space of “imaginary” environments, close to the training environment, which should make the agent robust when faced novel environments with unseen dynamics during evaluation. ### Ali Falaki Supervised by: Numa Dancause Université de Montréal Using artificial intelligence to personalize transcranial magnetic stimulation parameters In recent years, repetitive transcranial magnetic stimulation (rTMS) has shown great promise as a potential therapy for neurological and psychiatric disorders. rTMS is a safe non-invasive neurostimulation method to induce long-lasting changes in the brain via a wire coil that generates magnetic fields passing through the scalp. However, its clinical application is limited mainly because of the variability of induced responses across individuals. To address this variability, clinicians should be able to rapidly select the parameters of the stimulation, such as the frequency or the intensity, based on the responses evoked in a given individual. But these factors are too complex to study one by one manually. Our general goal is to develop an intelligent machine learning algorithm that effectively selects the optimal parameters of the stimulation based on each subject’s response. After this step, we will adapt this approach to create a user-friendly and flexible interface that can be used clinically. ### Jean-Pierre Glouzon Supervised by: Martin Smith Université de Montréal Clustering of human transcriptome for real-time gene expression profiling Identifying expression profiles from deep RNA sequencing (RNA-seq) analysis is a promising tool for precision medicine that can refine disease aetiology, improve risk stratification and increase diagnostic precision. However, the lengthy turnaround time required to generate and analyze data using next-generation sequencing technologies severely limit the diagnostic potential of RNAs-seq. There is thus a need for efficient and accurate RNA-seq technologies combined with gene expression pipelines and models to accelerate clinical decisions and improve patient management. We propose an approach to gene expression profiling in real-time based on an efficient representation and online clustering of transcript sequenced via Nanopore RNAseq. Our model leverages real-time raw directed RNA signals generated by Nanopore RNAseq to build accurate and efficient gene expression and transcriptome profiles avoiding error-prone based-called data analysis. ### Alex Hernandez-Garcia Supervised by: David Rolnick McGill University Deep learning for material discovery to fight climate change The current and expected consequences of climate change driven by anthropogenic greenhouse gas emissions are a major threat for humanity and, more generally, for the biodiversity and stability of the planet. Developing strategies of adaptation to and mitigation of these effects is thus of utmost importance. One of the ways in which artificial intelligence can help fight climate change is by accelerating scientific discoveries. In this multidisciplinary project, we propose to combine machine learning and chemistry to discover new materials to improve energy storage, optimize the energy from renewable sources or capture carbon dioxide, by leveraging the potential of deep neural networks to find the most promising materials among large sets of candidates. Progress in this direction will not only help reduce carbon emissions but also advance fundamental aspects of machine learning science. ### Xu Ji Supervised by: Yoshua Bengio Université de Montréal Generalization in Neural Networks Why is it that humans can switch tasks and apply knowledge learned in the first whilst learning the second, accurately and without forgetting, whereas neural network models cannot? Why do networks require orders of magnitude more training data to learn what a human can be taught with a single experience? Why is it that networks will still misclassify an image of a dog as a lawnmower, despite seeing thousands of examples of both? These questions relate to the fundamental problem of generalization, which is perhaps the most pressing problem currently facing the development of human-level intelligence. Without an effective solution, neural networks cannot learn in real time from naturally sequential real world data as effectively as humans do, which limits their performance, deployment options, and training efficiency. Generalization is therefore a problem with immense practical implications, as well as being interesting theoretically and from the perspective of biologically-inspired learning. This project seeks to understand and solve the questions posed above by finding new neural network training procedures. ### Katarzyna Jurewicz Supervised by: Becket Ebitz Université de Montréal Data-driven discovery of continual learning algorithms from neural populations Natural environments change unpredictably, and require continual learning, a skill that remains challenging even in cutting-edge artificial intelligence (AI). However, biological decision-makers have solved this problem. They have evolved algorithms that balance the reliable exploitation of well-tried options with the need to continue to explore and learn about alternatives. This is called the stability-flexibility dilemma and is a major open question in artificial intelligence and neuroscience. The goal of this project is to leverage large-scale neural datasets and machine learning methods to identify the biological algorithms for resolving the stability-flexibility dilemma. Understanding these algorithms could reveal basic neural mechanisms for cognitive flexibility, impact our understanding of diseases in which flexibility is compromised (e.g. addiction, depression, and obsessive compulsive disorder), and inform next-generation AI. ### Jacob Miller Supervised by: Guillaume Rabusseau Université de Montréal Structured language modeling with recurrent tensor networks Recent deep learning architectures have enabled massive improvements in the quality of neural language models, to the point that state-of-the-art models are now capable of generating essays, short computer programs, and even poetry with seemingly near-human quality. While this progress is impressive, further research has revealed important limitations in the high-level understanding of such models, largely arising from a difficulty in capturing certain syntactic and semantic structures present in natural language. To circumvent these limitations, this project proposes to make use of the novel capabilities of a new family of generative models built on quantum-inspired tensor networks. Small tensor network (TN) language models have already been shown capable of efficiently generating text which conforms to the structure of any user-specified grammar, and we aim to assess this capability in real-world settings using large-scale TN models trained on large language corpora. Beyond the immediate utility of these methods for source code generation, this will facilitate the development of similar methods for the automatic extraction of syntactic structure present in pretrained TN models. Such methods would open the door to transparent methods for capturing high-level semantic information, bringing such language models closer to genuine language understanding. ### Ashraf Uz Zaman Patwary Supervised by: Francesco Ciari Polytechnique Montréal Development of An Efficient Gradient Estimation Technique for Large-scale Traffic Assignment Model Optimization Intelligent transportation engineering is moving towards the implementation of sensor-embedded, connected vehicles and infrastructure, generating a large variety of complementary datasets in the process. While the machine learning literature does provide different methods for utilizing individual datasets for accurate predictions, large-scale, network-wide traffic assignment (TA) models can truly utilize the complementarity of these datasets while maintaining the interpretability and infrastructure of the underlying physical process. However, optimization of large-scale TA models is severely hindered by the curse of dimensionality, undesirable mathematical properties, and expensive function evaluations. To alleviate this problem, we propose to develop an efficient gradient estimation algorithm named iterative backpropagation (IB), for solving multi-source, high dimensional, large-scale, agent-based TA calibration or optimization problems. IB is inspired by the popular backpropagation through time (BPTT) algorithm used in recurrent neural network (RNN) training. It exploits the iterative structure of the TA solution procedure and simultaneously calculates the gradients while the TA process converges. IB requires no additional function evaluation and consequently, scales very well with higher dimensions. Having a similar structure as BPTT, IB is highly parallelizable and can benefit from the existing machine learning literature and implementations, ensuring sustainability and efficiency improvement of the algorithm in the long run. ### Janarthanan Rajendran Supervised by: Sarath Chandar Anbil Parthipan Polytechnique Montréal Towards Lifelong Reinforcement Learning Agents Lifelong learning agents that continually learn throughout their lifetime accumulate knowledge to achieve their current goals efficiently and prepare for future ones. Their wide ranging applications include dialog systems, autonomous vehicles, and household robots. In this project, we will focus on lifelong Reinforcement Learning (RL) agents which have long lifetimes and act in vast, complex, and non-stationary environments with very sparse and delayed rewards. Unlike most current RL agents, lifelong RL agents have a single start to their lifetime without any resets. They can visit only a small fraction of all the states in their environment, and their actions can have irreversible effects. Most of our current solution methods which focus on simpler settings are not effective in this setting. We propose to work on addressing some of the key challenges posed by lifelong learning to RL: 1) Discovering what to learn by learning General Value Functions (GVF), 2) Gathering relevant experience needed to learn by learning options guided by learned intrinsic motivation, 3) Learning and accumulating knowledge effectively by using modular and memory-augmented networks for GVF learning, and 4) Using the learned knowledge to infer more knowledge and achieve its goals by building approximate and abstract models. ### Jimin Rhim Supervised by: Derek Nowrouzezahrai McGill University Building human-AI trust for the long term: in the context of frictionless retail In April 2021, Canada proposed a federal budget of443.8 million for the nation’s artificial intelligence (AI) economy., among which, $185 million will support the commercialization of AI innovations. Contrary to the huge investment, only 44% of Canadians trust AI and robotics, making Canada one of the least trusting nations for the AI industry. If left unaddressed, the lack of user trust poses a major threat to Canada’s future prosperity with AI. For Canadians to take full advantage of the potential benefits of AI, it is integral to build human-AI trust. Consequently, the goal of the proposed research is to develop a temporal dynamic model of trust between human and AI by exploring humans’ behavioral patterns and expectations during repeated interactions with an AI system. The newly established McGill Retail Innovation Lab — a living lab featuring a frictionless convenience store — will be used to observed how trust is formed, maintained, and lost during human-AI interaction in the wild. In this proposed project, I will model the temporal dynamic of user trust in AI based on the accumulated data (e.g., navigation patterns, purchased products, visiting time, interaction duration, drop out rates, interaction with smart agents). Then, the effectiveness of the developed temporal trust dynamic model will be tested in the frictionless retail setting. The AI will prosper and provide its potential when the public’s trust in AI is formed. A new, temporal model of human-AI trust developed using the abovementioned longitudinal investigation will challenge the current rudimentary model of trust. The empirically validated model will subsequently inform how automation systems should be designed in the future with the long-term view of user trust in mind. This, in turn, will directly inform industry leaders and policymakers with practical means to help lift the distrust in AI and robotics in Canada and abroad. ### Hajime Shimao Supervised by: Maxime Cohen McGill University Implications of “Fairness” in Fair Machine Learning The issues surrounding fairness in prediction results generated by machine learning (ML) algorithms has attracted enormous interest from researchers in recent years. While numerous algorithms have been proposed and they can provide prediction results that are fair based on certain notions, the current literature lacks an understanding of potential impacts of these predictions on the behavior of prediction subjects. Their behavior in turn influences the data generating process for the future predictive task; therefore the iterative dynamics must be investigated to fully understand the social implication of fair-ML. The purpose of this research project is to examine the implications of fairness when a fair ML algorithm is used in realistic settings. This research project leverages a unique synergy that combines cross-discipline expertise in ML, economics, operation research, and information systems, which are strongly relevant to the problem at-hand. The overarching goal of this research project is to design a fair ML algorithm that takes into account the behavior of prediction subjects and welfare of relevant stakeholders when the algorithm is used in realistic settings. ### Bénédicte L. Tremblay Supervised by: Julie Hussin Université de Montréal L’apprentissage profond et les sciences omiques dans le combat contre l’infarctus du myocarde L’infarctus du myocarde (IM) résulte du blocage d’une artère du cœur, ce qui provoque la destruction d’une partie du muscle cardiaque. Les sciences omiques, qui étudient l’ensemble complexe des molécules qui composent le corps, permettent de mieux comprendre les maladies, dont l’IM. Par ailleurs, des modèles issus de l’intelligence artificielle (IA) ont le potentiel de prédire le risque de certaines maladies, dont l’IM. Toutefois, ces modèles sont difficiles à interpréter puisqu’ils fonctionnent comme des « boîtes noires » : ils génèrent des prédictions ou des recommandations, sans fournir d’explications ni de justifications. Or, l’interprétabilité biologique du modèle est essentielle pour le développement d’applications cliniques. Quelques études font état de l’interprétabilité biologique de méthodes d’IA, mais très peu concernent la santé. L’objectif de ce projet est d’évaluer l’interprétabilité biologique d’une nouvelle méthode d’IA qui utilise des données omiques pour prédire le risque d’IM. Le projet compte 2000 participants, dont 1000 avec des antécédents d’IM et 1000 sans antécédents d’IM. Ce projet permettra de mieux prédire le risque d’IM et, ultimement, d’aider au développement de programmes de prévention et de traitement. L’interprétabilité biologique de la méthode renforcera la confiance de la communauté médicale face à l’IA et favorisera son application en clinique. ### Yuan Yang Supervised by: Jerome Le Ny Polytechnique Montréal Learning for Haptic Shared Control in Human-Robot Teams The rapid advancements of decision support systems and autonomous systems are making robots more involved in physical collaboration with human teammates. In order to work alongside people safely and efficiently, robots need to correctly comprehend the time-varying states of their human partners and to appropriately adapt their behaviors in response. A promising solution to this challenge is haptic shared control systems that organize haptic interactions and share control authority between human and robot teammates by transferring suitable resistive/assistive forces to human operators as per their psychophysical status. For optimal haptic shared control, this project proposes to develop data-driven strategies to estimate quantitative models of human operators and to develop plug-and-play algorithms to coordinate a dynamic network of mobile robots in collaboration with human teammates. Specifically, we plan to leverage reinforcement learning techniques to identify certain passivity properties of human operators from our collected experimental data, and then synthesize the learned models into passivity-based human-robot teaming algorithms to improve the navigation and motion performance of robots. This project will contribute data-driven approaches to the control of transportation and logistics systems with humans in the loop. ### Undergraduate research initiation grants ### 1st edition ### Samuel Arseneault Supervisé.e par : Chantal Labbé HEC Montréal Générateur de données pour favoriser l’apprentissage en science des données. ### Kristina Atanasova Supervisé.e par : Alexandre Prat Université de Montréal Identification des interactions cellulaires de la barrière hémato-encéphalique par apprentissage automatique. ### Sol’Abraham Castaneda Ouellet Supervisé.e par : Didier Jutras-Aswad Université de Montréal Les effets du cannabidiol sur la cognition chez les personnes atteintes d’une dépendance à la cocaïne. ### Simon Chasles Supervisé.e par : François Major Université de Montréal Caractérisation de motifs structurels surreprésentés dans l’ARN. ### Laurence Beauregard Supervisé.e par : Dang Khoa Nguyen Université de Montréal Détection de crises épileptiques en combinant les techniques d’intelligence artificielle et les signaux physiologiques non-invasifs multimodaux. ### David Chemaly Supervisé.e par : Julie Hlavacek-Larrondo Université de Montréal State-of-the-art radio images of the Coma Cluster of Galaxies. ### Omar Chikhar Supervisé.e par : Julie Hlavacek-Larrondo Université de Montréal A novel machine learning approach to identifying cool core clusters. ### Léo Choinière Supervisé.e par : Numa Dancause Université de Montréal Hierarchical Bayesian Optimization for Stimulation Protocols in Cortical Neuroprostheses. ### Céline Boegler Supervisé.e par : Dominique Orban Polytechnique Montréal La méthode des résidus conjugués pour l’optimisation sans contraintes. ### Karl-Étienne Bolduc Supervisé.e par : Normand Mousseau Université de Montréal Développement de potentiels atomiques par apprentissage machine. ### Ariane Brucher Supervisé.e par : Phaedra Royle Université de Montréal Analyses statistiques de potentiels évoqués basées sur des modèles mixtes linéaires. ### Hugo Cordeau Supervisé.e par : Vasia Panousi Université de Montréal Utilisation des données satellitaires ainsi que de donnés de recensements afin de connaitre les éléments d’étalement urbain. ### Olivier Denis Supervisé.e par : Jean-François Arguin Université de Montréal Apprentissage machine pour l’analyse des données du LHC. ### Samuel Desmarais Supervisé.e par : Jean Provost Polytechnique Montréal Conception d’un algorithme de reconstruction d’images ultrasonores utilisant l’apprentissage profond afin de réduire le nombre de canaux nécessaire à l’obtention d’une image à contraste équivalent. ### Cauderic Deroy Supervisé.e par : Sébastien Hétu Université de Montréal Identification des signatures physiologiques de la violation des normes sociales. ### Guillaume Dupuis Supervisé.e par : Brunilde Sansò Polytechnique Montréal Projet sur le 5G et les villes intelligentes. ### Charlie Gauthier Supervisé.e par : Liam Paull Université de Montréal Enabling on-the-fly machine learning in Duckietown. ### Victor Geadah Supervisé.e par : Guillaume Lajoie Université de Montréal Impact of nonlinear activation functions on learning dynamics of recurrent networks. ### Rose-Marie Gervais Supervisé.e par : Frédéric Gosselin Université de Montréal Générer des métamères visuels utilisant la MEG et l’apprentissage profond pour évaluer l’encodage des propriétés visuelles spécifiques en mémoire à long terme. ### Élodie Labrecque Langlais Supervisé.e par : Jean Provost Polytechnique Montréal Imagerie de pulsatilité de localisation dynamique par ultrasons utilisant l’intelligence artificielle. ### Simon-Olivier Laperrière Supervisé.e par : Pierre L’Écuyer Université de Montréal Outils pour mesurer l’équidistribution de générateurs pseudoalétoires basés sur des récurrences modulo 2. ### Geoffroy Leconte Supervisé.e par : Dominique Orban Polytechnique Montréal Une méthode prédicteur-correcteur multi-précision pour l’optimisation quadratique convexe. ### Florence Ménard Supervisé.e par : Anne Gallagher Université de Montréal Optimization of data analysis strategies for the ELAN Project: a multimodal approach. ### Marco Mendoza Supervisé.e par : Vincent Arel-Bundock Université de Montréal Developing a machine-learning algorithm to predict citizens’ fiscal preferences. ### Neshma Metri Supervisé.e par : Pierre Majorique Léger HEC Montréal Source-level EEG connectivity correlates of immersion during high-fidelity vibrokinetically-enhanced cinema viewing. ### Andrei Mircea Romascanu Supervisé.e par : Jackie Cheung McGill University Reinforcement Learning Rewards for Text Generation. ### Olivier Parent Supervisé.e par : Roberto Araya Université de Montréal Computation model of diendritic nonlinearities in layer 5 pyramidal neurons. ### Maria Sadikov Supervisé.e par : Michel Côté Université de Montréal Utilisation de transfert d’apprentissage pour la caractérisation du graphène. ### Christopher Scarvelis Supervisé.e par : Prakash Panangaden McGill University Convex Relaxations for Neural Network Training. ### Joey St-Arnault Supervisé.e par : Marina Martinez Université de Montréal Analyse multivariée de la cinématique pour créer un lien entre les mouvements comportementaux observées et ceux générés par neurostimulation corticale. ### Danny Tran Supervisé.e par : Brunilde Sansò Polytechnique Montréal Projet 5G et villes intelligentes. ### Anton Volniansky Supervisé.e par : Jean-François Tanguay Université de Montréal Banque de données des issues cliniques à court et long termes des Échafaudages Vasculaires Biorésorbables comparativement aux Stents pharmacoactifs de 2e génération. ### Charles Wilson Supervisé.e par : Paul Charbonneau Université de Montréal Prédiction du cycle solaire par assimilation de données. ### 2nd edition ### Adrien Adam Supervisé.e par : Benjamin De Leener Polytechnique Montréal Développement d’une méthode de super-résolution pour l’imagerie quantitative de susceptibilité néonatale. ### Rodrigo Chavez Zavaleta Supervisé.e par : Sarath Chandar Anbil Parthipan Polytechnique Montréal Understanding the Dynamics of Non-saturating Recurrent Units. ### Cheng Chen Supervisé.e par : Aditya Mahajan McGill University Regret in learning the optimal linear quadratic regulator: empirical comparison of Thompson sampling and adaptive control algorithms. ### Feng Yang Chen Supervisé.e par : David Alexandre Saussié Polytechnique Montréal Learning based visual waypoint detection for agile drone flight. ### Ghassen Cherni Supervisé.e par : Sofiane Achiche Polytechnique Montréal Développement d’un chatbot pour mieux engager les patients dans leur gestion des barrières à une bonne observance aux antirétroviraux. ### Fanny Beltran Supervisé.e par : Benjamin De Leener Polytechnique Montréal Développement d’une méthode basée sur l’apprentissage automatique pour la reconstruction d’images IRM “sparse” du cerveau chez le nouveau-né. ### Valérie Bibeau Supervisé.e par : Bruno Blais Polytechnique Montréal Conception d’un réseau de neurones pour prédire la puissance des agitateurs à partir de données massives issues de simulations. ### Ludovic Bilodeau-Laflamme Supervisé.e par : Brunilde Sansò Polytechnique Montréal Modèles statistiques de la distribution des délais d’accès dans un réseau mobile. ### Geneviève Bock Supervisé.e par : Sofiane Achiche Polytechnique Montréal Conception et réalisation d’un agent conversationnel intelligent (chatbot) pour aider les personnes vivant avec le VIH à mieux gérer leurs traitements anti-VIH. ### Félix-Antoine Constantin Supervisé.e par : Samuel-Jean Bassetto Polytechnique Montréal L’agenda santé satisfaisant. ### Vivienne Crowe Supervisé.e par : Julie Hussin Université de Montréal Investigation of SARS-CoV-2 infection using a quantitative approach. ### Andjela Dimitrijevic Supervisé.e par : Benjamin De Leener Polytechnique Montréal Algorithme validant la segmentation du cerveau chez les enfants (2-8 ans) à partir d’images IRM en utilisant des réseaux adverses génératifs (GAN). ### Parfait Djimefo Supervisé.e par : Samuel Pierre Polytechnique Montréal Modèle de reconnaissance faciale pour les personnes de minorités visibles et les groupes de populations. ### Roxanne Drainville Supervisé.e par : Marina Martinez Université de Montréal Combiner les effets de la stimulation corticale et spinale pour améliorer la marche après une lésion médullaire. ### Marilou Farmer Supervisé.e par : Jalbert Jonathan Polytechnique Montréal Programmation et diffusion des courbes Intensité-Durée-Fréquence des précipitations. ### Sam Finestone Supervisé.e par : Sarath Chandar Anbil Parthipan Polytechnique Montréal Modular Neural Networks for Lifelong Learning. ### Victor Gaudreau-Blouin Supervisé.e par : François Leduc-Primeau Polytechnique Montréal Simulateur de réseaux de neurones profonds implémentés sur matériel non fiable. ### Sarah Hafez Supervisé.e par : Christian Dorion HEC Montréal Deep Learning Methods for Factor Investing. ### Tamara Herrera Fortin Supervisé.e par : Dang Khoa Nguyen Université de Montréal Identifying Patients’ and Caregivers’ Needs and Preferences: the Key to Developing Successful Seizure Detectors. ### Alexander Iannantuono Supervisé.e par : Adam Oberman McGill University Accelerated algorithm for SGD, applied to deep neural networks and reinforcement learning. ### Guillaume Jones Supervisé.e par : Mario Jolicoeur Polytechnique Montréal Génération d’un modèle à l’échelle du génome des cellules cancéreuses ovariennes. ### Hugues Martin Supervisé.e par : Mario Jolicoeur Polytechnique Montréal Caractérisation de la chimiorésistance par apprentissage machine du transcriptome des cellules cancéreuses ovariennes. ### Gabriela Moisescu Supervisé.e par : Doina Precup McGill University Temporal Abstraction in Reinforcement Learning. ### Sacha Morin Supervisé.e par : Guy Wolf Université de Montréal PHATE-NET. ### Stéfan Nguyen Supervisé.e par : Philippe Dixon Université de Montréal Biomechanical analysis of walking in outdoor environements using wearable sensors. ### Mathilde Ricard Supervisé.e par : Sébastien Le Digabel Polytechnique Montréal CHPO: Constrained Hyperparameter Optimization / Optimisation sous contraintes des hyper-paramètres. ### Patrice Rollin Supervisé.e par : Iwan Meier HEC Montréal Conception d’une base de données financières pour l’analyse des fonds communs de placement. ### Myriam Sahraoui Supervisé.e par : Karim Jerbi Université de Montréal Analyse de données cérébrales MEG combinant analyses spectrales et apprentissage machine. ### Monssaf Toukal Supervisé.e par : Dominique Orban Polytechnique Montréal Online Automatic Optimization of Software for Big Data. ### Alina Weinberger Supervisé.e par : Karim Jerbi Université de Montréal Oscillatory brain dynamics under light and deep anesthesia: Predicting states of consciousness using machine learning techniques. ### Paul Xing Supervisé.e par : Jean Provost Polytechnique Montréal Correction d’aberration en microscopie de localisation ultrasonore par apprentissage profond. ### Xin Yuan Zhang Supervisé.e par : Brunilde Sansò Polytechnique Montréal 5G, IoT and Smart cities project. ### Masters excellence scholarships ### Alexandre Adam Supervisé.e par : Laurence Perreault Levasseur Université de Montréal Mesurer l’expansion de l’Univers avec l’apprentissage automatique Le taux d’expansion de l’Univers est une observable importante pour contraindre les modèles cosmologiques qui retracent l’évolution de l’Univers depuis le Big Bang. Récemment (2018), l’équipe du satellite Planck a publié une valeur dérivée des mesures du rayonnement fossile émis lorsque l’Univers n’était âgé que de 300,000 ans. La valeur trouvée contredit les mesures locales du paramètre, faites à partir de la vitesse de fuite des supernovas Ia et des céphéides se trouvant près de la Voie lactée. Nous proposons d’investiguer ce problème via une troisième méthode de mesure qui, jusqu’à maintenant, possédait une précision limitée par la faible quantité connue de quasar situé derrière une galaxie selon notre ligne de vue, telle que l’image du quasar est multipliée par l’effet de lentille gravitationnelle. La précision de cette méthode est limitée en grande partie par la reconstruction de la distribution de masse de la galaxie-lentille. Les avancées récentes des algorithmes d’apprentissage automatiques ont permis de démontrer qu’un réseau neuronal convolutionnel (CNN) pouvait accomplir la