Portrait de Yoshua Bengio

Yoshua Bengio

Membre académique principal
Chaire en IA Canada-CIFAR
Professeur titulaire, Université de Montréal, Département d'informatique et de recherche opérationnelle
Fondateur et Conseiller scientifique, Équipe de direction
Sujets de recherche
Apprentissage automatique médical
Apprentissage de représentations
Apprentissage par renforcement
Apprentissage profond
Causalité
Modèles génératifs
Modèles probabilistes
Modélisation moléculaire
Neurosciences computationnelles
Raisonnement
Réseaux de neurones en graphes
Réseaux de neurones récurrents
Théorie de l'apprentissage automatique
Traitement du langage naturel

Biographie

*Pour toute demande média, veuillez écrire à medias@mila.quebec.

Pour plus d’information, contactez Marie-Josée Beauchamp, adjointe administrative à marie-josee.beauchamp@mila.quebec.

Reconnu comme une sommité mondiale en intelligence artificielle, Yoshua Bengio s’est surtout distingué par son rôle de pionnier en apprentissage profond, ce qui lui a valu le prix A. M. Turing 2018, le « prix Nobel de l’informatique », avec Geoffrey Hinton et Yann LeCun. Il est professeur titulaire à l’Université de Montréal, fondateur et conseiller scientifique de Mila – Institut québécois d’intelligence artificielle, et codirige en tant que senior fellow le programme Apprentissage automatique, apprentissage biologique de l'Institut canadien de recherches avancées (CIFAR). Il occupe également la fonction de conseiller spécial et directeur scientifique fondateur d’IVADO.

En 2018, il a été l’informaticien qui a recueilli le plus grand nombre de nouvelles citations au monde. En 2019, il s’est vu décerner le prestigieux prix Killam. Depuis 2022, il détient le plus grand facteur d’impact (h-index) en informatique à l’échelle mondiale. Il est fellow de la Royal Society de Londres et de la Société royale du Canada, et officier de l’Ordre du Canada.

Soucieux des répercussions sociales de l’IA et de l’objectif que l’IA bénéficie à tous, il a contribué activement à la Déclaration de Montréal pour un développement responsable de l’intelligence artificielle.

Étudiants actuels

Collaborateur·rice alumni - McGill
Collaborateur·rice alumni - UdeM
Collaborateur·rice de recherche - Cambridge University
Superviseur⋅e principal⋅e :
Doctorat - UdeM
Visiteur de recherche indépendant
Co-superviseur⋅e :
Doctorat - UdeM
Visiteur de recherche indépendant
Superviseur⋅e principal⋅e :
Collaborateur·rice de recherche - N/A
Superviseur⋅e principal⋅e :
Doctorat - UdeM
Collaborateur·rice de recherche - KAIST
Collaborateur·rice alumni - UdeM
Collaborateur·rice alumni - UdeM
Co-superviseur⋅e :
Doctorat - UdeM
Co-superviseur⋅e :
Doctorat - UdeM
Doctorat - UdeM
Stagiaire de recherche - UdeM
Doctorat - UdeM
Doctorat - UdeM
Superviseur⋅e principal⋅e :
Collaborateur·rice alumni - UdeM
Postdoctorat - UdeM
Superviseur⋅e principal⋅e :
Collaborateur·rice de recherche - UdeM
Collaborateur·rice alumni - UdeM
Postdoctorat - UdeM
Superviseur⋅e principal⋅e :
Collaborateur·rice alumni
Collaborateur·rice alumni - UdeM
Superviseur⋅e principal⋅e :
Doctorat - UdeM
Collaborateur·rice alumni - UdeM
Doctorat - UdeM
Co-superviseur⋅e :
Collaborateur·rice de recherche - UdeM
Doctorat - UdeM
Superviseur⋅e principal⋅e :
Doctorat - UdeM
Superviseur⋅e principal⋅e :
Postdoctorat - UdeM
Superviseur⋅e principal⋅e :
Visiteur de recherche indépendant - UdeM
Doctorat - UdeM
Superviseur⋅e principal⋅e :
Collaborateur·rice de recherche - Ying Wu Coll of Computing
Collaborateur·rice de recherche - University of Waterloo
Superviseur⋅e principal⋅e :
Collaborateur·rice alumni - Max-Planck-Institute for Intelligent Systems
Collaborateur·rice de recherche - UdeM
Co-superviseur⋅e :
Doctorat - UdeM
Postdoctorat - UdeM
Visiteur de recherche indépendant - UdeM
Doctorat - UdeM
Superviseur⋅e principal⋅e :
Visiteur de recherche indépendant
Superviseur⋅e principal⋅e :
Collaborateur·rice alumni - UdeM
Collaborateur·rice alumni - UdeM
Postdoctorat
Co-superviseur⋅e :
Visiteur de recherche indépendant - Technical University of Munich
Doctorat - UdeM
Co-superviseur⋅e :
Visiteur de recherche indépendant
Superviseur⋅e principal⋅e :
Collaborateur·rice alumni - UdeM
Postdoctorat - UdeM
Co-superviseur⋅e :
Doctorat - UdeM
Superviseur⋅e principal⋅e :
Collaborateur·rice de recherche
Collaborateur·rice de recherche - UdeM
Doctorat - McGill
Superviseur⋅e principal⋅e :
Doctorat - UdeM
Superviseur⋅e principal⋅e :
Collaborateur·rice alumni - McGill
Superviseur⋅e principal⋅e :

Publications

Catalyst GFlowNet for electrocatalyst design: A hydrogen evolution reaction case study
Efficient and inexpensive energy storage is essential for accelerating the adoption of renewable energy and ensuring a stable supply, despit… (voir plus)e fluctuations in sources such as wind and solar. Electrocatalysts play a key role in hydrogen energy storage (HES), allowing the energy to be stored as hydrogen. However, the development of affordable and high-performance catalysts for