reconstruction de la lentille 10 millions de fois plus rapidement que les algorithmes conventionnels. Cette preuve de concept arrive juste à temps pour permettre l’analyse de la quantité phénoménale de données qui sera produite par les télescopes à champs larges dans la prochaine décennie. Nous devrons aussi adapter des architectures comme les machines à inférences récurrentes (RIM) pour automatiser le processus de reconstruction. Les besoins scientifiques de notre mission nécessitera d’adapter l’architecture de nos modèles pour l’estimation des incertitudes. ### Hatim Belgharbi Supervisé.e par : Jean Provost Polytechnique Montréal Microscopie de localisation par ultrasons fonctionnelle 3D (fULM) L’imagerie fonctionnelle cérébrale permet de mieux comprendre quelles régions du cerveau sont impliquées dans différents types de tâches. Il est possible de réaliser ce type d’analyse à l’aide, par exemple, de l’imagerie par résonance magnétique, mais à une résolution spatiotemporelle limitée (de l’ordre du millimètre et de la seconde). Plus récemment, une autre technique, la microscopie de localisation 2D a permis de drastiquement augmenter la résolution spatiale des ultrasons (5 millièmes de millimètre), mais puisqu’elle requiert la détection de microbulles injectées individuelles (approuvées en clinique), sa résolution temporelle était insuffisante pour détecter l’activation du cerveau (dans l’ordre des minutes). Le laboratoire de Jean Provost a récemment développé une nouvelle technique d’imagerie appelée Microscopie de Localisation Ultrasonore Dynamique 3D (dMLU-3D), qui permet d’atteindre la même résolution spatiale en trois dimensions plutôt que deux et aussi une résolution élevée pour les phénomènes périodiques (de l’ordre de la milliseconde). La technique permet la visualisation de la microvasculature cérébrale (morphologie), mais la visualisation de l’activité cérébrale n’a pas encore été développée (fonction). La modélisation de ce qui caractérise une activation cérébrale dépend de plusieurs paramètres non linéaires dont il n’existe pas de vérité terrain à l’échelle de la microvasculature in-vivo, alors l’utilisation d’un réseau de neurones convolutionnel (CNN) s’avère pertinente à cette application. Ce projet vise à montrer qu’il est possible de faire de l’imagerie fonctionnelle (détecter l’activité ou le manque d’activité cérébrale) dans tout le cerveau de rongeur à l’aide de l’approche dMLU-3D avec une résolution spatiotemporelle encore jamais atteinte avec d’autres méthodes comparables. Des expériences seront réalisées afin de révéler et de corréler l’activité des régions visuelles thalamiques et corticales du cerveau du modèle murin suivant la présentation de stimuli visuels. Par la suite, ces résultats seront comparés avec ceux obtenus chez des modèles animaux de la schizophrénie (développemental, pharmacologique, lésionnel ou génétique) afin de vérifier l’hypothèse que ce désordre est caractérisé par une altération des connexions entre le cortex visuel et le thalamus. Ce projet serait la toute première démonstration de la faisabilité de l’imagerie fonctionnelle cérébrale par ultrasons superrésolus en 2D et en 3D, permettant la cartographie de l’activation cérébrale de la totalité du cerveau de rongeur ou d’autres petits animaux, tel le chat, pour des études pré-cliniques permettant à terme de mieux comprendre certaines pathologies et menant potentiellement à un meilleur diagnostic ou même traitement. C’est d’autant plus prometteur étant donné qu’aucune autre modalité d’imagerie peut atteindre une résolution aussi fine, avec une profondeur d’imagerie suffisante et ce, de manière non invasive. ### Marie-Hélène Bourget Supervisé.e par : Julien Cohen-Adad Polytechnique Montréal Segmentation automatique d’images histologiques par apprentissage profond Les axones de la matière blanche sont le prolongement des neurones, et constituent les autoroutes du système nerveux central. Une gaine lipidique, la myéline, entoure ces axones permettant la conduction plus rapide de l’influx nerveux. Des maladies neurodégénératives comme la sclérose en plaques ou encore des traumatismes menacent l’intégrité des axones myélinisés, ce qui peut mener à des déficits sensoriels ou moteurs tels que la douleur ou la paraplégie. Afin de développer de nouveaux traitements, les chercheurs en neurosciences ont besoin de quantifier avec précision la morphométrie de ces axones (taille, épaisseur de myéline, etc.). Mon laboratoire d’accueil NeuroPoly a développé le logiciel AxonDeepSeg permettant de faire la segmentation automatique de neurones sur des images histologiques par des algorithmes d’apprentissage profond. Cependant, AxonDeepSeg manque de robustesse vis-à-vis de la variabilité qui peut exister selon les paramètres d’acquisition et la qualité des images ainsi que selon les espèces. Ce projet vise donc à développer des modèles robustes de segmentation de neurones par l’adaptation et l’implémentation de méthodes innovantes de segmentation par apprentissage profond (Adaptation de domaine, MixUp, FiLM). Le potentiel de généralisation des algorithmes développés sera validé à l’aide de bases de données de microscopie incluant diverses modalités d’imagerie (optique, électronique à balayage, électronique en transmission), espèces, organes et pathologies. De plus, les modèles développés et les données générées seront rendus publics en accès libre et documentés afin de permettre à de nombreux chercheurs et cliniciens en neurosciences de les utiliser. Cet outil permettra également de faire la validation d’autres modalités d’imagerie essentielles dans la recherche sur les maladies neurodégénératives comme l’imagerie par résonance magnétique quantitative non-invasive, et ainsi augmenter la quantité de données utilisables par les chercheurs. ### Joëlle Cormier Supervisé.e par : Valérie Bélanger HEC Montréal Analyse du transport d’urgence aérien dans les régions éloignées du Québec Dans un objectif d’offrir des soins spécialisés à l’ensemble de sa population, le Québec peut compter sur le programme d’Évacuation aeromédicales du Québec (EVAQ) mis en place par le gouvernement. L’offre de service permet de transférer des patients depuis les différentes régions du Québec vers des centres spécialisés de Québec et Montréal afin de leur offrir les soins nécessaires, le tout entouré d’une équipe médicale adaptée à leur condition et leur niveau d’urgence. Plusieurs des services offerts par l’EVAQ ont connu une augmentation de la demande durant la dernière décennie. La présente recherche vise à bâtir un outil de simulation qui permettra de simuler différentes utilisations des ressources. L’analyse des différents scénarios permettra de faire des recommandations à l’ÉVAQ sur les actions à prendre afin d’offrir le meilleur niveau de service possible aux populations des régions. Il y a beaucoup à apprendre sur le modèle instauré au Québec, tant au niveau de la planification stratégique des appareils et des trajets, qu’au niveau de la coordination et des opérations au quotidien. La densité de population, les distances à franchir et les conditions météorologiques difficiles sont des facteurs déterminants à considérer dans leur unicité. ### Edward Hallé-Hannan Supervisé.e par : Sébastien Le Digabel Polytechnique Montréal Optimisation de l’entraînement des réseaux de neurones profonds à partir d’extensions de l’algorithme MADS sur les hyperparamètres de type variable de catégorie Ce projet de maîtrise vise à optimiser l’entraînement des réseaux de neurones profonds à partir d’extensions de l’algorithme MADS sur les hyperparamètres de type variable de catégorie. Ces hyperparamètres sont généralement choisis de manière arbitraire ou heuristique. Or, la plupart des algorithmes d’optimisation développés solutionnent des problèmes où les variables sont de type continu ou entier. En d’autres mots, il existe peu de méthodes d’optimisation pouvant traiter efficacement les variables de catégorie. Cependant, puisque ces variables sont discrètes, il est possible de construire et d’explorer un espace de variables discrétisées avec les méthodes d’optimisation dites recherche directe. Le projet de recherche a pour objectif d’adapter les récents développements de l’algorithme MADS (« Mesh Adaptive Direct Search ») aux variables de catégorie, notamment pour le traitement des contraintes et l’intégration d’un treillis anisotrope dynamique. Plus précisément, nous nous intéressons à optimiser plus rigoureusement les hyperparamètres des réseaux de neurones profonds, afin d’entraîner plus intelligemment les modèles d’intelligence artificielle. Plus particulièrement, les hyperparamètres étudiés seront : la fonction de perte ; les extensions et les modifications à l’algorithme de rétropropagation (ADAM, RMSProp, etc.) ainsi que les régulateurs (LASSO, « Ridge regression », etc.). Les mécanismes développés pourront également servir à modéliser la topologie des réseaux (nombres de couches, nombres de neurones, etc.) En effet, dans le cadre de l’algorithme MADS, le traitement des variables de catégorie pourraient s’étendre à des variables discrètes, dont la valeur modifie la dimension du problème. En pratique, le système résultant permettra donc, pour la première fois, d’optimiser simultanément les hyperparamètres reliés à l’entraînement et ceux reliés à la topologie. ### Dongyan Lin Supervisé.e par : Blake Richards McGill University Analyzing mouse hippocampal « time cell » activities during memory task with machine learning approaches Previous studies have identified hippocampal “time cells” in CA1 that bridge the temporal gap between discontiguous events by firing in tiling patterns during the delay period of memory tasks, such as alternative maze (Pastalkova et al., 2008) and object-odor pairing tasks (MacDonald et al., 2011). However, recent findings have argued that this tiling might be an analysis artifact due to cell-sorting because it also appears in tasks with no memory load (Salz et al., 2016). To address this discrepancy, our collaborators have collected calcium recordings in mouse hippocampal CA1 region during trial unique, nonmatch-to-location (TUNL) task (Talpos et al., 2010) and showed tiling patterns. Our objective is to use computational methods to determine if these patterns are meaningful. To do this, we will first train decoders on the calcium recordings to decode sample for each trial, with temporal sequences preserved (i.e. sorted tiling columns) or shuffled (i.e. randomized columns). If the tiling patterns are indeed meaningful, we would expect to see higher accuracy of the decoder in the preserved sequences. Our next step is to construct a simulated reinforcement learning agent on simulated TUNL task to see whether there exists a consistent tiling pattern in the activities of the neural networks of the reinforcement learning agent. If so, it would suggest that these patterns play a role in preserving information about the sample location during the delay period as a solution to the task. If not, it would suggest that the tiling patterns previously observed in memory tasks could merely be a ubiquitous artifact. Our findings would have a significant impact on the current view of hippocampal “time cells” as well as the functional segregation of the brain. ### Yiqun (Arlene) Lu Supervisé.e par : Guillaume-Alexandre Bilodeau Polytechnique Montréal Jumpy, Hierarchical and Adversarial Variational Video Prediction This project is in the context of intelligent transportation systems. To improve road user detection and tracking, we want to predict their position in future frames using video prediction. However, predicting high fidelity videos over long time scale is notoriously difficult. Current video prediction models either diverges from real samples after a few frames or fail to capture stochasticity in the videos, resulting in bad prediction performance for long videos. In order to overcome this difficulty, new models with ability to do jumpy or hierarchical video prediction are proposed by the AI community. In this proposal, we propose to further develop these ideas and explore new models for stochastic video prediction that is able to do jumpy predictions in a hierarchical manner. We mainly want to explore two research problems: (1) How to do stochastic jumpy video predictions. (2) How to combine jumpy prediction with temporal abstraction. ### Andrei Lupu Supervisé.e par : Doina Precup McGill University Emergent Behaviour in Multi-Agent Reinforcement Learning This project aims for the investigation of intricate emergent behaviours in large scale multi-agent reinforcement learning (MARL). Of particular concern are the behaviours of agents in settings where they are tightly interdependent to the point of nearly composing a single entity. Such settings will draw strong inspiration from biological systems, and be achieved either through a shared common reward or through complex and necessary interactions. Because large interconnected populations of agents present a novel collection of settings complete with new challenges, this project will force a rethinking of well-established reinforcement learning practices, all while probing the limits of their scalability. Furthermore, enabling MARL systems that simultaneously achieve large population scales and appropriate complexity will allow for better modelling of intricate phenomena that have been out of reach of previous artificial intelligence methods. This would potentially result in far-reaching benefits in other scientific disciplines, thus broadening the range of applications of reinforcement learning and simultaneously opening it to easier idea cross-pollination from other fields. These settings will be studied empirically by analyzing the behaviour of existing MARL algorithms, and by comparing and contrasting them to new approaches that allow for more complex interactions between agents. The analysis of the results will be performed quantitatively on the basis of standard reinforcement learning and game theoretic methodology, and qualitatively in light of the principles of behavioural biology. The implementation of the environments and the MARL models will be done with modularity and concurrency in mind and the code-base will then be openly released. ### Nicholas Meade Supervisé.e par : Siva Reddy McGill University Stylistic Controls for Neural Text Generation Deep learning-based approaches to text generation have proven effective in recent years, with many models able to generate realistic text, often exhibiting higher-order structure. While these models produce high-quality samples, there is usually little control provided over what is specifically generated. Recently, work has begun in this area, but much remains to be explored. This application proposes research towards controllable text generation by implementing a variety of stylistic controls that can be used to influence what is sampled from a neural language model. In my previous work, we developed a conditional generative model for music. We demonstrated that we could control for a variety of characteristics during generation by providing the model with an additional externally-specified input called the control signal. For instance, in this work, we trained a model using a composer-based control signal. This signal identified the composer of each piece on which the model was trained. After training, we used the control signal to produce samples of music in the style of specific composers, for instance, Bach and Beethoven. Based on my previous work with music, we are now interested in implementing a similar set of controls for generating text. Such a set of stylistic controls would extend the practical utility of text generated from neural language models. We plan to explore generation methods involving supervised controls and latent (disentangled) controls. ### Marie-Eve Picard Supervisé.e par : Pierre Rainville Université de Montréal Utilisation d’approches d’apprentissage automatique pour l’identification d’une signature cérébrale de l’expression faciale de la douleur L’expression faciale est un outil important pour communiquer diverses informations, notamment la manifestation d’un état de douleur, la présence d’une menace immédiate dans l’environnement et un éventuel besoin d’aide. Les dimensions sensorielle (intensité) et affective (caractère déplaisant) de la douleur peuvent être encodées dans les mouvements faciaux. Les techniques d’analyse jusqu’à présent utilisées pour examiner la relation entre l’expression faciale et l’activité cérébrale lors de l’expérience de la douleur possèdent plusieurs limitations statistiques par rapport à l’évaluation de l’activité cérébrale spatialement distribuée. L’objectif principal du projet proposé est de mieux comprendre les mécanismes neuronaux qui sous-tendent l’expression faciale de la douleur. Des données d’imagerie par résonance magnétique fonctionnelle (IRMf) seront utilisées pour analyser les changements dans l’activité cérébrale en réponse à des stimuli douloureux (mais non dommageables). Plus spécifiquement, ce projet vise à utiliser des approches d’apprentissage automatique (c’est-à-dire l’analyse de modèles multivariés) pour développer une signature cérébrale de l’expression faciale de la douleur afin de prédire les changements faciaux en réponse à des stimuli douloureux dans différents contextes : douleur phasique (stimulation courte), douleur tonique (stimulation longue), et modulation des dimensions sensorielle et affective de la douleur. En bref, ce projet permettra de résoudre certaines lacunes des analyses univariés précédemment utilisées afin de déterminer avec une meilleure précision les bases neurales de l’expression faciale de la douleur et de faire progresser de manière significative notre compréhension des mécanismes cérébraux qui sous-tendent la communication non verbale. ### Myriam Prasow-Émond Supervisé.e par : Julie Hlavacek-Larrondo Université de Montréal Les premières images d’exoplanètes orbitant autour de naines blanches, d’étoiles à neutrons et de trous noirs Les binaires X, formés d’une étoile orbitant autour d’un objet compact stellaire compact (naine blanche, étoile à neutrons ou trou noir), sont des laboratoires fantastiques pour comprendre la physique dans des conditions extrêmes. Au cours des dernières décennies, les binaires X ont fait l’objet d’une multitude d’études dans diverses longueurs d’onde, conduisant à des avancées remarquables dans le domaine de la physique de l’accrétion, ainsi que dans la compréhension de la formation de jets de particules relativistes dans de puissants champs magnétiques. Les binaires X sont aussi d’excellents laboratoires pour comprendre les explosions de type supernova ainsi que l’effet de ces explosions sur le système et son environnement. En effet, la présence d’une étoile à neutrons ou d’un trou noir dans ces systèmes implique directement que l’étoile (et ses potentielles planètes) survivent à ces explosions. Plusieurs études montrent que les planètes et les naines brunes peuvent exister dans une multitude d’environnements, tels que celles qui orbitent très proche de leur étoile hôte (Jupiters chaudes) ou celles qui orbitent à des distances de centaines d’unités astronomiques de l’étoile. Ces découvertes montrent que la formation et la survie des planètes sont mal comprises. Par conséquent, ce projet amène un nouveau point de vue, soit celui des conditions extrêmes. Bref, on pourra étudier plusieurs binaires X et des données des télescopes NIRC2/KECK (visible) et NOEMA (millimétrique) ont déjà été acquises en 2018, et d’autres demandes de temps sont en cours. Selon une analyse préliminaire, la présence d’objets astrophysiques est confirmée, et donc ce projet garantit des résultats surprenants pour la communauté de l’astrophysique. ### Chence Shi Supervisé.e par : Jian Tang HEC Montréal Addressing the retrosynthesis problem using a graph-to-graph translation network Retrosynthesis analysis, which aims to identify a set of reactant graphs to synthesize a target molecule, is a fundamental problem in computational chemistry and is of central importance to the organic synthesis planing as well as drug discovery. The problem is challenging as the search space of all possible transformations is very huge. For decades, people have been seeking to assist chemists in retrosynthesis analysis with modern computing algorithms. Most existing machine learning works on this task rely on reaction templates that define the subgraph patterns of a set of chemical reactions, which require expensive graph isomorphism and suffer from poor generalization on unseen molecule structures. To address the above limitations, in this project, we formulate the retrosynthesis prediction as a graph-to-graph translation task, i.e., translating a product graph to a set of reactant graphs, and propose a novel template-free approach to tackle the problem. We will show that our method excludes the need of domain knowledge, and scales well to large datasets. We will also empirically verify the superiority of our method on the benchmark data set. ### Shi Tianyu Supervisé.e par : Luis Miranda-Moreno McGill University A Multi-agent Decision and Control Framework for Mixed-autonomy Transportation System As the autonomous vehicle becomes more and more popular. Recently, there has been a new emphasis on traffic control in the context of mixed-autonomy, where only a fraction of vehicles are connected autonomous vehicles and interacting with human-driven vehicles. As in a mixed autonomy system, there are several challenges. The first challenge is how to encourage different agents’ cooperation so as to maximize the total returns of the whole system. For example, when there is a gap in front of the adjacent line of the autonomous vehicle, if the autonomous vehicle cuts in immediately, the surrounding vehicle in the adjacent line will also decrease its speed sharply, which will end up a shock wave in traffic flow. Instead, if the autonomous vehicle learns to cooperate with other agents, it will adjust its speed steadily and try to mitigate the negative impact on the whole system. The second challenge is how to improve the communication efficiency in multi-agent system. As autonomous vehicle has different characteristics with human-driven agent, for example, their reacting time and action may be different. Therefore, how to formalize personalized policy for each agent is also worth to explore. The third challenge is how to explore expert knowledge (e.g. green wave, max pressure, actuated control) in transportation domain to improve the training efficiency and performance. Our overall goal of this project is to design effective decision and control framework for an efficient and safe mixed autonomy system by mitigating the shockwave and improving the transportation efficiency. To address the aforementioned problems, we will develop a novel multi agent decision framework based on deep reinforcement learning to improve the decision making and control performance of the agents in mixed autonomy system. ### Rey Wiyatno Supervisé.e par : Liam Paull Université de Montréal Exploiting Experiences and Priors in Semantic Visual Navigation Robotics has always been anticipated to revolutionize the world. However, despite the significant progress over the past few decades, robots have yet to be able to reliably navigate within an unstructured indoor environment. Semantic visual navigation is the task of navigating within a possibly unknown environment using only visual sensors, such as asking a household robot agent to “go to the kitchen”. Traditional “modular” methods combine a Simultaneous Localization and Mapping (SLAM) component with separate search, planning, and control modules. However, these methods do not scale well to large environments, and require significant engineering efforts. Alternatively, end-to-end “learning” solutions produce agent policies that directly infer actions from camera frames, by applying Deep Reinforcement Learning (DRL) techniques on large-scale datasets. Nevertheless, these policies tend to be reactive, do not explicitly exploit scene geometry, and are not data efficient. Furthermore, both modular and learning-based approaches do not sufficiently exploit knowledge from past task instances to improve subsequent search performance in both repeated environments as well as unseen yet similar environments. Our project explores the learning and use of spatial-semantic priors for more efficient semantic visual navigation. We aim to devise a framework that learns, updates, and exploits a topological-semantic map between discovered locations and objects within. We hypothesize that these advances will result in agents that generalize better to unseen similar environments, as well as becoming increasingly more efficient during repeated search queries within the same environment. ### Chengyuan Zhang Supervisé.e par : Lijun Sun McGill University Statistical Modeling Framework to Understand Dynamic Traffic Patterns from Video Data Video-based traffic monitoring systems, as the backbone of modern Intelligent Transportation Systems (ITS), is playing an essential role in sensing traffic conditions and detecting abnormal events/incidents. Semantically understanding traffic scenes and automatically mining the traffic patterns from video data of a static camera can help with traffic situation analysis and anomaly events warning. Given a video of a dynamic traffic scene with several different behaviors happening simultaneously, we want the ITS to learn and understand: “How many typical traffic patterns are in the video? How to semantically interpret these patterns? What are the rules governing the transitions between these patterns?”In this project, we will mainly focus on traffic patterns recognition and anomaly detection from video data, we will: (i) construct representation learning model to extract efficient features; and (ii) develop an unsupervised learning framework based on Bayesian nonparametrics to automatically learn the traffic patterns. ### Doctoral excellence scholarships ### Md Rifat Arefin Supervisé.e par : Irina Rish Université de Montréal Developing Biologically inspired Deep Neural Network for Continual Lifelong Learning We humans are able to continually learn throughout our lifetime which is called lifelong learning. This capability is also crucial for computational systems interacting in the real world and processing continuous streams of data. However, the current deep learning systems struggle to continually acquire the incremental information available over time from non-stationary data distributions. They tend to forget the knowledge which is acquired earlier upon learning the new one – such a problem is called catastrophic forgetting. In this project, we will study biological factors of lifelong learning and their implications for the modelling of biologically motivated neural network architectures that can improve life-long learning capability of computational systems by reducing catastrophic forgetting problem. ### Sumana Basu Supervisé.e par : Doina Precup McGill University Off Policy Batch Reinforcement Learning for Healthcare Artificial Intelligence (AI) has an increasing impact on our everyday life, one being in health care. Today most of the successful applications of AI in healthcare are for diagnosis or prediction, but not for the treatment. But AI agents also have the potential for sequential decision making such as assisting doctors in reassessing treatment options, as well as in surgery. The branch of AI that is a natural fit for handling such sequential decision-making problems is known as Reinforcement Learning (RL).So far most of the successful applications of RL have been in the video game environments. But there are relatively fewer applications of RL in healthcare. One of the reasons is that unlike games, in healthcare the RL agents cannot interact with the environment to explore new possibilities to learn the optimal treatment policy. Trying new treatment options on patients without knowing their consequences is not only unethical but also can be fatal. So, the agent has to learn retrospectively from previously collected batches of data. In RL literature, this is called Off-Policy Learning. Challenges in off-policy evaluation, sparse reward, non-stationary data, and sample inefficiency are some of the roadblocks for using RL safely and successfully in healthcare. During my Ph.D. I aim to tackle some of these challenges in the context of healthcare. ### Christopher Beckham Supervisé.e par : Christopher Pal Polytechnique Montréal Unsupervised representation learning Unsupervised representation learning is concerned with using deep learning algorithms to extract ‘useful’ features (latent variables) from data without any external labels or supervision. This addresses one of the issues with supervised learning, which is the cost and lack of scalability in obtaining labeled data. The techniques developed in this field have broad applicability, especially with regard to training smart ‘AI agents’ and domains where obtaining labeled data is difficult.’Mixup’ (Zhang et al) is a recently-proposed class of data augmentation techniques which involve augmenting a training set with extra ‘virtual’ examples by constructing ‘mixes’ between random pairs of examples in the training set and optimizing some objective on those mixed examples. While the original mixup algorithm simply performed these mixes in input space (which comes with a few limitations) for supervised classification, recent work (Verma et al, Yaguchi et al) proposed performing these mixes in the latent space of the classifier instead, achieving superior results to the original work.One intuitive way to think about ‘latent space mixing’ is that we can imagine that the original data is generated by *many* latent variables, the possible configurations of which increase exponentially as the number of latent variables increases. Because of this we only see a *very small* subset of those configurations in our training set. Therefore, mixup can be seen as allowing the network to explore *novel* combinations of the latent variables it has inferred (which may not already be present in the training set), therefore making the network more robust to novel configurations of latent states (i.e. novel examples) at test time. Empirical results from the works cited corroborate this hypothesis. The first stage of my PhD was exploring mixup in the context of unsupervised representation learning (building on the work of Verma et al, which I also co-authored), in which the goal is to learn useful latent variables from unlabeled data. This was done by leveraging ideas from adversarial learning and devising an algorithm which is able able to mix between encoded states of real inputs and decoding them into realistic-looking inputs indistinguishable from the real data. We showed promising results both qualitatively and quantitatively, and recently published our findings at the NeurIPS 2019 conference. Some preliminary experiments suggest that one of our proposed variants of ‘unsupervised mixup’ has a connection to ‘disentangled learning’, which explores the inference of latent variables which are conceptually ‘atomic’ but can be arbitrarily composed together to produce more abstract concepts (which is similar to how we as humans structure information in the brain). This lays the groundwork for some more exciting research to pursue during my PhD. ### Xinyu Chen Supervisé.e par : Nicolas Saunier Polytechnique Montréal City-Scale Traffic Data Imputation and Forecasting with Tensor Learning With recent advances in sensing technologies, large-scale and multidimensional urban traffic data are collected on a continuous basis from both traditional fixed traffic sensing systems (e.g., loop detectors and video cameras) and emerging crowdsourcing/floating sensing systems (e.g., GPS trajectory from taxis/buses and Google Waze). These data sets have provided us with unprecedented opportunities for sensing and understanding urban traffic dynamics and developing efficient and reliable smart transportation solutions. For example, forecasting the demand and states (e.g., speed, volume) of urban traffic is essential to a wide range of intelligent transportation system (ITS) applications such as trip planning, travel time estimation, route planning, traffic signal control, to name just a few. However, there are two critical issues that undermine the use of these data sets in real-world applications: (1) the missing data and noisy nature make it difficult to get the true signal, and (2) it is computationally expensive to process large-scale data sets for online applications (e.g., traffic prediction). The goal of this project is to develop new framework to better model local consistencies in spatiotemporal traffic data, such as the {sensor dependencies} and {temporal dependencies} resulting from traffic flow dynamics. The scientific objectives are to: (1) Develop nonconvex low-rank matrix/tensor completion models considering spatiotemporal dependencies/correlations (e.g., graph Laplacian [spatial] and time series [temporal]) and traffic domain knowledge (e.g., fundamental diagram, traffic equilibrium, and network flow conservation). (2) Incorporate Gaussian process kernels and neural network structure ### Abhilash Chenreddy Supervisé.e par : Delage Erick HEC Montréal Inverse Reinforcement Learning with Robust Risk Preference RL/IRL methods provide powerful tools for solving a wide class of sequential decision-making problems under uncertainty. However, the practical use of these techniques as a computational tool has been limited historically owing to multiple factors like the presence of high-dimensional continuous state and action spaces in many real-world decision problems, the stochastic and noisy nature of the real world systems compared to the simulated environments, and the indifference of traditional reward and utility functions to the risk preference of the agent. I am excited about the possibility of directing my future research towards building risk-aware MDP models as they would provide stronger reliability guarantees than their risk-neutral counterparts. one typical modeling premise in RL/IRL is to optimize the expected utility (i.e., an assumption that humans are risk-neutral), which deviates from actual human behaviors under ambiguity. Recent work suggests such an effort can provide stable solutions for high-dimensional state space problems, thus making them more applicable for practical use cases.As an effort in this direction, under the guidance of Prof. Erick Delage, I am working towards developing risk-aware IRL/RL algorithms for portfolio selection problems. Applications that I am interested in include, but are not limited to, i) learning the agent’s risk profile using inverse learning methods and ii) Risk sensitive exploration in RL setting. Our work tries to formulate the inverse learning model from a distributionally robust optimization (DRO) point of view where the agent performs at least as well as the expert in terms of the risk-sensitive objective. We plan to achieve this by building an ambiguity set for the expert’s risk preference and train the agent to learn by taking a worst-case approach, thus shielding the agent from the ambiguity in the underlying risk distribution. ### Chloé Bourquin Supervisé.e par : Jean Provost Polytechnique Montréal Mesure de la pulsatilité cérébrale et son impact sur la cognition chez la souris vasculairement compromise par imagerie ultrasonore Les maladies cardiovasculaires peuvent être à l’origine d’un vieillissement cérébral accéléré. Les artères, telles l’aorte ou les carotides, sont riches en fibres élastiques, permettant d’adoucir les fluctuations de la pression sanguine (ou pulsatilité) lors du cycle cardiaque dans les vaisseaux cérébraux en aval. Avec l’âge et la maladie, les artères deviennent plus rigides, entraînant une augmentation de la pulsatilité en aval et menant à des altérations microvasculaires. Cartographier la pulsatilité dans l’ensemble du réseau vasculaire cérébral pourrait donc devenir un biomarqueur permettant de diagnostiquer les maladies neurodégénératives. Jusqu’à récemment, suivre l’évolution du pulse dans le réseau vasculaire cérébral n’était pas possible : la microscopie optique ne permet que la mesure des micro-vaisseaux à la surface du cerveau, tandis que l’IRM haut champ permet d’imager un cerveau entier mais n’a pas une résolution spatiotemporelle et sensibilité suffisantes pour mesurer de petits vaisseaux. Une nouvelle technique ultrasonore pourrait relever ce défi : la Microscopie par Localisation Ultrasonore (MLU). Basée sur la localisation et le suivi de microbulles injectées comme agents de contraste, elle permet de cartographier les vaisseaux avec une résolution de l’ordre de 5 µm dans l’ensemble du cerveau. Cependant, cette méthode nécessite de suivre les microbulles durant 10 minutes pour finalement n’obtenir qu’une unique image de la vascularisation cérébrale. Notre objectif est de parvenir à rendre cette méthode dynamique, en la synchronisant avec l’ECG et la respiration afin d’obtenir non pas une image unique mais un film d’au moins une pulsation cardiaque, afin d’observer les variations de vitesse du flux sanguin au cours du cycle, et d’en déduire la pulsatilité. Cette nouvelle méthode permettra de démontrer pour la première fois la variation de la pulsatilité dans le cerveau entier, d’établir un lien de corrélation entre l’augmentation de la pulsatilité et les pertes de cognition ainsi que les dommages cérébraux et d’établir la mesure de la pulsatilité comme biomarqueur pour suivre l’évolution de maladies cardiovasculaires et/ou neurodégénératives. ### Theophile Demazure Supervisé.e par : Pierre-Majorique Léger HEC Montréal Apprentissage profond et classification des états cognitifs pour la modulation en temps réel des interactions humain-machine en milieu automatisé. L’univers du travail est en train d’être profondément modifié. Des technologies comme la robotique et des applications de l’intelligence artificielle s’intègrent de plus en plus dans les tâches du travail. L’objectif de cette recherche est de prendre en compte l’humain dans un environnement composé de machines. Ces dernières ne sont pas capables de percevoir que l’employé, avec lequel elles collaborent, est fatigué, absent mentalement, ou tout simplement distrait. Un collègue, dans ce cas-ci, s’ajusterait ou le préviendrait afin qu’il reprenne ses esprits. La machine, quant à elle, poursuivrait son activité sans s’ajuster augmentant le risque d’accident ou d’erreur. Pour répondre à ce problème, ce projet porte sur le développement d’un système qui s’adapte à l’état cognitif de son utilisateur comme la fatigue, la charge mentale ou la fatigue.Les interfaces cerveau-machines utilisent des mesures neurophysiologiques de l’être humain pour surveiller, s’adapter, ou se faire contrôler. À l’intérieur, des algorithmes d’apprentissage machine permettent de classifier l’état cognitif à partir des données capturées en temps réel. En utilisant les signaux électriques dégagés par le cerveau et la dilatation de la pupille, il est possible de discriminer entre plusieurs états dans le temps la situation de l’opérateur. Le prototype développé pourra ainsi donner l’ordre à d’autres machines de ralentir la cadence ou de prévenir lorsque l’employé avec qui elles collaborent semble fatigué ou peu vigilant. Ce prototype sera développé et évalué en laboratoire dans un environnement contrôler. Il s’agit d’une preuve de concept à l’entreprise. Les interfaces cerveau-machines sont aujourd’hui principalement utilisées en médecine pour des prothèses, système d’assistance à la parole ou fauteuil roulants. Les retombées sont majoritairement en sécurité au travail (transport, manufacture) et dans l’optimisation de l’interaction humain-machine (collaboration humain-machine). ### Sébastien Henwood Supervisé.e par : François Leduc-Primeau Polytechnique Montréal Coded Neural Network Les réseaux de neurones profonds connaissent un engouement généralisé en ce début de décennie. Or, les progrès dans ce domaine s’accompagnent d’une hausse des besoins en capacité de calcul devançant la loi de Moore. Dans ce contexte, on cherche à proposer un ensemble de méthodes permettant d’optimiser les besoins en énergie de réseaux de neurones profonds en prenant en compte les caractéristiques physiques (mémoire, processeur, etc) du système accueillant le réseau pour son usage final. L’objectif est d’avoir une méthode suffisamment générale pour s’adapter aux tâches et réseaux variés que les concepteurs pourraient vouloir déployer dans leurs applications, et réduisant la charge énergétique selon un compromis capacité du réseau/énergie contrôlable. Ces travaux permettraient d’une part de gagner en énergie sur les systèmes des utilisateurs (par exemple, téléphone intelligent), ce faisant favorisant les usages déconnectés. D’autre part, on cherche à toucher les utilisations en data-centers, si voraces en énergie. Ce projet de recherche permettra à terme de tirer parti au mieux des ressources allouées à l’apprentissage automatique dans sa phase d’exploitation, pour s’assurer de son acceptabilité sociale d’une part et de sa viabilité technique et économique d’autre part. ### Jad Kabbara Supervisé.e par : Jackie Cheung McGill University Computational Investigations of Pragmatic Effects in Language This thesis focuses on natural language processing (NLP), specifically computational pragmatics, using deep learning methods. While most NLP research today focuses on semantics (literal meaning of words and sentences), my research takes a different approach: I focus on pragmatics which deals with intended meaning of sentences, one that is context-dependent. Correctly performing pragmatic reasoning is at the core of many NLP tasks including information extraction, summarization, machine translation, sentiment/stance analysis. My goal is to develop computational models where pragmatics is a first-class citizen both in terms of natural language understanding and generation. I have already made strong progress toward this goal: I developed a neural model for definiteness prediction [COLING 2016] — the task of determining whether a noun phrase should be definite or indefinite — in contrast to prior work relying on heavily-engineered linguistic features. This has applications in summarization, machine translation and grammatical error correction. I also introduced the new task of presupposition triggering detection [ACL 2018 — best paper award] which focuses on detecting contexts where adverbs (e.g. “again”) trigger presuppositions (e.g.,“John came again” presupposes “he came before”). This work is important because it is a first step towards language technology systems capable of understanding and using presuppositions and because it constitutes an interesting testbed for pragmatic reasoning. Moving forward, I propose to examine the role of pragmatics, particularly presuppositions, in language understanding and generation. I will develop computational models and corpora that incorporate this understanding to improve: (1) summarization systems e.g. in a text rewriting step to learn how to appropriately allocate adverbs in generated sentences to make them more coherent and (2) reading comprehension systems where pragmatic effects are crucial for the proper understanding of texts and where systems can answer questions of pragmatic nature whose answers are not found explicitly in the text. By the end, the thesis would present the first study on presuppositional effects in language to enable pragmatically-empowered natural language understanding and generation systems ### Caroline Labelle Supervisé.e par : Sébastien Lemieux Université de Montréal Enhancing the Drug Discovery Process: Bayesian Inference to evaluate Efficacy Characteristics of potential Drug Through Uncertainty During the multi-phase drug-discovery process, many compounds are tested in various assays which generates a great deal of data from which Efficacy Metrics (EM) can be estimated. Compounds are selected with the aim of identifying at least one sufficiently potent and efficient to go into preclinical testing. This selection is based on the EM meeting a specific threshold or by comparison to other compounds. Current analysis methods suggest point estimates of EM and hardly consider the inevitable noise present in experimental observations, thus failing to report the uncertainty on the EM and precluding its use during compound selection. We propose to extend our previously introduced statistical methods (EM inference and pairwise comparison) to the ranking of a panel of compounds and to combinatorial analysis (multiple compounds tested simultaneously). Given an EM threshold, we aim at identifying the compounds with the highest probability of meeting that criteria. We use a hierarchical Bayesian model to infer EM from dose-response assays (single- and multi-doses), yielding empirical distributions for EM of interest rather than single point estimates. The assay’s uncertainty can thus be propagated to the EM inference and to compound selection. We are thus able to identify all compounds of an experimental dose-response dataset with at least 1% chance of being amongst the best for various given EM, and to characterize the effects of each compounds of a combinatorial assay. This novel methodology is developed and applied to the identification of novel compounds able to inhibit cellular growth of leukemic cells. ### Sébastien Lachapelle Supervisé.e par : Simon Lacoste-Julien Université de Montréal Uncertainty in Operations Research, Causality and Out-of-Distribution Generalization My research focuses on two main directions: widening the operations research toolbox using recent advances in deep learning and learning causal structures. Both aspects have the potential to be useful in various applications, for example the optimization of railway operations, gene expression studies as well as the understanding of different protein interactions in human cells.Together with Emma Frejinger and its team at the CN chair, we developed a methodology which allows to predict tactical solutions given only partial knowledge of the problem using deep neural networks. We demonstrated the efficiency of the approach on the problem of booking intermodal containers on double-stack trains. Moreover, we are currently working to apply machine learning techniques to standard operations research problems such as the knapsack and the travelling salesman problem in hope of gaining insight about classical algorithms to solve them. More recently, I have been interested in the nature of causal reasoning and how machines could acquire it. Typical machine learning systems are good at finding statistical dependencies in data, but often lack the causal understanding which is necessary to predict the effect of an intervention (e.g. the effect of a drug on the human body). Together with my co-authors, we developed « Gradient-Based Neural DAG Learning », a causal discovery algorithm which aims at going beyond simple statistical dependencies. We showed the algorithm was capable of finding known causal relationships between multiple proteins in human cells. In the future, I will work to make machine learning more adaptive and able to reuse past knowledge in order to learn new patterns faster. This is something humans do all the time, but which is hard for current algorithms. I believe causality is part of the answer, but other frameworks like meta-learning, transfer learning and reinforcement learning are going to be necessary. Apart from bringing us closer to human-level intelligence, making progress in this direction would benefit many applications. For instance, if a machine learning system is used to predict tactical solutions to a railway optimization problem, the distribution of problems it faces might shift due to changes in trade legislation, hence rendering the predicted solutions far from optimal. We should aim to build systems which can adapt to a changing world quickly. ### Antoine Boudreau LeBlanc Supervisé.e par : Bryn Williams-Jones Université de Montréal Bioéthique écosystémique et mégadonnées: santé, agriculture et écologie Les problèmes actuels sont globaux, liant société, économie et environnement à la santé. L’antibiorésistance par exemple provient d’un mésusage d’antibiotiques en santé et en agriculture qui vient réduire l’efficacité de ceux-ci. Pour attaquer ce problème, de larges collaborations entre médecins, agriculteurs et écologistes deviennent nécessaires, mais demeurent limitées par bons nombres de défis techniques (ex. : partage de données) et éthiques (consentement, sécurité) apparaissant dès l’intégration les données et les connaissances pour intervenir de façon concertée. L’objectif de cette thèse est d’étudier ces enjeux affectant la circulation des données entre santé, agriculture et écologie afin de proposer un modèle de gouvernance des données maximisant l’accès et la protection des données pour appuyer la recherche, la surveillance et l’intervention tout en maintenant la confiance des fournisseurs de données. Ce projet fondera son analyse éthique sur une cartographie des relations entre les intervenants clés pouvant supporter un réseau de partage de données entre la santé, l’agriculture et l’écologie. Quatre études de cas sont amorcées et permettent de décrire le processus de constitution de ce réseau aux niveaux interministériel, intersectoriel, interprofessionnel, interpersonnel (certification éthique obtenue). Le devis ethnographique réalisé en étroite collaboration avec ces 4 milieux d’accueil supportera l’écriture d’un cadre de gouvernance par théorisation ancrée. Il sera ensuite comparé aux initiatives internationales (Danemark, Angleterre, États-Unis). Cette thèse permettra d’appuyer la mise en œuvre de réseaux structurants de partage de données intersectorielles au niveau de la médecine vétérinaire au Québec et jettera les bases d’un cadre de gouvernance pour l’interconnexion des bases de données entre organisations et secteurs. ### Maude Lizaire Supervisé.e par : Guillaume Rabusseau Université de Montréal Connexions entre réseaux récurrents, automates pondérés et réseaux de tenseurs pour l’apprentissage avec données séquentielles À plusieurs reprises dans l’histoire, des découvertes ont été faites parallèlement par plusieurs scientifiques. On n’a qu’à penser au calcul infinitésimal développé indépendamment par Newton sous l’influence de ses travaux sur les lois universelles du mouvement et Leibniz inspiré par le principe philosophique de l’infiniment petit. À l’intersection entre plusieurs disciplines, ce type de découvertes n’atteignent leur plein potentiel que grâce à la contribution des différentes expertises. Dans cet ordre d’idées, de nombreuses équivalences peuvent être tracées entre les formalismes développés en physique et en intelligence artificielle. En particulier, une méthode pilier de la formulation moderne employée en physique quantique, les réseaux de tenseurs, peut être reliée aux réseaux récurrents, l’une des principales familles de modèles adaptés aux données structurées en apprentissage profond. Ces derniers sont également connectés aux automates pondérés, qui sont des modèles au cœur des méthode formelles et de vérification en informatique théorique. L’exploration des liens entre ces trois méthodes (réseaux de tenseurs, réseaux récurrents et automates pondérés) permet de tirer profit des garanties théoriques offertes par les méthodes formelles, de l’expressivité et des nombreuses applications des réseaux récurrents, tout en faisant le pont avec les débouchés des réseaux de tenseurs dans les domaines des matériaux et de l’informatique quantiques. Le projet vise ainsi à créer des passerelles entre ces différentes disciplines et exploiter les progrès faits dans l’une au profit des autres. ### Elena Massai Supervisé.e par : Marina Martinez Université de Montréal Neuroprosthesis development to recover the gait after spinal cord injury in rats Spinal Cord Injury (SCI) interrupts the communication between the brain and the spinal locomotor networks, causing leg paralysis. When SCI is incomplete (iSCI), some nerve fibers survive the lesion and patients with iSCI can eventually regain some motor abilities. The goal of this study is to assess in the rat model whether combined brain and spinal stimulation can lead to a superior locomotion recovery after spinal cord injury. Artificial Intelligence (AI) techniques will be employed to track the motor activity, drive the stimulation and optimize the strategy in real time. By refining the spatiotemporal stimulation parameters, the intelligent algorithm will help the rat’s brain to generate leg trajectory that features a better clearance of the ground during swing, stronger leg extension and higher posture during stance. We expect that optimized neuroprosthetic stimulation will result in locomotor patterns that are more similar to intact rats and will facilitate the recovery of voluntary control of locomotion. The results will provide a framework for the future development of efficient neuromodulation interfaces and prosthetic approaches for rehabilitation. ### Antoine Moevus Supervisé.e par : Benjamin De Leener Polytechnique Montréal Quantitative susceptibility mapping framework for assessing cortical development in neonates after severe deoxygenation at birth Hypoxic ischemic encephalopathy (HIE) is a newborn brain pathology that is common but, unfortunately, not well understood. HIE affects 1.5 per 1000 live births in developed countries and is the leading cause of death and devastating sequelae in terms of neonates cognitive, behavioural, and physical disabilities. The most effective clinical treatment, therapeutic hypothermia, improves the survival rate; however, the repercussions of HIE remain unclear for survivors. As of today, the understanding of altered cortical growth mechanisms after HIE is incomplete but promising non-invasive magnetic resonance imaging (MRI) technique, called quantitative susceptibility mapping (QSM), provide new brain biomarkers that can help understand how HIE affects the brain development. Yet, because cortical development of neonates is rapid and sophisticated, standard clinical neurological imaging tools, such as MRI templates, are not suited for neurodevelopmental analysis in neonates. Therefore, we propose to implement new methods for solving the QSM reconstruction problem and improve the common MRI template by developing adaptive age-based longitudinal templates. We will adopt a data-driven strategy with deep learning in order to create a new framework for the pediatric and neurology communities. ### Alexis Montoison Supervisé.e par : Dominique Orban Polytechnique Montréal Méthodes multi-précision pour l’optimisation et l’algèbre linéaire Ce projet de recherche a pour but de développer des méthodes capables de basculer d’une précision machine à l’autre durant la résolution de problèmes d’optimisation de grande taille, et d’effectuer l’essentiel des opérations en basse précision où elles sont peu coûteuses et requièrent peu d’énergie. Nos résultats préliminaires indiquent des économies énergétiques pouvant aller jusqu’à 90% sur certains problèmes. Ces méthodes s’appliquent notamment à la biologie des systèmes, qui requiert des solutions en quadruple précision, et au machine learning, où la demi précision est de plus en plus populaire. Sur les plateformes spécialisées émergentes gérant nativement ces nouvelles précisions, comme les cartes graphiques Turing de Nvidia qui implémentent la demie-précision ou encore le processeur IBM Power9 qui implémente la quadruple précision, ces méthodes seront à même d’exploiter au maximum le bénéfice du travail en multi-précision. À l’ère des données massives et de l’explosion de l’information, des algorithmes permettant des économies d’énergie significatives sur les plateformes adéquates sont un investissement pour l’avenir du Canada, en termes du volume de données exploité et de l’environnement. ### Amine Natik Supervisé.e par : Guillaume Lajoie Université de Montréal Decomposition of information encoded in learned representations of recurrent neural networks The human brain contains billions of neurons that communicate with each other through trillions of synapses, enabling us to learn new skills, solve complex tasks and understand intricate concepts. Everything we do such as walking, eating, communicating, and learning, is a function of these neurons firing in certain patterns, in specific locations. This sophisticated biological neural network is the outcome of millions of years of evolution. Recent advances in deep learning have proposed several artificial neural network architectures for solving complex learning tasks, by taking simplified inspiration from neural circuits in our brains. Examples of these include convolutional neural networks for image and audio processing, recurrent neural networks for sequence learning and autoencoders for dimensionality reduction. Both biological and artificial networks rely on efficient calibration of synapses (or connection weights) to match desired behaviours. This adjustment is how a network « learns », but is a complicated task that is not well understood. An important substrate of networks after learning is the internal low dimensional representation found in the joint activity of neural populations that emerge upon performing a learned task. The present research aims to explore and further investigate these internal representations, and address the question of how do structural properties of network connectivity impact the geometry, dimensionality and learning mechanisms encoded by these internal features. We plan to answer this question by leveraging multidisciplinary data exploration tools from graph signal processing, dimensionality reduction, representation learning and dynamical systems. We expect that this project will allow us to gain better understanding of how natural and artificial neural networks solve complicated tasks, which in turn will help us find methodological ways to improve existing structures, and build new models, but more from a deeper understanding rather than trial and error. ### Cédric Poutré Supervisé.e par : Manuel Morales Université de Montréal Statistical Arbitrage of Internationally Interlisted Stocks In this project, we will investigate a novel form of statistical arbitrage that will combine artificially created financial instruments in a high-frequency world, meaning that we will operate in the millisecond timeframe. These instruments will be constructed in such a way that they will offer very interesting statistical properties that will enable us to exploit violations in the law of one price in the Canadian and American markets. This arbitraging activity is essential, since it is making them more efficient by eliminating mispricing in equities that are quoted on both markets. The novel strategy will be tested on a large basket of equities on three trading venues in North America and given that we are working in high-frequency, this means that millions of market observations are ingested and analyzed daily by our trading algorithms. In order to be proactive in the markets, to make extremely fast and accurate predictions, and because of the complex nature of financial data and its abundance, we will be relying on machine learning algorithms to guide our trading decisions. ### Carter Rhea Supervisé.e par : Julie Hlavacek-larrondo Université de Montréal A Novel Deep Learning Approach to High-Energy Astrophysics Despite machine learnings recent rise to stardom in the applied sciences, the astronomy community has been reluctant to accept it. We propose to gently introduce several forms of machine learning to the community through the study of the hot gas pervasive in galaxy clusters. Currently, emission spectra from galaxy clusters are studied by fitting physical models to them and using those models to extract relavent physical parameters. Unforunately, there are several inherent pitfalls with this method. We plan to train different algorithms — from a random forest classifier to a convolutional neural network — to parse the necessary thermodynamic variables from the emission spectra. The fundamental goal of this project is to create and open-source pipeline and suite of tutorials which integrate machine learning into the study of galaxy clusters. ### Charly Robinson La Rocca Supervisé.e par : Emma Frejinger Université de Montréal Learning solutions to the locomotive scheduling problem Given a set of demands on a railway network, how should one assign locomotives to trains in order to minimize total costs and satisfy operational constraints? This question is critical for Canada’s largest railway company: Canadian National Railways. Given the size of their network, even a small relative gain in efficiency would produce significant savings. The goal of this research is to explore recent advances in machine learning in order to efficiently solve the locomotive assignment problem. The idea is to train a neural network on precomputed solutions of the problem with the aim of learning the correct configuration of locomotives for a given train. By combining both integer programming and deep learning, the computational time can be reduced by at least an order of magnitude compared to integer programming alone. This is a solution that is significantly more efficient and practical for train operators. ### Davood Wadi Supervisé.e par : Sylvain Sénécal HEC Montréal Cognition-Based Auto-Adaptive Website User Interface in Real Time A personal message that is designed specifically for the need and taste of consumers has always been the goal of media outlets, retailers, and social activists. Here at Tech3Lab, we are launching this massive study of personalization in an unprecedented way: by analyzing neurophysiological and psychophysiological signals of the body to determine the best possible look and feel on websites to improve user experience and best convey the intended message. Previously, auto-adaptive website personalization was carried out mostly by guesswork and theory, in which there is no real evidence for the parameters used. Thanks to the equipment in Tech3Lab, such as EEG, fNIRS, physiological measurement instruments, and eye tracking measures, we are able to base our adaptive system on direct signals from the body. This interdisciplinary study of cognitive neuroscience, marketing, and data science has the potential to revolutionize the approach of designers, developers, and editors to website design by studying auto-adaptive websites using direct body measures. ### Zichao Yan Supervisé.e par : William Hamilton McGill University Bridging the gap between structures and functions: learning interpretable graph neural representation of RNA secondary structures for functional characterization Cells are the basic units of life and their activity is regulated by many delicate subcellular processes that are crucial to their survival. Therefore, it is important to gain more insights into the complex control mechanisms at play, both to obtain a better fundamental understanding of biology, and to help understand diseases caused by defects in these mechanisms. We are particularly interested in the regulatory roles played by RNA molecules in the post-transcriptional phase such as subcellular localization and RNA-protein interactions. RNA secondary structures, a representation of how RNA sequences fold onto themselves, can have a significant impact on the molecule’s regulatory functions through its interaction with various mediating agents such as proteins, RNAs and small molecules. Therefore, in order to fully exploit RNA secondary structures to better understanding of their functions, we propose a novel framework of an interpretable graph neural representation of RNAs, which may ultimately lead us to the design of RNA based therapeutics for disease such as neurodegenerative disorders and cancers, the success of which would crucially depend on our capability of understanding the relations between RNA structures and functions. ### Internship grants: Data to tell ### Anaïs Babio Université de Montréal Stage chez Synapse-C, spécialité science des données ### Marc Boulanger Université de Montréal Stage chez Radio-Canada, spécialité communication ### Stephanie Cairns McGill University Stage au CIRANO, spécialité science des données ### Ève Campeau-Poirier Université de Montréal Stage chez Synapse-C, spécialité science des données ### André-Anne Côté HEC Montréal Stage chez Synapse-C, spécialité communication ### Ambre Giovanni Concordia Stage chez Le Devoir, spécialité science des données ### Philippe Robitaille-Grou Université de Montréal Stage chez Le Devoir, spécialité science des données ### Catherine Soum Université de Montréal Stage au CIRANO, spécialité communication ### Jérémie Tousignant Université de Montréal Stage chez Radio-Canada, spécialité science des données ### Postdoc-entrepreneur program ### Marco Bonizzato Supervised by: Marina Martinez Université de Montréal Company: NeuralDrive NeuralDrive is a medical-device start-up with a revolutionary view on AI-based neurostimulation therapy for people with spinal cord injury. The device applies neurostimulation of the brain, nerves and muscles to improve the efficiency of motor training, reverse paralysis and enable walking again. ### Yann-Seing Law-Kam Cio Supervised by : Sofiane Achiche Polytechnique Montréal Company: DesignBot inc. The project goal is to develop a software solution aimed at alleviating the burden on engineers and designers moving from product ideas to functional proof of concept. The software guides engineers via a methodology, tested using research and artificial intelligence (AI) algorithms, that allows them to identify the key functionalities, properties and means needed to ensure their technology product is functional. ### Postdoctoral research funding ### Ammar Alsheghri Supervisé.e par : François Guilbault Polytechnique Montréal Deep Learning Approach to generate patient-specific teeth / Approche d’apprentissage profond pour générer des dents de remplacement spécifiques au patient. Dental offices are faced with hundreds of thousands of dental reconstructions per year. Each dental reconstruction typically requires a dental professional to manually design and input the characteristics of the tooth to be reconstructed. Consequently, this time-consuming process is difficult to reproduce between professionals and hence leads to great variability in quality. This project will use Deep Learning approaches to develop a new methodology that automatically designs patient-specific teeth. Using a dataset of roughly 5,000 digitalized arches as a gold standard, neural networks will be trained to generate and/or deform mesh models to yield a volumetric surface representing the tooth to be reconstructed in its spatial context. The resulting integrated system will be designed to continuously learn. Indeed, teeth generated by the system can be modified by a dental professional making a restoration. The resulting modification will then be used to retrain the network and increase its effectiveness. This project aims to generate dental restorations in a few seconds through artificial intelligence, replacing the current manual process that can take between 30-90 minutes. ### Kartik Ahuja Supervisé.e par : Ioannis Mitliagkas Université de Montréal Theory and Methods for OOD Generalization and Robust Learning. “Existing machine learning models are trained using empirical risk minimization (ERM). These models are known to generalize well when the test and the train distribution are similar. In many real-life applications, we expect models to be robust to scenarios when the train and test distributions are different, i.e., out-of-distribution (OOD) generalization. ERM based models often have poor OOD generalization. In this project, we aim to build theory and methods for OOD generalization. In recent works, it has been shown that incorporating principles of causality into traditional machine learning is the key to addressing OOD. In our recent work, we showed that incorporating causality into ERM formulation transforms a standard optimization problem into solving for the Nash equilibrium of a special game. Building on these recent works, we will explore the following problems in this project. i. Translate the standard PAC learnability notions from standard generalization to OOD generalization. ii. Develop methods using game theory and robust optimization that provably learn the predictors that exhibit OOD. iii. Incorporate causality into other areas, e.g., reinforcement learning, adversarial learning and unsupervised learning.” ### Taoli Cheng Supervisé.e par : Aaron Courville Université de Montréal Physics-inspired Deep Generative Modeling and New Physics Search for High Energy Physics. After finding the Higgs Boson in 2012, the Large Hadron Collider (LHC) at CERN, which explores the frontier of elementary particle physics, has put its focus on searching for new physics signals. However, it’s challenging to find rare signals given the large amount of data produced at the LHC. The modern approach of using machine learning assisted methods comes into aid. This project will focus on physics-inspired generative modeling. Taking advantage of the domain knowledge of underlying physics theories, physics-aware neural nets learn powerful representations by implementing physics laws within the architecture. A physics-aware generative model that is based on a set of fundamental physics laws will simply conserve these laws and bring realistic modeling of physics events. At the same time, the generative models will be able to be used for model-agnostic novelty detection, which assists new physics search at the LHC. ### Nehme El-Hachem Supervisé.e par : Vincent-Philippe Lavallée Université de Montréal Targeting leukemia stem cells using computational systems biology approaches. My research project at CHU Sainte Justine Research Center will primarily exploit single-cell RNA sequencing analytical tools to get first insights into the functional impact of mutations driving stem cell self-renewal, a critical step in the development of acute myeloid leukaemia (AML) a type of leukemia originating in the bone marrow from immature myeloid progenitors and affecting both children and adults. Our novel computational approaches will efficiently integrate sequencing data with large pharmaco-genomic databases to identify drug repurposing opportunities that can target specific molecular hits required for leukemia transformation. ### Jessie Galasso-Carbonnel Supervisé.e par : Houari Sahraoui Université de Montréal L’assistance au développement du logiciel au temps des données massives. L’ère du cloud et du big data a pour conséquence l’augmentation du besoin de logiciels complexes dans de nombreux domaines comme ceux de la santé ou de l’énergie. Dans ce contexte, automatiser autant que possible la création, la correction et plus généralement la manipulation de code est devenu un enjeu important. De nombreux travaux se sont intéressés, ces dernières années, à l’utilisation de techniques d’optimisation comme les algorithmes évolutionnistes ou l’intelligence artificielle pour la correction automatique de bogues ou la génération de code. Cependant, ces travaux se concentrent sur des cas très spécifiques et offrent peu de possibilités de généralisation. L’idée principale de notre projet est de voir les tâches d’écriture, de correction, de sophistication et de génération de code comme un continuum dans lequel il est possible de réduire de manière substantielle la partie des connaissances fournie par le développeur et de compenser cette réduction par l’abstraction de connaissances à partir des dépôts de données massives sur les logiciels. Nous proposons d’explorer plusieurs techniques d’optimisation (programmation génétique multiobjectif) et d’apprentissage automatique (code embedding et réseaux de neurones profonds), ainsi que leur combinaison, pour l’abstraction de ces connaissances et leur restitution en fonction de la tâche courante de développement. ### Emma Glennon Supervisé.e par : Timothée Poisot Université de Montréal Algorithmic outbreak detection for low-resource settings. “Infectious disease outbreaks are most easily controlled when detected quickly, but a lack of access to diagnostic resources can make early detection difficult. This problem is compounded in low-resource settings and those with underfunded public health infrastructure, which face challenges to both detecting and controlling infectious disease. This project aims to develop algorithms and simple tools for outbreak detection and identification. Symptom data is already cheaply and routinely collected in many low-resource settings. We propose to use data science techniques to automatically process this data, as well as to create accessible software tools to help public health agencies access these quantitative insights. Put together, this work will help public health agencies in low-resource settings identify and control emerging outbreaks. “ ### Kevin Kovalchik Supervisé.e par : Etienne Caron Centre hospitalier universitaire Mère-Enfant (CHU Sainte-Justine) Machine learning-assisted identification of SARS-CoV-2 epitopes from mass spectrometry data. The current COVID-19 pandemic highlights the importance of rapid vaccine development platforms. A key step in vaccine development is the identification of viral peptide antigens presented by major histocompatibility complexes (MHC) on cell surfaces, termed MHC-associated peptides (MAPs). Currently, mass spectrometry (MS) is the only platform which allows for the direct, systematic and unbiased identification of MAPs from clinical samples (MS-immunopeptidomics). Using MS data from the SysteMHC Atlas, the largest public repository of MS-immunopeptidomics data, we will build machine learning models for the accurate prediction of MS fragmentation patterns and retention times of MAPs, key features used in peptide identification. After validation, these models will be integrated as core components of a complete peptide identification workflow for MS data. This workflow will be used to identify SARS-Cov-2 epitopes from MS data of infected cells and tissues as part of a collaboration with several academic (IRIC, CHUSJ), private (Nexelis and Trans-Hit Bio) and federal organizations (NRC) to support validation of COVID-19 vaccine efficacy. Beyond the immediate application, the proposed workflow will be applicable to any peptide-centric MS data and will find broad applications in immuno-oncology and systems biology. ### Hiroshi Mamiya Supervisé.e par : Erica Moodie McGill University Building precision retail grocery strategies to promote healthy food purchasing using massive consumer panel data. Diets consisting of high-calorie, nutrition-poor foods are one of the leading causes of morbidity and mortality, increasing risk of obesity, type 2 diabetes, cardiovascular disease, infectious diseases and more. We aim to leverage the idea of precision medicine to develop precision retail strategies so as to develop geographically or even individually targeted strategies to promote healthy food purchasing choices. To do so requires massive data and, consequently, efficient data storage and analytic capacities. In an unprecedented approach, I will devote my postdoctoral research to implementing precision retail in health using a decade of individual consumer data from a large, population-based sample to develop efficient analytic methods to develop approaches to motivate healthy diets in vulnerable populations. ### Alexandre Payeur Supervisé.e par : Guillaume Lajoie Université de Montréal Identifying and guiding learning dynamics in the brain using brain-machine interfaces. Brain-machine interfaces (BMIs) are an emerging technology with great potential for helping patients with paralysis or motor disabilities. They rely on the brain’s learning capabilities and machine learning to enable brain circuits to control devices such as a computer cursor or a prosthetic arm. Beyond their clinical benefits, implanted BMIs offer unique access to the brain’s learning process itself, because the recorded neural activity exclusively controls the output. The situation is thus similar to artificial networks, i.e. relevant network states are observed and learning rules can be studied in an end-to-end fashion. As a tool for basic neuroscience, BMIs could provide deep insights into the principles of cortical learning. Adopting an approach at the interface of neuroscience and machine learning, we propose to exploit knowledge about the training and optimization of artificial neural networks to better understand learning in the motor cortex, and to develop algorithms that interact more seamlessly with the brain for robust BMIs. ### Ramesh Ramasamy Pandi Supervisé.e par : Yossiri Adulyasak HEC Montréal GPU-based Data-driven Framework for Real-time Dispatching of Autonomous Mobility-on-Demand. This IVADO project is primarily motivated by the emerging concept of Autonomous mobility-on-demand (AMoD). AMoD is a transformative and rapidly developing mode of mobility service that offers smart and efficient passenger transportation using self-driving vehicles while concurrently reducing the negative externalities such as congestion and pollution. The main goal of this project is to develop a generic framework for AMoD that integrates novel GPU-based data-driven algorithms to perform non-myopic real-time dispatching of large-scale ride-sharing systems that essentially deals with tens of thousands of customers per hour. Specifically, we plan to develop the GPU-accelerated optimization approaches, deep learning (DL) algorithms for efficient demand prediction, and DL-based predictive fleet control policies to perform non-myopic real-time dispatching of large-scale ride-sharing systems with multiple modes of vehicles. We will conduct simulations on real-world transportation networks to fully analyze the feasibility and effectiveness of the proposed framework. ### Wenshuo Wang Supervisé.e par : Lijun Sun McGill University Interaction-Aware Decision-Making for Autonomous Driving in Urban Environment. Autonomous driving will soon be sufficiently reliable and affordable to replace most human driving, providing independent mobility to non-drivers, reducing driver stress, and offering a panacea for urban problems. With recent advances in autonomous driving technology, prototype vehicles are already running on highways. However, given the enormous complexity of driving tasks in cities, it is widely acknowledged that fully autonomous will take decades to achieve for urban driving environments. Mixed traffic where autonomous vehicles share a traffic space with human drivers is inevitable before achieving fully autonomous. Therefore, it becomes a critical question to developing autonomous driving decision-making frameworks that can effectively learn and understand the intent of human drivers and adapt to their driving styles on public roads. The overall objective of this project is to develop a close-loop interaction-aware decision-making framework and algorithms for autonomous driving in a complex urban environment, with a particular focus on urban intersections, by leveraging human intent prediction. This project will achieve four sub-objectives progressively: (1) process video-based sequential data for complex scenarios, (2) learn multi-agent spatial interaction representations, (3) predict human driver intents in space and time to support decision making, (4) integrate a predictable decision-maker to form a close-loop interaction-aware framework. ### Funding of fundamental research projects ### Charles Audet (chercheur principal) Équipe : Sébastien Le Digabel, Michael Kokkolaras, Miguel Diage Martinez Polytechnique Montréal Combining machine learning and blackbox optimization for engineering design The efficiency of machine learning (ML) techniques relies on many mathematical foundations, one of which being optimization and its algorithms. Some aspects of ML can be approached using the simplex method, dynamic programming, line-search, Newton or quasi-Newton descent techniques. But there are many ML problems that do not possess an exploitable structure necessary for the application of the above methods. The objective of the present proposal is to merge, import, specialize and develop blackbox optimization (BBO) techniques in the context of ML. BBO considers problems in which the analytical expressions of the objective function and/or of the constraints defining an optimization are unavailable. The most frequent situation is when these functions are computed through a time-consuming simulation. These functions are often nonsmooth, contaminated by numerical noise and can fail to produce an usable output. Research in BBO is in constant growth since the last 20 years, and has seen a variety of applications in many fields. The research projects will be bidirectional. We plan to use and develop BBO techniques to improve the performance of ML algorithms. Conversely, we plan to deploy ML strategies to improve the efficiency of BBO algorithms. ### Julien Cohen-Adad, Polytechnique Montréal Équipe : Yoshua Bengio, Joseph Cohen, Nicolas Guizard, Kawin Setsompop, Anne Kerbrat, David Cadotte Physics-informed deep learning architecture to generalize medical imaging tasks The field of AI has flourished in recent years; in particular deep learning has shown unprecedented performance for image analysis tasks, such as segmentation and labeling of anatomical and pathological features. Unfortunately, while dozens of deep learning papers applied to medical imaging get published every year, most methods are tested in single-center: in the rare case where the code is publicly available, the algorithm usually fails when applied to other centers, which is the “real-world” scenario. This happens because images from different centers have different features than the images used to train the algorithm (contrast, resolution, etc.). Another issue limiting the performance potential of deep learning in medical imaging is that little data and few manual labels are available, and the labels are themselves highly variable across experts. The main objective of this project is to push the generalization capabilities of medical imaging tasks by incorporating prior information from MRI physics and from the inter-rater variability into deep learning architectures. A secondary objective will be to disseminate the developed methods to research and hospital institutions via open-source software (www.ivadomed.org), in-situ training and workshops. ### Patricia Conrod, Université de Montréal Équipe : Irina Rish, Sean Spinney A neurodevelopmentally-informed computational model of flexible human learning and decision making The adolescent period is characterized by significant neurodevelopmental changes which impact on reinforcement learning and the efficiency with which such learning occurs. Our team has modelled passive-avoidance learning using a bayesian reinforcement learning framework. Results indicated that parameters estimating individual differences in impulsivity, reward sensitivity, punishment sensitivity and working memory, best predicted human behaviour on the task. The model was also sensitive to year-to-year changes in performance (cognitive development), with individual components of the learning model showing different developmental growth patterns and relationships to health risk behaviours. This project aims to expand and validate this computer model of human cognition to: 1) Better measure neuropsychological age/delay; 2) understand how learning parameters contribute to human decision making processes on more complex learning tasks; 3) simulate better learning scenarios to inform development of targeted interventions that boost human learning and decision making; and 4) inform next generation artificial intelligence models of lifelong learning. ### Numa Dancause, Université de Montréal Équipe : Guillaume Lajoie, Marco Bonizzato Novel AI driven neuroprosthetics to shape stroke recovery Stroke is the leading cause of disability in occidental countries. After stroke, patients often have abnormally low activity in the part of the brain that controls movements, the motor cortex. However, the malfunctioning motor cortex receives connections from multiple spared brain regions. Our general hypothesis is that neuroprostheses interfacing with the brain can exploit these connections to help restore adequate motor cortex activation after stroke. In theory, brain connections can be targeted using new electrode technologies, but this problem is highly complex. It cannot be done by hand, one patient at a time. We need automated stimulation strategies to harness this potential for recovery. Our main objective is thus to develop an algorithm that efficiently finds the best residual connections to restore adequate excitation of the motor cortex after stroke. In animals, we will implant hundreds of electrodes in the diverse areas connected with the motor cortex. The algorithm will learn the pattern of stimulation that is the most effective to increase activity in the motor cortex. For the first time, machine learning will become a structural part of neuroprosthetic design. We will use these algorithms to create a new generation of neuroprostheses that act as rehabilitation catalyzers. ### Michel Denault, HEC Montréal Équipe : Dominique Orban, Pierre-OIivier Pineau Paths to a cleaner Northeast energy system through approximate dynamic programming Our main research question is the design of greener energy systems for the American Northeast (Canada and USA). Some of the sub questions are as follows. How can renewable energy penetrate the markets? Are supplementary power transmission lines necessary ? Can energy storage improve the intermittency problems of wind and solar power? Which greenhouse gases (GHG) reductions are achievable ? What is the cost of such changes ? Crucially, what is the path to a better system ? To support the transition to this new energy system, our proposition is : 1. to model the evolution of the Northeast power system as a Markov Decision process (MDP), including crucial uncertainties, e.g. on technological advances and renewable energy cost; 2. to solve this decision process with dynamic programming and reinforcement learning techniques; 3. to derive energy/environmental policy intelligence from our computational results. Our methodological approach relies on two building blocks, an inter-regional energy model and a set of algorithmic tools to solve the model as an MDP. ### Vincent Grégoire, HEC Montréal Équipe : Christian Dorion, Manuel Morales, Thomas Hurtut Learning the Dynamics of the Limit Order Book Modern financial markets are increasingly complex. A particular topic of interest is how this complexity affects how easily investors can buy or sell securities at a fair price. Many have also raised concerns that algorithms trading at high frequency could create excess volatility and crash risk. The central objective of our research agenda is to better understand the fundamental forces at play in those markets where trading speed is now measured in nanoseconds. Our project seeks to lay the groundwork, using big data, visualization, and machine learning, to answer some of the most fundamental questions in the literature on market structure. Ultimately, we envision an environment in which we could learn the behavior of the various types of agents in a given market. Once such an environment is obtained, it would allow us to better understand, for instance, the main drivers of major market disruptions. More importantly, it could allow us to guide regulators in the design of new regulations, by testing them in a highly realistic simulation setup, thereby avoiding the unintended consequences associated with potential flaws in the proposed regulation. ### Mehmet Gumus, McGill University Équipe : Erick Delage, Arcan Nalca, Angelos Georghiou Data-driven Demand Learning and Sharing Strategies for Two-Sided Online Marketplaces The proliferation of two-sided online platforms managed by a provider is disrupting the global retail industry by enabling consumers (on one side) and sellers (on the other side) to interact in exponential ways. Evolving technologies such as artificial intelligence, big data analytics, distributed ledger technology, and machine learning are posing challenges and opportunities for the platform providers with regards to understanding the behaviors of the stakeholders – consumers, and third-party sellers. In this proposed research project, we will focus on two-sided platforms for which demand-price relationship is unknown upfront and has to be learned from accumulating purchase data, thus highlighting the importance of the information-sharing environment. In order to address this problem, we will focus on the following closely connected research objectives: 1.Identify the willingness-to-pay and purchase decisions (i.e., conversion rate) of online customers based on how they respond to the design of product listing pages, online price and promotion information posted on the page, shipping and handling prices, and stock availability information. 2.Determine how much of the consumer data is shared with the sellers and quantify the value of different information sharing configurations – given the sellers’ optimal pricing, inventory (product availability), and product assortment (variety) decisions within a setting. ### Julie Hussin, Université de Montréal Équipe : Sébastien Lemieux, Matthieu Ruiz, Yoshua Bengio, Ahmad Pesaranghader Interpretability of Deep Learning Approaches Applied to Omics Datasets The high-throughput generation of molecular data (omics data) nowadays permits researchers to glance deeply into the biological variation that exists among individuals. This variation underlies the differences in risks for human diseases, as well as efficacy in their treatment. This requires combining multiple biological levels (multi-omics) through flexible computational strategies, including machine learning (ML) approaches, becoming highly popular in biology and medicine, with a particular enthusiasm for deep neural networks (DNNs). While it appears like a natural way to analyze complex multi-omics datasets, the application of such techniques to biomedical datasets poses an important challenge: the black-box problem. Once a model is trained, it can be difficult to understand why it gives a particular response to a set of data inputs. In this project, our goal is to train and apply state-of-the-art ML models to extract accurate predictive signatures from multi-omics datasets while focusing on biological interpretability. This will contribute to building the trust of the medical community in the use of these algorithms and will lead to deeper insights into the biological mechanisms underlying disease risk, pathogenesis and response to therapy. ### Jonathan Jalbert, Polytechnique Montréal Équipe : Françoise Bichai, Sarah Dorner, Christian Genest Modélisation des surverses occasionnées par les précipitations et développement d’outils adaptés aux besoins de la Ville de Montréal La contamination fécale des eaux de surface constitue l’une des premières causes de maladies d’origine hydrique dans les pays industrialisés et dans les pays en voie de développement. En zone urbaine, la contamination fécale provient majoritairement des débordements des réseaux d’égouts combinés. Lors de précipitations, l’eau pluviale entre dans le réseau d’égouts et se mélange à l’eau sanitaire pour être acheminée vers la station d’épuration. Si l’intensité des précipitations dépasse la capacité de transport du réseau, le mélange des eaux pluviales et sanitaires est alors directement rejeté dans le milieu récepteur sans passer par la station d’épuration. Ces débordements constituent un risque environnemental et un enjeu de santé publique. À l’heure actuelle, les caractéristiques des événements pluvieux occasionnant des surverses sont incertaines. Ce projet de recherche vise à tirer profit des données sur les surverses récemment rendues publiques par la Ville de Montréal pour caractériser les événements de précipitations occasionnant des surverses sur son territoire. Cette caractérisation permettra, d’une part, d’estimer le nombre de surverses attendues pour le climat projeté des prochaines décennies. D’autre part, elle sera utilisée pour dimensionner les mesures de mitigation, tels que les bassins de rétention et les jardins de pluie. ### Nadia Lahrichi, Polytechnique Montréal Équipe : Sébastien Le Digabel, Andrea Matta, Nicolas Zufferey, Andrea Lodi, Chunlong Yu Reactive/learning/self-adaptive metaheuristics for healthcare resource scheduling The goal of this research proposal is to develop state-of-the-art decision support tools to address the fundamental challenges of accessible and quality health services. The challenges to meeting this mandate are real, and efficient resource management is a key factor in achieving this goal. This proposal will specifically focus on applications related to patient flow. Analysis of the literature shows that most research focuses on single-resource scheduling and considers that demand is known; Patient and resource scheduling problems are often solved sequentially and independently. The research goal is to develop efficient metaheuristic algorithms to solve integrated patient and resource scheduling problems under uncertainty (e.g., demand, prole, and availability of resources). This research will be divided into three main themes, each of them investigating a different avenue to more efficient metaheuristics: A) learning approaches to better explore the search space; B) blackbox optimization for parameter tuning; and C) simulation-inspired approaches to control the noise induced by uncertainty. ### Eric Lécuyer, Université de Montréal Équipe : Mathieu Blanchette, Jérôme Waldispühl, William Hamilton Deciphering RNA regulatory codes and their disease-associated alterations using machine learning The human DNA genome serves as an instruction guide to allow the formation of all the cells and organs that make up our body over the course of our lives. Much of this genome is transcribed into RNA, termed the ‘transcriptome’, that serves as a key conveyor of genetic information and provides the template for the synthesis of proteins. The transcriptome is itself subject to many regulatory steps for which the basic rules are still poorly understood. Importantly, when these steps are improperly executed, this can lead to disease. This project aims to utilize machine learning approaches to decipher the complex regulatory code that controls the human transcriptome and to predict how these processes may go awry in different disease settings. ### Gregory Lodygensky, Université de Montréal Équipe : Jose Dolz, Josée Dubois, Jessica Wisnowski Next generation neonatal brain segmentation built on HyperDense-Net, a fully automated real-world tool There is growing recognition that major breakthroughs in healthcare will result from the combination of databanks and artificial intelligence (AI) tools. This would be very helpful in the study of the neonatal brain and its alterations. For instance, the neonatal brain is extremely vulnerable to the biological consequences of prematurity or birth asphyxia, resulting in cognitive, motor, language and behavioural disorders. A key difference with adults is that key aspects of brain-related functions can only be tested several years later, hindering greatly the advancement of neonatal neuroprotection. Researchers and clinicians need objective tools to immediately assess the effectiveness of a therapy that is given to protect the brain without waiting five years to see if it succeeded. Neonatal brain magnetic resonance imaging can bridge this gap. However, it represents a real challenge as this period of life represents a unique period of intense brain growth (e.g. myelination and gyrification) and brain maturation. Thus, we plan to improve our existing neonatal brain segmentation tools (i.e. HyperDense-Net) using the latest iterations of AI tools. We will also develop a validated tool to determine objective brain maturation in newborns. ### Adam Oberman, McGill University Équipe : Michael Rabbat, Chris Finlay, Levon Nurbekyan Robustness and generalization guarantees for Deep Neural Networks in security and safety critical applications Despite impressive human-like performance on many tasks, deep neural networks are surprisingly brittle in scenarios outside their previous experience, often failing when new experiences do not closely match their previous experiences. This ‘failure to generalize’ is a major hurdle impeding the adoption of an otherwise powerful tool in security- and safety-critical applications, such as medical image classification. The issue is in part due to a lack of our theoretical understanding of why neural networks work so well. They are powerful tools but less interpretable than traditional machine learning methods which have performance guarantees but do not work as well in practice. This research program will aim to address this ‘failure to generalize’, by developing guarantees of generalization, using notions of the complexity of a regularized model, corresponding to model averaging. This approach will be tested in computer vision applications, and will have near-term applications to medical health research, through medical image classification and segmentation. More broadly, the data science methods developed under this project will be applicable to a wide variety of fields and applications, notably wherever reliability and safety are paramount. ### Liam Paull, Université de Montréal Équipe : Derek Nowrouzezahrai, James Forbes Differentiable perception, graphics, and optimization for weakly supervised 3D perception An ability to perceive and understand the world is a prerequisite for almost any embodied agent to achieve almost any task in the world. Typically, world representations are hand-constructed because it is difficult to learn them directly from sensor signals. In this work, we propose to build the components so that this map-building procedure is differentiable. Specifically, we will focus on the perception (grad-SLAM) and the optimization (meta-LS) components. This will allow us to backpropagate error signals from the 3D world back to the sensor inputs. This enables us to do many things, such as regularize sensor data with 3D geometry. Finally, by also building a differentiable rendering component (grad-Sim), we can leverage self-supervision through cycle consistency to learn representations with no or sparse hand-annotated labels. Combining all of these components together gives us the first method of world representation building that is completely differentiable and self-supervised. ### Gilles Pesant, Polytechnique Montréal Équipe : Siva Reddy, Sarath Chandar Anbil Parthipan Investigating Combinations of Neural Networks and Constraint Programming for Structured Prediction L’intelligence artificielle occupe une place de plus en plus importante dans de nombreuses sphères d’activité et dans notre quotidien. En particulier, les réseaux de neurones arrivent maintenant à assimiler puis à accomplir des tâches auparavant réservées aux humains. Cependant lorsqu’une tâche nécessite le respect de règles structurantes complexes, un réseau de neurones éprouve parfois beaucoup de mal à apprendre ces règles. Or un autre domaine de l’intelligence artificielle, la programmation par contraintes, a précisément été conçue pour trouver des solutions respectant de telles règles. Le but de ce projet est donc d’étudier des combinaisons de ces deux approches à l’intelligence artificielle afin de plus facilement apprendre à accomplir des tâches sous contraintes. Dans le cadre du projet, nous nous concentrerons sur le domaine du traitement de la langue naturelle mais nos travaux pourraient aussi s’appliquer à des tâches dans d’autres domaines. ### Jean-François Plante, HEC Montréal Équipe : Patrick Brown, Thierry Duchesne, Nancy Reid, Luc Villandré Statistical inference and modelling for distributed systems Statistical inference requires a large toolbox of models and algorithms that can accommodate complex data structures. Modern datasets are often so large that they need to be stored on distributed systems, with the data stored across a number of nodes with limited bandwidth between them. Many complex statistical models cannot be used with such complex data, as they rely on the complete data being accessible. In this project, we will advance statistical modeling contributions to data science by creating solutions that are ideally suited for analysis on distributed systems. More specifically, we will develop spatio-temporal models as well as accurate and efficient approximations of general statistical models that are suitable for distributed data, and as such, scalable to massive data. ### Wei Qi, McGill University Équipe : Xue (Steve) Liu, Max Shen, Michelle Lu Deals on Wheels: Advancing Joint ML/OR Methodologies for Enabling City-Wide, Personalized and Mobile Retail Moving forward to a smart-city future, cities in Canada and around the world are embracing the emergence of new retail paradigms. That is, retail channels can further diversify beyond the traditional online and offline boundaries, combining the best of the both. In this project, we focus on an emerging mobile retail paradigm in which retailers run their stores on mobile vehicles or self-driving cars. Our mission is to develop cross-disciplinary models, algorithms and data-verified insights for enabling mobile retail. We will achieve this mission by focusing on three interrelated research themes: Theme 1 – Formulating novel optimization problems of citywide siting and inventory replenishment for mobile stores. Theme 2 – Developing novel learning models for personalized demand estimation. Theme 3 – Integrating Theme 1 and Theme 2 by proposing a holistic algorithmic framework for joint and dynamic demand learning and retail operations, and for discovering managerial insights. The long-term goal is to thereby advance the synergy of operations and machine learning methodologies in the broad contexts of new retail and smart-city analytics. ### Marie-Ève Rancourt, HEC Montréal Équipe : Gilbert Laporte, Aurélie Labbe, Daniel Aloise, Valérie Bélanger, Joann de Zegher, Burcu Balçik, Marilène Cherkesly, Jessica Rodríguez Pereira Humanitarian Supply Chain Analytics Network design problems lie at the heart of the most important issues faced in the humanitarian sector. However, given their complex nature, humanitarian supply chains involve the solution of difficult analytics problems. The main research question of this project is “how to better analyze imperfect information and address uncertainty to support decision making in humanitarian supply chains?”. To this end, we propose a methodological framework combining data analysis and optimization, which will be validated through real-life applications using multiple sources of data. First, we propose to build robust relief networks under uncertainty in demand and transportation accessibility, due to weather shocks and vulnerable infrastructures. We will consider two contexts: shelter location in Haiti and food aid distribution planning in Southeastern Asia. Second, we propose to embed fair cost sharing mechanisms into a collaborative prepositioning network design problem arising in the Caribbean. Classic economics methods will be adapted to solve large-scale stochastic optimization problems, and novel models based on catastrophic insurance theory will be proposed. Finally, a simulation will be developed to disguise data collection as a serious game and gather real-time information on the behavior of decision makers during disasters to extrapolate the best management strategies. ### Saibal Ray, McGill University Équipe : Maxime Cohen, James Clark, AJung Moon Retail Innovation Lab: Data Science for Socially Responsible Food Choices In this research program, we propose to investigate the use of artificial intelligence techniques, involving data, models, behavioral analysis, and decision-making algorithms, to efficiently provide higher convenience for retail customers while being socially responsible. In particular, the research objective of the multidisciplinary team is to study, implement, and validate systems for guiding customers to make healthy food choices in a convenience store setting, while being cognizant of privacy concerns, both online and in a brick-and-mortar store environment. The creation of the digital infrastructure and decision support systems that encourage people and organizations to make health-promoting choices should hopefully result in a healthier population and reduce the costs of chronic diseases to the healthcare system. These systems should also foster the competitiveness of organizations operating in the agri-food and digital technology sectors. A distinguishing feature of this research program is that it will make use of a unique asset – a new “living-lab”, the McGill Retail Innovation Lab (MRIL). It will house a fully functioning retail store operated by a retail partner with extensive sensing, data access, and customer monitoring. The MRIL will be an invaluable source of data to use in developing and validating our approaches as well as a perfect site for running field experiments. ### Léo Raymond-Belzile, HEC Montréal Équipe : Johanna Nešlehová, Alexis Hannart, Jennifer Wadsworth Combining extreme value theory and causal inference for data-driven flood hazard assessment The IPCC reports highlight an increase in mean precipitation, but the impact of climate change on streamflow is not as certain and the existing methodology is ill-equipped to predict changes in flood extremes. Our project looks into climate drivers impacting flood hazard and proposes methodological advances based on extreme value theory and causal inference in order to simulate realistic streamflow extremes at high resolution. The project will also investigate how climate drivers impact the hydrological balance using tools from machine learning for causal discovery to enhance risk assessment of flood hazard. ### Nicolas Saunier, Polytechnique Montréal Équipe : Francesco Ciari, Catherine Morency, Martin Trépanier, Lijun Sun Bridging Data-Driven and Behavioural Models for Transportation Transportation data is traditionally collected through travel surveys and fixed sensors, mostly on the roadways: such data is expensive to collect and has limited spatial and temporal coverage. In recent years, more and more transportation data has become available on a continuous basis from multiple new sources, including users themselves. This has fed the rise of machine learning methods that can learn models directly from data. Yet, such models often lack robustness and may be difficult to transfer to a different region or period. This can be alleviated by taking advantage of domain knowledge stemming from the properties of the flow of people moving in transportation systems with daily activities. This project aims to develop hybrid methods relying on transportation and data-driven models to predict flows for all modes at different spatial and temporal scales using multiple sources of heterogeneous data. This results in two specific objectives: 1. to learn probabilistic flow models at the link level for several modes based on heterogeneous data; 2. to develop a method bridging the flow models (objective 1) with a dynamic multi-agent transportation model at the network level. These new models and methods will be developed and tested using real transportation data. ### Yvon Savaria, Polytechnique Montréal Équipe : François Leduc-Primeau, Elsa Dupraz, Jean-Pierre David, Mohamad Sawan Ultra-Low-Energy Reliable DNN Inference Using Memristive Circuits for Biomedical Applications (ULERIM) Recent advances in machine learning based on deep neural networks (DNNs) have brought powerful new capabilities for many signal processing tasks. These advances also hold great promises for several applications in healthcare. However, state-of-the-art DNN architectures may depend on hundreds of millions of parameters that must be stored and then retrieved, resulting in a large energy usage. Thus, it is essential to reduce their energy consumption to allow in-situ computations. One possible approach involves using memristor devices, a concept first proposed in 1971 but only recently put in practice. Memristors are a very promising way to implement compact and energy-efficient artificial neural networks. The aim of this research is to advance the state-of-the-art in the energy-efficient implementation of deep neural networks using memristive circuits and introducing DNN-specific methods to better manage uncertainty inherent to integrated circuit fabrication. These advances will benefit a large number of medical applications for which portable devices are required to perform a complex analysis of the state of the patient, and also benefit generally the field of machine learning by reducing the amount of energy required to apply it. Within this project, the energy improvements will be exploited to improve the signal processing performance of an embedded biomedical device for the advanced detection of epileptic seizures. ### Alexandra M. Schmidt, McGill University Équipe : Jill Baumgartner, Brian Robinson, Marília Carvalho, Oswaldo Cruz, Hedibert Lopes Flexible multivariate spatio-temporal models for health and social sciences Health and social economic variables are commonly observed at different spatial scales of a region (e.g. districts of a city or provinces of a country), over a given period of time. Commonly, multiple variables are observed at a given spatial unit resulting in high dimensional data. The challenge in this case is to consider models that account for the possible correlation among variables across space or space and time. This project aims at developing statistical methodology that accounts for this complex hierarchical structure of the observed data. And inference procedure follows the Bayesian paradigm meaning that uncertainty about the unknowns in the model is naturally accounted for. The project is subdivided into four sub projects that range from the estimation of a social economic vulnerability index for a given city to the spatio-temporal modelling of multiple vector borne diseases. The statistical tools proposed here will help authorities with the understanding of the dynamics across space and time of multiple diseases, and assist with the decision making process of evaluating how urban policies and programmes will impact the urban environment and population health, through a lens of health equity. ### David Stephens, McGill University Équipe : Yu Luo, Erica Moodie, David Buckeridge, Aman Verma Statistical modelling of health trajectories and interventions Large amounts of longitudinal health records are now collected in private and public healthcare systems. Data from sources such as electronic health records, healthcare administrative databases and data from mobile health applications are available to inform clinical and public health decision-making. In many situations, such data enable the dynamic monitoring of the underlying disease process that governs the observations. However, this process is not observed directly and so inferential methods are needed to ascertain progression. The objective of the project is to build a comprehensive Bayesian computational framework for performing inference for large scale health data. In particular, the project will focus on the analysis of records that arise in primary and clinical care contexts to study patient health trajectories, that is, how the health status of a patient changes across time. Having been able to infer the mechanisms that influence health trajectories, we will then be able to introduce treatment intervention policies that aim to improve patient outcomes. ### An Tang, Université de Montréal Équipe : Irina Rish, Guy Wolf, Guy Cloutier, Samuel Kadoury, Eugene Belilovsky, Michaël Chassé, Bich Nguyen Ultrasound classification of chronic liver disease with deep learning Chronic liver disease is one of the top ten leading causes of death in North America. The most common form is nonalcoholic fatty liver disease which may evolve to nonalcoholic steatohepatitis and cirrhosis if left untreated. In many cases, the liver may be damaged without any symptoms. A liver biopsy is currently required to evaluate the severity of chronic liver disease. This procedure requires the insertion of a needle inside the liver to remove a small piece of tissue for examination under a microscope. Liver biopsy is an invasive procedure with a risk of major complications such as bleeding. Ultrasound is ideal for screening patients because it is a safe and widely available technology to image the whole liver. Our multi-disciplinary team is proposing the use of novel artificial intelligence techniques to assess the severity of chronic liver disease from ultrasound images and determine the severity of liver fat, inflammation, and fibrosis without the need for liver biopsy. This study is timely because chronic liver disease is on the rise which means that complications and mortality will continue to rise if there is no alternative technique for early detection and monitoring of disease severity. ### Guy Wolf, Université de Montréal Équipe : Will Hamilton, Jian Tang Unified approach to graph structure utilization in data science While deep neural networks are at the frontier of machine learning and data science research, their most impressive results come from data with clear spatial/temporal structure (e.g., images or audio signals) that informs network architectures to capture semantic information (e.g., textures, shapes, or phonemes). Recently, multiple attempts have been made to extend such architectures to non-Euclidean structures that typically exist in data, and in particular to graphs that model data geometry or interaction between data elements. However, so far, such attempts have been separately conducted by largely-independent communities, leveraging specific tools from traditional/spectral graph theory, graph signal processing, or applied harmonic analysis. We propose a multidisciplinary unified approach (combining computer science, applied mathematics, and decision science perspectives) for understanding deep graph processing. In particular, we will establish connections between spectral and traditional graph theory applied for this task, introduce rich notions of intrinsic graph regularity (e.g., equivalent to image textures), and enable continuous-depth graph processing (i.e., treating depth as time) to capture multiresolution local structures. Our computational framework will unify the multitude of existing disparate attempts and establish rigorous foundations for the emerging field of geometric deep learning, which is a rapidly growing field in machine learning. ### COVID-19: IVADO projects and initiatives ### Projects funded by IVADO ### Digital clinical trials to accelerate the evaluation of colchicine therapy Team : Digital clinical trials to accelerate the evaluation of colchicine therapy Jean-Claude Tardif (Director, Research Centre, Montréal Heart Institute and Professor, Université de Montréal) and Frédéric Lesage (Professor, Polytechnique Montréal) This project, which has already received approval from Health Canada, the Québec Ministry of Health and Social Services and the Montréal Heart Institute’s Ethics Committee, seeks to evaluate a colchicine-based treatment, including its impact on mortality rates and pulmonary complications. The scope of the task (recruitment of a cohort of 6,000 subjects) and the extremely tight deadline (as quickly as possible) for this type of project require the implementation of new digital recruitment and follow-up tools. Funding amounts: IVADO:$125,000
scale ai: $100,000 TransMedTech Institute:$50,000

Colcorona clinical trial

Polytechnique Montréal news release

### Accelerating the search for a drug for COVID-19

Team : Yoshua Bengio (Scientific Director, Mila and IVADO, and Professor, Université de Montréal) and Mike Tyers (Principal Investigator, IRIC)

In a prerequisite for drug development, this project seeks to identify molecules that may specifically associate with SARS-CoV-2. To do this, researchers from Mila and IRIC will first use neural networks to automatically generate billions of potential molecules. An enhancement algorithm will then be used to select the most promising ones for biological evaluation and possible clinical trials.

Funding amounts:
IVADO: $100,000 scale ai:$125,000

### Modelling of animal reservoirs of pathogens

Team : Timothée Poisot (Assistant Professor, Université de Montréal) and Colin Carlson (Visiting Professor, Université de Montréal)

The current COVID-19 pandemic, like others before it, originated in a host animal. However, the ecology, origin and development of these hosts and their viruses remain largely unknown. In order to address this shortcoming, this project aims to model animal populations that act as reservoirs for these pathogens in order to complete knowledge of the disease and to anticipate future resurgences, or outbreaks of new viruses.

IVADO: $11,000 ### Covid-19 critical-care digital visualization board Team : Philippe Doyon-Poulin (Chercheur IVADO et professeur adjoint, Polytechnique) et Philippe Jouvet (Intensiviste pédiatre du CHU Sainte-Justine et professeur titulaire de clinique, Université de Montréal) In a pandemic, the number of intensive care inpatients increases rapidly and the management of medical resources is critical to the success of care. The purpose of this project is to develop a digital board to visualize the health status of patients in intensive care units and the allocation of medical resources so as to respond in real time to the needs produced by the COVID-19 crisis. This digital tool will be transferred to the Pediatric Intensive Care Unit at CHU Sainte-Justine and the Intensive Care Unit at the Jewish General Hospital. IVADO funding amount:$30,600

Polytechnique Montréal article

### Identifying the Achilles heel of SARS-CoV-2

François Major (Principal Investigator, IRIC)

Using an algorithm based on machine-learning techniques, this project seeks to develop a protocol for better understanding the structural components involved in the vital functions of SARS-CoV-2 or any other RNA virus. This technique will make it possible to produce a list of therapeutic targets to counter their replication and proliferation, thus offering new perspectives for the development of drugs to be used in current or future clinical studies.

### Interconnecting COVID-19 data

Team : Interconnecting COVID-19 data David Ardia (Researcher, IVADO and Assistant Professor, HEC Montréal) and Emanuele Guidotti (PhD Student, Université de Neuchâtel)

Numerous COVID-19-related databases exist, but no virtual platform currently incorporates a significant proportion of these sources. This makes it difficult to do a global analysis of them, and to make connections between this often-medical information and external factors, especially socio-political ones. In this perspective, this international project aims to develop a multifactorial open-source platform, enabling the integration and continuous addition of new information.

IVADO funding amount: $10,000 Learn more ### Projects supported by IVADO ### Interactive therapeutic target-prediction portal Team : Tariq Daouda (Postdoctoral Researcher, Massachusetts General Hospital – Harvard Medical School) and Maude Dumont-Lagacé (Scientific Coordinator, ExCellThera) The goal of this project is to provide the scientific community with a platform to predict potential targets for a vaccine against COVID-19. This interactive platform uses an algorithm’s ability to predict which parts of the virus will be exposed on the surface of infected cells and thus generates a list of potential targets. This algorithm, developed by Tariq Daouda in the laboratories of Sébastien Lemieux and Claude Perreault, has already been used successfully, enabling the current situation to be approached from a different angle. Offered to researchers through a portal, it will make it possible to accelerate the development of vaccines against COVID-19, but also against other emerging viruses. News release ### Lightening the healthcare community’s load through dialogue systems Alexis Smirnov (CTO, Dialogue) Many telemedicine tasks (such as responding to 811) involve healthcare professionals. This project proposes to set up several standalone telephone assistance solutions to free up these experts who are currently in high demand, whether to answer citizens’ routine questions, do follow-ups, make appointments or help navigate through healthcare facilities. Funding amount:$500,000

Find out more

### Improved prognosis using chest X-rays

Joseph Paul Cohen (Postdoctoral Researcher, Université de Montréal)

Chester is an existing prototype of a radiology assistant that can recognize certain pneumonia-related characteristics. During the current pandemic, this project aims to improve Chester’s disease predictions with the aim of enhancing the management of patient care. How will this be done? By combining artificial intelligence and image recognition, while widely disseminating a public database of clinical metadata for a large number of COVID-19 cases (as well as SARS and other pneumonia cases).

Find out more

Ressources

### Julie Hussin

Senior Researcher, ICM

“This project aims to analyze the viral sequences at different stages of the evolution of the virus and thus identify indicators associated with the geographical regions where patients have tested positive for COVID-19.”

“Data-efficient deep learning to better model immune response: (…) building an open-source platform leveraging the latest AI technologies to model pathways in the immune system in order to better predict immune response. (…) we work on AI approaches that can contribute to the process of vaccine design(…)”.

Find out more

### Michaël Chassé

Researcher, CHUM

• Creation of a biobank

“The main objective of this Québec-wide infrastructure is to provide researchers with the samples and data they need for their work. This will facilitate the co-ordination of research and support efforts for the development of new disease biomarkers, with a view to creating vaccines and drugs.”

Find out more

### Guy Wolf

Assistant Professor, Université de Montréal

• Omics profiling of COVID-19 progression mechanisms and specific analysis of immune responses in young patients

“This project [will] provide a mechanistic understanding of SARS-CoV-2 virus progression to assess the risk of specific medical profiles and patients, as well as to help identify binding targets for potential antiviral agents and vaccines. (…) An example of an active research question is to understand the apparent resilience of young children to severe infection, which is somewhat atypical for such epidemics.”

Find out more

### AlayaCare

Creation of a new, free COVID-19 screening device with a self-administered questionnaire assessing healthcare workers’ symptoms prior to a client visit.

Find out more

### Brainbox AI

Creation of a free HVAC (heating, ventilation and air conditioning) optimization service in response to COVID-19, using a “zone by zone” approach supported by cloud computing technologies.

Find out more

### CAI Global

Establishment of an economic and industrial impact forecasting model that, with the help of private databases, evaluates each economic sector of a city, region, RCM or other, in order to describe the situation and its risk factors.

Find out more

### EdLive

EdLive makes its distance-learning technology available to schools and businesses. Thanks to this initiative and the collaboration of EdLive partners, several thousand students across Québec are taking courses online.

Find out more

### IBM

L’assistant Watson pour les citoyens est maintenant disponible gratuitement pour aider les gouvernements et les institutions de santé à répondre aux questions courantes sur la COVID-19.

Apprenez-en plus

### Institut national du sport du Québec

Mental health capsules online. Tips and strategies for better coping with this high-risk period for stress.

Find out more

### Streamscan

Implementation of a free cybersecurity monitoring service to safeguard the security of companies’ and organizations’ IT equipment during this crisis.

Find out more

### Thales

Launch of a COVID-19 rapid response call by Thales and its artificial intelligence (AI) research centre cortAIx.

Find out more

### Valital

Free use of the Valital recruitment platform to more quickly find candidates or volunteers in the medical and research fields.

Find out more

### Imene Abid

Supervisé.e par : Pierre-Majorique Léger

HEC Montréal

Générateur de données synthétiques pour améliorer l’apprentissage en science des données

### Simon Chamorro

Supervisé.e par : Christopher Pal

Polytechnique Montréal

Navigational Assistant for the Visually Impaired (NAVI)

Supervisé.e par : Chahé Nerguizian

Polytechnique Montréal

Méthodes d’apprentissage automatique dans l’aide à l’élaboration de plans d’interventions en éducation

### Omar Chikhar

Supervisé.e par : Marc Fredette

HEC Montréal

Automation of signal processing methods for feature construction on physiological signals

### Léo Choinière

Supervisé.e par : Julie Hussin

Institut de cardiologie de Montréal (ICM)

Traitement de Données Génomiques par Différentes Architectures de Réseaux de Neurones.

### Anas Bouziane

Polytechnique Montréal

Reverse-engineering of and migration towards scalable NoSQL data architecture

### Valérie Daigneault

Supervisé.e par : Frédéric Gosselin

Université de Montréal

Intégration et traitement temporel de la vision dans le cerveau lors de la reconnaissance d’attributs faciaux : modélisation de données MEG et comportementales en apprentissage machine.

### Etienne Denis

Supervisé.e par : William Hamilton

McGill University

Multi-Relational Link Prediction Using Graph Neural Networks (SEARL)

### David Teddy Diffo Nguemetsing

Supervisé.e par : Numa Dancause

Université de Montréal

Learning algorithms for functional cortical neurostimulation

### Sandra Ferland

Supervisé.e par : Cisek Paul

Université de Montréal

Les mécanismes neuronaux de la prise de décision

### Aude Forcione-Lambert

Supervisé.e par : Guy Wolf

Université de Montréal

Probing learned network structure in a multi-task setting

### Dominique Fournelle

Supervisé.e par : Julie Hussin

Université de Montréal

Annotation des chromosomes sexuels de l’ornithorynque par apprentissage automatique

### Martine Francoeur

Supervisé.e par : Olivier Bahn

HEC Montréal

Projet de stage sur la modélisation du secteur énergétique du Mexique

### Enora Georgeault

Supervisé.e par : Marie-Ève Rancourt

HEC Montréal

Modèles prédictifs de l’allocation des dons de la Croix-Rouge canadienne en réponse aux feux de forêt

### William Glazer-Cavanagh

Polytechnique Montréal

Automatic integration and deployment of AI models

### Alexandre Gravel

Supervisé.e par : Bernard Gendron

Université de Montréal

Méthodes lagrangiennes pour la résolution de problèmes de conception de réseaux

### Rose Guay Hottin

Supervisé.e par : Marina Martinez

Université de Montréal

Un agent d’apprentissage pour une neuroprothèse cortico-spinale Marina

### Alice-Marie Hamelin

Supervisé.e par : Michel Gamache

Polytechnique Montréal

Outil de planification en temps réel pour les mines souterraines

### Jérémie Huppé

Supervisé.e par : Maleknaz Nayebi

Polytechnique Montréal

Automated communication analysis for Software-aided emergency management

### Arnaud L’Heureux

Supervisé.e par : Alain Tapp

Université de Montréal

Utilisation de réseau profond pour la simplification automatique de textes

### Julien Leissner-Martin

Supervisé.e par : Jean-François Arguin

Université de Montréal

Using Deep Learning to Identify Electrons at the Large Hadron Collider (LHC)

### Anthony Lemieux

Supervisé.e par : Serge McGraw

Centre hospitalier universitaire Mère-Enfant (CHU Sainte-Justine)

Investigation de dérégulations épigénétiques héritables par approches computationnelles

### Rui Ze Ma

Supervisé.e par : Franz Bernd Lang

Université de Montréal

Investigation of systematic errors in genome assembly algorithms

### Mohammed Mahmoud

Supervisé.e par : Mohamed Ouali

Polytechnique Montréal

Prediction of Fiber Quantity and Quality in Forest Supply Chains Using Artificial intelligence Methods

### Filip Milisav

Supervisé.e par : Karim Jerbi

Université de Montréal

Studying social influence using a neuroimaging and data science approach

### Alexandre Morinvil

Supervisé.e par : Giovanni Beltrame

Polytechnique Montréal

IA sécuritaire dans les essaims de drones : Développer une approche permettant aux petits essaims de drones de suivre les humains en toute sécurité

### Derek Ojeda Centeno

Supervisé.e par : Brunilde Sansò

Polytechnique Montréal

Simulation multiniveaux pour les applications des villes intelligentes

### Pierrick Pascal

Supervisé.e par : Sébastien Le Digabel

Polytechnique Montréal

Création d’une interface Julia pour NOMAD pour l’ajustement automatique des hyper-paramètres d’algorithmes d’optimisation.

### Justin Pelletier

Supervisé.e par : Julie Hussin

Institut de cardiologie de Montréal (ICM)

Évaluation de scores de risque polygénique selon le sexe et la structure populationnelle

### Pierre-Elie Personnaz

Supervisé.e par : Dominique Orban

Polytechnique Montréal

Traitement de la dégénérescence par régularisation en optimisation continue

### Marie-Eve Picard

Supervisé.e par : Pierre Jolicoeur

Université de Montréal

Analyses multivariées des interactions entre différents processus attentionnels (EEG): une approche orientée sur les données.

### Myriam Prasow-Émond

Supervisé.e par : Julie Hlavacek-Larrondo

Université de Montréal

Étude de l’amas de galaxies supermassif MACSJ1447.7+0827

Supervisé.e par : Jean-Francois Cordeau

HEC Montréal

Évaluation empirique de méthodes de prévision de la demande

Supervisé.e par : Brunilde Sansò

Polytechnique Montréal

Visualisation et Analyse de données des réseaux des villes intelligentes.

Supervisé.e par : Liam Paull

Université de Montréal

Duckietown AI Driving Olympics.

### Daniel Tomasso

Supervisé.e par : Dang Khoa Nguye

Centre hospitalier de l’Université de Montréal (CHUM)

Epileptic seizure detection by combining smart wear monitoring and artificial intelligence techniques.

### Fama Tounkara

Supervisé.e par : Franco Lepore

Université de Montréal

Validation d’une batterie de tests visuels comme aide au diagnostic de troubles neurologiques.

### Étienne Tremblay

Supervisé.e par : Réjean Plamondon

Polytechnique Montréal

Application heuristique des sciences des données à la théorie cinématique des mouvements humains.

### Anton Volniansky

Supervisé.e par : Jean-François Tanguay

Institut de cardiologie de Montréal (ICM)

Banque de données des issues cliniques à court et long termes des Échafaudages Vasculaires Biorésorbables comparativement aux Stents pharmacoactifs de 2e génération

Supervisé.e par : Brunilde Sansò

Polytechnique Montréal

Performances des micro-PMUs dans les Villes Intelligentes

### Tiphaine Bonniot de Ruisselet

Supervisé.e par : Dominique Orban

Polytechnique Montréal

Accélération de méthodes d’optimisation pour les problèmes volumineux par évaluation inexact

Nous nous intéressons aux problèmes d’optimisation continue, non convexe et sans contraintes dans lesquels l’évaluation des valeurs de l’objectif et de son gradient sont obtenues à l’issue d’un processus coûteux. Nous supposons que l’on peut obtenir à moindre coûts des approximations de l’objectif et de son gradient à un niveau de précision souhaité. Nous regarderons l’impact de ces hypothèses sur la convergence et la complexité de méthodes d’optimisation classiques ainsi que les économies pouvant être effectuées sur le temps de calcul et la consommation énergétique. Cette étude est motivée, entre autre, par les problèmes d’inversion sismique dont la taille peut avoisiner les centaines de millions de variables et dont la fonction et le gradient peuvent être approximés par la résolution d’un problème aux moindres carrés linéaires. L’économie de temps de calcul et d’énergie est un enjeu majeur de l’ère de l’intelligence artificielle et de l’exploration des données volumineuses et cette approche est nouvelle est prometteuse en terme de retombées économiques et environnementales.

### Stephanie Cairns

McGill University

Oberman Mathematical approaches to adversarial robustness and confidence in DNN

Deep convolutional neural networks are highly effective at image classification tasks, achieving higher accuracy than conventional machine learning methods but lacking the performance guarantees associated with these methods. Without additional performance guarantees, for example error bounds, they cannot be safely used in applications where errors can be costly. There is a consensus amongst researchers that greater interpretability and robustness are needed. Robustness can be to differences in the data set where the models can be deployed, or even robustness to adversarial samples: perturbations of the data designed deliberately by an adversary to lead to a misclassification.

In this project, we will study reliability in two contexts: (i) developing improved confidence in the prediction of the neural network, using modified losses to improve confidence measures (ii) modified losses which result in better robustness to adversarial examples. The overall goal of the project is to lead to more reliable deep learning models.

### Enora Georgeault

Supervisé.e par : Marie-Ève Rancourt

HEC Montréal

Modèles prédictifs de l’allocation des dons de la Croix-Rouge canadienne en réponse aux feux de forêt

Au Canada, les inondations et les feux de forêt sont les catastrophes naturelles qui provoquent le plus de dégâts. Les efforts de la Croix-Rouge canadienne (CRC) visant à atténuer les impacts des feux de forêt dépendent fortement de la capacité des organisations à planifier, à l’avance, les opérations logistiques de secours. Le premier objectif du projet est l’élaboration de modèles permettant de prédire l’allocation des dons en argent aux bénéficiaires, selon les caractéristiques socio-démographiques de la région et du bénéficiaire ainsi que selon les caractéristiques des feux (sévérité et type). Le second objectif sera de comprendre les facteurs qui ont un impact significatif sur les besoins de la CRC lors d’une réponse à un feu de forêt, afin de faciliter la planification des opérations logistiques et les appels de financement.

### Bhargav Kanuparthi

Supervisé.e par : Yoshua Bengio

Université de Montréal

h detach Modifying the LSTM Gradient Towards Better Optimization

Recurrent neural networks are known for their notorious exploding and vanishing gradient problem (EVGP). This problem becomes more evident in tasks where the information needed to correctly solve them exist over long time scales, because EVGP prevents important gradient components from being back-propagated adequately over a large number of steps. We introduce a simple stochastic algorithm (\textit{h}-detach) that is specific to LSTM optimization and targeted towards addressing this problem. Specifically, we show that when the LSTM weights are large, the gradient components through the linear path (cell state) in the LSTM computational graph get suppressed. Based on the hypothesis that these components carry information about long term dependencies (which we show empirically), their suppression can prevent LSTMs from capturing them. Our algorithm\footnote{Our code is available at https://github.com/bhargav104/h-detach.} prevents gradients flowing through this path from getting suppressed, thus allowing the LSTM to capture such dependencies better. We show significant improvements over vanilla LSTM gradient based training in terms of convergence speed, robustness to seed and learning rate, and generalization using our modification of LSTM gradient on various benchmark datasets.

### Vincent Labonté

Supervisé.e par : Michel Gagnon

Polytechnique Montréal

Extraction de connaissances en français basée sur une traduction des textes en anglais combinée à l’utilisation d’outils développés pour l’anglais

Plusieurs institutions gouvernementales rendent disponible sur leurs sites web un très grand volume de documents qui ne sont écrits que dans la langue officielle du pays. Or, de plus en plus ces institutions désirent transformer ces documents en une base de connaissances, déployée en un ensemble de données ouvertes intégrées au Web sémantique. C’est le cas notamment du ministère de la Culture et des Communications du Québec, qui met à la disposition du public un répertoire du patrimoine culturel du Québec, très riche en informations textuelles, mais qu’il est malheureusement difficile d’intégrer aux données des autres acteurs culturels du Québec, ou de lier à toutes les connaissances patrimoniales qui sont déjà présentes dans le réseau de données ouvertes Linked Open Data (LOD).

Plusieurs travaux ont déjà été proposés pour soutenir l’effort d’extraction de connaissances à partir de textes : des annotateurs sémantiques, qui identifient dans un document les entités qui y sont citées (personnes, organisations, etc.) et les lient à leur représentation dans une base de connaissances du LOD; des extracteurs de relations, capables d’extraire du texte des relations entre deux entités (par exemple, « X est l’auteur du roman Y »); des extracteurs d’événements et d’informations temporelles. Dans la très grande majorité des cas, ces outils ont été développés pour l’anglais, ou offrent de piètres performances lorsqu’appliqués au français.

Nous proposons donc d’explorer une approche qui consiste à produire, à partir d’un corpus de documents en français, une version équivalente traduite sur laquelle seront appliqués les outils déjà existants pour l’anglais (le service Syntaxnet de Google, par exemple). Cela implique qu’il faudra tenir compte des erreurs et inexactitudes qui résulteront de l’étape de traduction. Pour y arriver, des techniques de paraphrase et de simplification de texte seront explorées, l’hypothèse ici étant que des phrases simples sont plus faciles à traduire et que cette simplification n’aura pas d’impact majeur sur la résolution de la tâche si la sémantique est préservée lors de cette simplification. On notera aussi que certains aspects de la langue, comme l’anaphore, perturbent la traduction (le module de traduction aura du mal à choisir entre les pronoms « it » et « he » pour traduire le pronom « il »). Il faudra dans ces cas mesurer précisément leur impact et proposer des solutions de contournement.

En bref, le projet proposé permettra de déterminer dans quelle mesure les services de traduction actuellement disponibles préservent suffisamment le sens du texte pour pouvoir exploiter des outils développés pour une autre langue. L’hypothèse que nous désirons valider est que leurs lacunes peuvent être comblées par certains prétraitements du texte original, et que ces prétraitements peuvent être implémentée à faibles coûts (en temps et en ressources).

### Thomas MacDougall

Supervisé.e par : Sébastien Lemieux

Université de Montréal

Use of Deep Learning Approaches in the Activity Prediction and Design of Therapeutic Molecules

The proposed research is to employ Deep Learning and Neural Networks, which are both fields of Machine Learning, to more accurately predict the effectiveness, or “activity”, of potential therapeutic molecules (potential drugs). We are primarily concerned with predicting a given molecule’s ability to inhibit the growth of primary patient cancer cells (cells taken directly from a patient). The Leucegene project at the Institut de Recherche en Immunologie et Cancérologie (IRIC) has tested the activity of a large number of compounds in inhibiting the growth of cancer cells from patients afflicted with acute myeloid leukemia. The proposed research will use this activity data, along with several other data sources, to build an algorithm that can better predict the effectiveness that a molecule will have in inhibiting cancer cell growth. This means that before a molecule is even synthesized in a chemistry lab, a good estimation of its effectiveness as a therapeutic compound can be made, almost instantly. The first approach is to use Neural Networks and “representation learning”, in which features of the molecules that are important to improving activity are identified automatically by the algorithm. This will be done by representing the molecules as graphs and networks. Another approach that will be taken is the use of “multi-task learning” in which the prediction accuracy of an algorithm can be improved if the same algorithm is trained for multiple tasks on multiple datasets. The « multiple tasks » that will be focused on are multiple, but related, drug targets that are essential to cancer cell growth. Moving beyond activity prediction alone, these machine learning architectures will be expanded to design new chemical structures for potential drug molecules, based on information that is learned from drug molecules with known activities. These approaches have the capacity to improve the predictions about whether molecules will make effective drugs, and to design new molecules that have even better effectiveness than known drugs. Research progress in this area will lower the cost, both in money and time, of the drug development process.

### Bhairav Mehta

Supervisé.e par : Liam Paull

Université de Montréal

Attacking the Reality Gap in Robotic Reinforcement Learning

As Reinforcement Learning (RL) becomes an increasingly popular avenue of research, one area that stands to be revolutionized is robotics. However, one prominent downside of applying RL in robotics scenarios is the amount of experience today’s RL algorithms require to learn. Since these data-intensive policies cannot be learned on real robots due to time constraints, researchers turn to fast, approximate simulators. Trading off accuracy for speed can cause problems at test time, and policies that fail to transfer to the real world fall prey to the reality gap: the differences between training simulation and the real-world robot. Our project focuses on theoretically analyzing this issue, and provides practical algorithms to improve safety and robustness when transferring robotic policies out of simulation. We propose algorithms that use expert-collected robot data to learn a simulator, allowing for better modeling of the testing distribution and minimizing the reality gap upon transfer. In addition, we study the transfer problem using analysis tools from dynamical systems and continual learning research, looking for indicators in neural network dynamics and optimization that signal when the reality gap is likely to pose an issue. Lastly, we use the analysis to synthesize an algorithm which optimizes for the metrics that signal good, “transferable” policies, allowing safer and more robust sim-to-real transfer.

### Timothy Nest

Supervisé.e par : Karim Jerbi

Université de Montréal

Leveraging Machine Learning and Magnetoencephalography for the Study of Normal and atypical states of Consciousness

Understanding the neural processes and network dynamics underlying conscious perception is a complex yet important challenge that lies at the intersection between cognitive brain imaging, mental health, and data science. Magnetoencephalography (MEG) is a brain imaging technique that has many qualities favorable to investigating conscious perception due to its high temporal resolution and high signal to noise ratio. However MEG analyses across space, time and frequency is challenging due to the extreme high-dimensionality of variables of interest, and susceptibility to overfitting. Furthermore, high-computational complexity limits the ease with which investigators might approach some cross-frequency coupling metrics believed to be important for conscious perception and integration, across the whole brain. To mitigate such challenges, researchers frequently rely on a variety of multivariate feature extraction and compression algorithms. However, these techniques still require substantial tuning, and are limited in their application to the kinds of high-order tensor structures encountered in MEG. New methods for the study of conscious perception with MEG are thus needed.

In this project, we will leverage very recent advances in computer science and machine learning that extend algorithms currently used in neuroimaging research, to extreme high-dimensional spaces. Taken together, the proposed research will apply state-of-the-art techniques in machine-learning and electrophysiological signal processing to overcome current obstacles in the study of the brain processes that mediate conscious perception. This work will constitute an important contribution to neuroimaging methodology, neuropharmacology, and psychiatry. Beyond expanding our understanding of healthy cognition, this research may ultimately provide novel paths to the study of psychiatric disorders that involve altered conscious perception, such as Schizophrenia.

### Jacinthe Pilette

Supervisé.e par : Jean-François Arguin

Université de Montréal

Recherche de nouvelle physique au Grand collisionneur de hadrons (LHC) à l’aide de l’apprentissage profond

Le Grand collisionneur de hadrons (LHC) se situe au cœur de la recherche fondamentale en physique. Avec sa circonférence de 27 km, celui-ci constitue le plus grand et plus puissant accélérateur de particules au monde. Ceci en fait le meilleur outil afin d’étudier le domaine de l’infiniment petit. C’est d’ailleurs au LHC que le boson de Higgs fut découvert, menant à l’obtention du prix Nobel de physique en 2013.

Cependant, le modèle standard, référence qui dicte les lois régissant les particules et leurs interactions, possède plusieurs lacunes que les physiciens et physiciennes n’ont toujours pas réussi à combler. Plusieurs théories furent élaborées, mais aucune d’entre elles ne fut observée au LHC. Face à ce défi, la communauté de physique des particules devra utiliser une nouvelle approche.

Le groupe ATLAS de l’Université de Montréal s’est ainsi tourné vers l’intelligence artificielle. Le projet élaboré par cette collaboration, et l’objectif principal de cette recherche est de développer un algorithme d’apprentissage profond qui permettrait de détecter des anomalies dans les données. L’algorithme développé sera ensuite utilisé sur les données du détecteur ATLAS dans l’espoir de découvrir des signaux de nouvelle physique et d’améliorer notre compréhension de l’univers.

### Léa Ricard

Supervisé.e par : Emma Frejinger

Université de Montréal

Modélisation de la probabilité d’acceptation d’une route dans un contexte de covoiturage

Le covoiturage touche aux algorithmes fréquemment étudiés de tournées de véhicule, de ramassage et de livraison avec fenêtres de temps et de transport à la demande dynamique. Toutefois, très peu d’études s’attardent au contexte où les conducteurs et les passagers peuvent rejeter une proposition de route. Alors que le rejet d’une route proposée est rare lorsque les conducteurs sont des professionnels, c’est plutôt la norme dans un contexte de covoiturage. La modélisation de la probabilité d’acceptation d’une route se pose alors comme un problème central dans le développement d’une application mobile de covoiturage de qualité.

Le modèle d’apprentissage automatique développé devra estimer, selon les caractéristiques de l’utilisateur (notamment s’il est conducteur ou passager) et les routes alternatives proposées, la probabilité d’acceptation d’une route. De prime abord, cette modélisation pose deux défis :

(1) La façon dont les acceptations et les refus sont collectés pose un problème de type logged bandit. À ce titre, plusieurs propositions peuvent être offertes en même temps et un utilisateur peut en accepter plusieurs. De plus, les offres peuvent être activement refusées, simplement ignorées ou acceptées. Puisque les offres sont affichées séquentiellement, celles qui apparaissent en premier ont plus de chance d’attirer l’attention de l’utilisateur. L’ordre des propositions a donc probablement une influence sur la probabilité d’acceptation.
(2) Le comportement des nouveaux utilisateurs, pour qui très peu d’information est disponible, devra être inféré à partir des clients similaires de longue date. Ceci est en soi un problème difficile.

### Alexandre Riviello

Supervisé.e par : Jean-Pierre David

Polytechnique Montréal

Hardware Acceleration of Speech Recognition Algorithms

Speech recognition has become prevalent in our lives in recent years. Personal assistants, such as Amazon’s Alexa or Apple’s Siri are such examples. With the rise of deep learning, speech recognition algorithms gained a lot of precision. This is due, mostly, to the use of neural networks. These complex algorithms, used in the context of a classification task, can distinguish between different characters, phonemes or words. However, they require lots of computations, limiting their use in power-constrained devices, such as smartphones. In my research, I will attempt to find hardware-friendly implementations of these networks. Deep learning algorithms are usually written in high-level languages using frameworks such as Torch or Tensorflow. To generate hardware-friendly representations, models will be adapted, using these frameworks. For example, recent findings have shown that basic networks can use weights and activations represented over 1 or 2 bits and retain their accuracy. The reduction of the precision of the network parameters is called quantization. This concept will be one of the many ways used to simplify the networks. Another aspect of this research will be to revisit methods of representing voice features. Traditionally, spoken utterances were converted to Mel Frequency Cepstrum Coefficients (MFCCs) which are essentially values representing signal power over a frequency axis. These coefficients are calculated roughly every 10 ms and are then sent to the model network. A representation of lower precision can greatly reduce the computational costs of the network. The overall goal of the research will be to improve the calculation speed and to diminish the power consumption of speech recognition algorithms.

### Lluis E. Castrejon Subira

Supervisé.e par : Aaron Courville

Université de Montréal

Self-Supervised Learning of Visual Representations from Videos.

### Francis Banville

Supervisé.e par : Timothée Poisot

Université de Montréal

Réseaux d’interactions écologiques et changements climatiques : inférence et modélisation par des techniques d’apprentissage automatique

### Avishek Bose

Supervisé.e par : William Hamilton

McGill University

Domain Agnostic Adversarial Attacks for Security and Privacy.

### Elodie Deschaintres

Supervisé.e par : Catherine Morency

Polytechnique Montréal

Modélisation des interactions entre les modes de transport par l’intégration de différentes sources de données

### Laura Gagliano

Polytechnique Montréal

Artificial Neural Networks and Bispectrum for Epileptic Seizure Prediction.

### Ellen Jackson

Supervisé.e par : Hélène Carabin

Université de Montréal

Evaluation of a Directed Acyclic Graph for Cysticercosis using Multiple Methods.

### Mengying Lei

Supervisé.e par : Lijun Sun

McGill University

Spatial-Temporal Traffic Pattern Analysis and Urban Computation Applications based on Tensor Decomposition and Multi-scale Neural Networks.

### Tegan Maharaj

Supervisé.e par : Christopher Pal

Polytechnique Montréal

Deep ecology: Bringing together theoretical ecology and deep learning.

### Antoine Prouvost

Supervisé.e par : Andrea Lodi

Polytechnique Montréal

Learning to Select Cutting Planes in Integer Programming.

### Matthew Schlegel

Supervisé.e par : Martha White

University of Alberta

Representing the World Through Predictions in Intelligent Machines.

### Jing Xu

Supervisé.e par : Guillaume-Alexandre Bilodeau

Polytechnique Montréal

Computer Vision for Safe Interactions between Humans and Intelligent Robots.

### Olivia Gélinas

Polytechnique Montréal

Stage chez Le Devoir, spécialité science des données.

### Justine Pépin

Polytechnique Montréal

Stage chez Le Devoir, spécialité science des données.

### Sandrine Vieira

UQAM

Stage chez Le Devoir, spécialité communication.

### Selçuk Güven

Supervisé.e par : Philippe Langlais

Université de Montréal

Dépistage et diagnostic des troubles de la parole et du langage chez les enfants utilisant l’IA

Entreprise : LinguAI

Le but de ce projet est de créer une plateforme Web où les cliniciens seront guidés dans la différenciation des troubles de la parole et du langage chez les enfants en posant simplement quelques questions de fond et en analysant les échantillons de parole des enfants pour y déceler les erreurs de parole et de langage. La solution proposée utilisera une reconnaissance de la parole qui est suffisamment précise dans les  » troubles de la parole  » qui sera développée au cours de ce projet et ensuite un algorithme sera déployé pour l’analyse détaillée des erreurs. Une partie de ce projet était le projet postdoctoral du boursier. Les objectifs à long terme de ce projet sont de rendre cet outil accessible aux parents de ces enfants également.

Supervisé.e par : Sébastien Lemieux

Université de Montréal

Développement d’outils basés sur l’apprentissage automatique pour soutenir la modulation thérapeutique de l’épissage alternatif

Entreprise : BioBenchAI

Les gènes humains sont constitués de séquences codantes pour des protéines (les exons) interrompues par des séquences non-codantes (les introns). Les séquences codantes sont jointes selon différentes combinaisons lors d’une étape importante de l’épissage alternatif (EA). Lors de maladies, l’EA est souvent dérégulé mais l’impact fonctionnel d’une telle dérégulation et les stratégies pour la corriger restent largement inconnus en raison de sa complexité et du manque d’outil informatique pour analyser de grands ensembles de données d’EA. Nous proposons de développer un ensemble d’outils informatiques utilisant des algorithmes d’apprentissage automatique qui nous permettront de comprendre l’impact d’une dérégulation de l’EA dans la pathogenèse et ainsi d’aider aux développements de nouvelles stratégies thérapeutiques ciblant les mécanismes de l’EA.

### Jhelum Chakravorty

Supervisé.e par : Doina Precup

McGill University

Temporal abstraction in multi-agent environment

Temporal abstraction refers to the ability of an intelligent agent to reason, act and plan at multiple time scales. The question of how to obtain and reason with temporally abstract representations has been extensively studied in classical planning and control theory, and more recently it has become an important topic in reinforcement learning, especially through the framework of options. The theoretical development of options is based on the framework of Semi-Markov Decision Processes (SMDPs), in which an agent interacts with its environment by observing states and taking actions. As a result of an action, the agent receives an immediate reward, and transitions to a new state drawn from some distribution, after a certain period of time which is also drawn stochastically. Both the state and the dwell-time distribution are dependent only on the agent’s state and action. However, in many cases of practical importance, an agent may be faced either with more general environments, in which the environment may be partially observable, or there may be multiple agents acting in the environment. For example, in energy markets or in transportation there may be many agents, who would be interacting with each other and making decisions without being able to observe relevant information except at specific time points.We propose to focus on establishing a mathematical framework for temporal abstraction which would work in Decentralized Partially Observable Markov Decision Processes. In a multi-agent system, agents take decision and exchange their information at designated decision epochs. In general, the decision epochs are given by the realizations of a random sequence. Instead of looking at every instant of time, when an action is taken by an agent, we are interested in the Decentralized Semi-Markov Decision Processes (Dec-SMDPs), in which a Partially Observable Markov Decision Process (POMDP) corresponding to an agent is embedded between any two successive decision epochs. In between two such decision epochs, each agent chooses actions so as to maximize the total return over a finite or infinite horizon, i.e., it solves a POMDP problem. The optimal decision epochs are chosen based on a given criterion, e.g., exchanging information at some goal states fixed a priori or when the increase of reward from the last decision epoch is less than a threshold. The overall performance, which is to be maximized through such sequential decision making consists of two rewards. The exchange of information in encouraged by an extrinsic reward along with an intrinsic reward that is maximized in between two consecutive decision epochs.We would like to investigate two aspects of this problem setup. First, we are interested in formally establishing the framework for Partially Observable Semi-Markov Decision Processes and its extension to decentralized (multi-agent) problems. We would like to investigate if under certain simplifying assumptions in the planning problem, the posterior beliefs (i.e., belief on the state of the environment based on past information and current action) exhibit certain monotonicity and symmetry properties so that we can infer what the structure of optimal policies could be. We also want to establish the general Dec-SMDP framework for modeling this problem and characterize its properties in comparison with SMDPs.In the subsequent analysis, we would like to investigate learning algorithms for these families of problems. We will build on standard reinforcement learning algorithms for temporal abstraction, such as option-critic, and provide extensions in our case that are consistent with the theoretical characterization of these problems. We will also examine the performance of both value-function-based and policy-gradient style algorithms in this context. We will compare the results that can be obtained using our framework to results in which each agent ignores the others and only tries to optimize myopically its own reward. We will use both standard simulated small problems from the multi-agent literature, designed to emphasize specific aspects, as well as larger scale domains that correspond to simulate transportation and energy markets, where multiple agents work in a cooperative setting to achieve a common task in a decentralized manner, e.g., self-driving cars and smart-grids. In such applications the agents occasionally communicate among themselves and use a common information to update a belief about the state of the world and a local information to decide about their individual policies and terminations of such policies.

### Eugene Belilovsky

Supervisé.e par : Aaron Courville

Université de Montréal

Towards Learning Language Based Navigation in Visually Rich 3-D Environments

A long term goal of artificial intelligence and robotics is a robot able to perform manual tasks by understanding language instructions or questions and using visual and other sensory input to navigate and interact in a complex environment to achieve it’s goals. Advances in machine learning have succeeded in important perceptual sub-tasks of this problem: object recognition, speech recognition, natural language processing among others. However, how to integrate these successes with sequential decision making and multi-modal reasoning across language, vision, and other modalities is an open question that has been difficult to study. Very recently visually rich 3-D simulations and tasks have arisen aimed to allow the development of algorithms for learning language directed navigation of robots. Even in these constrained simulations, straightforward application of existing machine learning and reinforcement learning techniques are unable to effectively tackle this new set of challenges. We aim to develop methods for these problems focusing on visual relational reasoning and ideas from human learning. We also strive to advance the nascent evaluation methodology of these algorithms. Besides making steps towards our ambitions of creating intelligent agents, methods developed to solve these tasks can be directly applied in household automation, robotic assistants, manufacturing, and autonomous driving.

### Glen Berseth

Supervisé.e par : Christopher Pal

Polytechnique Montréal

Visual Imitation Learning With Partial Information

For many control and decision-making tasks, it is complex to describe the desired behaviour we hope to elicit from a robot. Many complex tasks that we want robots to be able to perform are dependant on a skill that people acquire at a young age, imitation. The ability of animals to learn from demonstrations has triggered research across many disciplines. This work will push the possibilities of imitation learning by creating methods that will allow robots to learn from diverse video demonstration. Of particular interest are skills that involve interaction with objects in the real world. Imitation learning is a tough problem but is also a very important one. If we make enough progress on imitation learning average people could program robots by providing a few demonstrations of the desired task in the real world.

### Ricardo de Azambuja

Supervisé.e par : Giovanni Beltrame

Polytechnique Montréal

High Fidelity Data Collection for Precision Agriculture with Drone Swarms

### Benoit Delcroix

Supervisé.e par : Michel Bernier

Polytechnique Montréal

Un défi majeur dans le secteur du bâtiment est l’’absence de systèmes continus de suivi de la performance et d’évaluation des écarts de performance entre la situation observée et celle désirée. L’opération non-optimale des systèmes de Chauffage, Ventilation et Conditionnement d’Air (CVCA) entraîne des pertes énergétiques et de confort des occupants. Le secteur du bâtiment représente environ un tiers de la consommation énergétique au Québec et au Canada. Ainsi, des mesures d’’efficacité dans ce secteur produisent des impacts positifs majeurs. L’idée de ce projet est d’utiliser des méthodes d’apprentissage profond pour exploiter les larges jeux de données générés par les systèmes CVCA. Le but final est d’automatiser la détection et le diagnostic des anomalies, et d’optimiser l’opération des équipements. Les bénéfices incluent une gestion améliorée de l’énergie et une meilleure prise en compte du confort des occupants. Au terme de ce projet, des outils de détection / diagnostic / contrôle basés sur l’apprentissage profond seront développés et testés à des fins d’implantation dans des bâtiments réels.

Supervisé.e par : Michel Gendreau

Polytechnique Montréal

The increasing use of algorithmic decision-making in domains that affect people’s lives, has raised concerns about possible biases and discrimination that such systems might introduce. Recent concerns on algorithmic discrimination have motivated the development of fairness-aware mechanisms in the machine learning (ML) community and the operations research (OR) community, independently. While in fairness-aware ML, the focus is usually on ensuring that the inference and predictions produced by a learned model are fair, the OR community has developed methods to ensure fairness in solutions of an optimization problem. In this project, I plan to build on the complementary strengths of fairness methods in ML and OR to address these shortcomings in a fair data-driven decision-making system. I will apply this work to real-world problems in the areas of personalized education, employment hiring (business), social well-being (health), and network design (transportation). The advantage of my proposed system compared to the existing works is that it: 1) incorporates domain knowledge with data-driven probabilistic models, 2) detects and describes complex discriminative patterns, 3) returns a fair decision/policy, and 4) breaks negative/positive feedback loops.

### Kuldeep Kumar

Supervisé.e par : Michel Gendreau

Polytechnique Montréal

The increasing use of algorithmic decision-making in domains that affect people’s lives, has raised concerns about possible biases and discrimination that such systems might introduce. Recent concerns on algorithmic discrimination have motivated the development of fairness-aware mechanisms in the machine learning (ML) community and the operations research (OR) community, independently. While in fairness-aware ML, the focus is usually on ensuring that the inference and predictions produced by a learned model are fair, the OR community has developed methods to ensure fairness in solutions of an optimization problem. In this project, I plan to build on the complementary strengths of fairness methods in ML and OR to address these shortcomings in a fair data-driven decision-making system. I will apply this work to real-world problems in the areas of personalized education, employment hiring (business), social well-being (health), and network design (transportation). The advantage of my proposed system compared to the existing works is that it: 1) incorporates domain knowledge with data-driven probabilistic models, 2) detects and describes complex discriminative patterns, 3) returns a fair decision/policy, and 4) breaks negative/positive feedback loops.

### Elizaveta Kuznetsova

Supervisé.e par : Miguel Anjos

Polytechnique Montréal

The cumulative solar and wind power capacity integrated mainly into low and medium voltage grids in Canada represents 9% of total available power capacity in 2015, and is expected to more than double by 2040. This reality will create not only opportunities for sustainable energy production, but also challenges for the system operator due to the uncertain power fluctuations from supporting multiple prosumers (customers who can alternatively behave as energy consumers or producers). This project addresses the question of how to involve prosumers in the energy management process for provision of ancillary services in the grid (e.g. voltage control) while mitigating unsuitable emerging effects. The idea is to consider a three-layer optimization problem related to different voltage levels (high, medium and low). Grid incentives will be optimized at the high voltage level, while lower levels will optimize the dispatch among grid prosumers to maximize their involvement. An Agent-Based Modelling framework will provide a backbone for this multi-level optimization, enable bi-directional information flows, and make it possible to handle the challenges of high data volume and complexity.

### Tarek Lajnef

Supervisé.e par : Miguel Anjos

Polytechnique Montréal

The cumulative solar and wind power capacity integrated mainly into low and medium voltage grids in Canada represents 9% of total available power capacity in 2015, and is expected to more than double by 2040. This reality will create not only opportunities for sustainable energy production, but also challenges for the system operator due to the uncertain power fluctuations from supporting multiple prosumers (customers who can alternatively behave as energy consumers or producers). This project addresses the question of how to involve prosumers in the energy management process for provision of ancillary services in the grid (e.g. voltage control) while mitigating unsuitable emerging effects. The idea is to consider a three-layer optimization problem related to different voltage levels (high, medium and low). Grid incentives will be optimized at the high voltage level, while lower levels will optimize the dispatch among grid prosumers to maximize their involvement. An Agent-Based Modelling framework will provide a backbone for this multi-level optimization, enable bi-directional information flows, and make it possible to handle the challenges of high data volume and complexity.

### Neda Navidi

Supervisé.e par : Nicolas Saunier

Polytechnique Montréal

Learning driving behavior from smartphone location and motion sensors Monitoring and tracking vehicles and driving behavior are of great interest to better assess safety and understand the relationship with potential factors related to the infrastructure, vehicles and users. This has been implemented in recent years by car insurance to better assess their customers’ risk of crash and offer usage-based premium. Driver monitoring and analysis or driver behavior profiling is the process of automatically collecting driving data (e.g., location, speed, acceleration) and predicting the crash risk. These systems are mainly based on Global Positioning System (GPS), which suffers from accuracy issues, e.g. in urban canyons, and is insufficient to detect normal and risky driving events like steering and braking to assess the driving behaviours. To address this problem, researchers have proposed the integration of GPS, Inertial Navigation System (INS) and motion sensors, and map-matching (MM) in a single hybrid system. INS is fused with GPS and used during signal outages to provide continuous positioning (dead reckoning). Map matching is the process of estimating a user’s position on a road segment, which provides more contextual information like road geometry and conditions, historical risk of the segment and other drivers’ behaviour. The objective of this work is to improve the understanding of driver behaviour and crash risk by integrating location and motion data, driving events and road attributes using different machine learning algorithms.The objective of this work is to improve the understanding of driver behaviour and crash risk by integrating location and motion data, driving events and road attributes. The specific objectives are the following: 1) to detect risky driving events, namely hard acceleration/braking, compliance to signalization (e.g. speed limits), sharp steering, tailgating, improper passing and weaving from location and motion data using machine learning (ML); 2) to apply map-matching algorithms to extract road-related attributes; 3) to cluster driver behaviour based on the time series of location and motion data, detected driving events and road-related attributes.

### Nurit Oliker

Supervisé.e par : Bernard Gendron

Université de Montréal

We study the context of a transportation network manager who wants to take decisions on infrastructures, assets and resources to deploy in order to achieve its objectives. The network manager has to take into account that there are several classes of users, most of which pursue their own objectives within the rules stated by the manager, while others have objectives that are antagonist to those of the manager. Our goal is to develop methodology to help the transportation network manager. The application that motivates this research project is based on the transportation network design problem faced by a vehicle inspection agency who wants to inspect a maximum number of vehicles on a given territory under a limited budget. In such application, it is important to take into account the fact that some users will react to the installation of new vehicle inspection stations by diverting from their usual path to avoid inspection. Other applications of interest include the design of transportation networks that are resilient to major accidents and terrorist attacks. In this context, the network manager must anticipate potential threats posed by hostile users.

### Camilo Ortiz Astorquiza

Supervisé.e par : Emma Frejinger

Université de Montréal

The railway industry represents one of the most important means of freight transportation. In Canada only more than 900,000 tons of goods are moved every day. where one of the major companies of the sector is Canadian National Railways (CN). An important component in their overall structure is the locomotive fleet management. The high cost of each locomotive and the large number of them required to satisfy train schedules makes the locomotive planning highly valuable. This in turn, represents an environmental and macroeconomic effect of great importance. Although several variants of locomotive planning problems have been studied before there is still a huge gap between the state-of-practice and the state-of-the-art. Thus, we will first study an optimization model that is tailored for CN’s requirements. Moreover, we will investigate on the development of specialized solution methods that incorporate machine learning with operations research techniques to obtain optimal solutions within reasonable time. This will provide a tool for the partner company to better evaluate scenarios in the locomotive planning and give value to the data while representing an important scientific contribution for the optimization community.

### Musa Ozboyaci

Supervisé.e par : Sebastian Pechmann

Université de Montréal

Protein homeostasis describes the cells capability to keep its proteins in their correct shape and function through a complex regulatory system that integrates protein synthesis, folding and degradation. How cells maintain protein homeostasis is a fundamental phenomenon, an understanding of which has direct implications for prevention and treatment of severe human diseases such as Alzheimer’s and Parkinson’s. The protein quality control is regulated through specific enzymes called molecular chaperones that assist the (re)folding of proteins thus managing a complex and varied proteome efficiently. Although the specificity of interactions of these chaperones with their client proteins is known to be the key to the efficient allocation of protein quality control capacity, a significant yet unanswered question lies in rationalizing the principles of this specificity. This project aims to systematically define the principles of sequence specificity across eukaryotic chaperone network through a combination of molecular modelling and machine learning methods. To this end, the peptide sequences that confer chaperone specificity will be identified systematically using a robust docking procedure accelerated by a Random Forest model. To account for the conditional interdependencies of the energetic contributions of the peptide residues binding to the chaperone receptor and to capture them, probabilistic graphical models will be developed and deep learning methods will be applied to the large dataset obtained from docking simulations. This project, through the unique and rich dataset we will construct and the sophisticated analyses we will apply, will not only unravel the sequence specificity in protein homeostasis interactions during health and disease, but also provide the necessary guidelines for how it can be re-engineered for rational therapeutic intervention.

### Maximilian Puelma Touzel

Supervisé.e par : Guillaume Lajoie

Université de Montréal

Recurrent neural nets are neuroscience-inspired AI algorithms that are revolutionizing the machine learning of complex sequences. They help power a variety of widely used applications such as Google Translate and Apple’’s Siri. But they are also big, complicated models, and learning them is a delicate process, up to now requiring much fine-tuning to avoid the parameter adjustments getting out of control. The human brain also faces this stability problem when it learns sequences, but it has a robust, working solution that we are only beginning to understand. Bringing together experts in neuroscience, applied math, and artificial intelligence, we will adapt sophisticated methods for measuring stability from the mathematics of dynamical systems. We will develop learning algorithms that use this information to efficiently guide the learning, and will employ them in a neuroscience study that compares artificial and brain solutions to learning complex task sequences. Our goal is to unify and extend our understanding of how natural and artificial recurrent neural nets learn complex sequences.

### Raphael Harry Frederico Ribeiro Kramer

Supervisé.e par : Guy Desaulniers

Polytechnique Montréal

Facility location arises as an important field in combinatorial optimization with applications to logistics and data mining. In facility location problems (FLPs), one seeks to find the location of some supply points and to assign customers to those supply points so as to optimize a certain measure of performance. In data mining, several FLPs can be used with the purpose of modelling and solving clustering problems. The p-center problem (PCP) is an example of such type of problem, in which one seeks to find the location of p points (namely the centers) so as to minimize the maximum dissimilarity between any customer and its closest center. This problem is extremely difficult in practice. In a recent article co-authored by the candidate, the most classical variant of the PCP (namely the vertex PCP) is solved by an iterative algorithm for problems containing up to a million data points within reasonable time limits. This is more than 200x larger than previous algorithms. In this project we aim at extending some of the ideas used in that article to solve other classes of facility location problems for large datasets.

### Joshua Stipancic

Supervisé.e par : Aurélie Labbe

HEC Montréal

Road traffic crashes are a serious concern. Typically, dangerous locations in the road network are identified based on historical crash data. However, using crashes is not ideal, as crash data bases contain error and omissions and crashes are not perfect predictors of safety. Our earlier work demonstrates how mobile sensor data, such as GPS travel data collected from regular drivers, can be used to substitute crash data in the safety management process within Quebec City. However, advanced statistical models must be developed to convert the collected sensor data into predicted crash counts at sites throughout the network. This project proposes three advancements to crash models developed in previous work. First, methods for imputing missing data will be proposed and explored. The effect of these methods on the final predicted crash counts will also be quantified. Second, techniques for expanding analysis to an entire road network will be developed. Third, the developed models will be tested on additional datasets in Montreal and Toronto. The ability to predict levels of safety with mobile sensor data is a substantial contribution to the field of transportation.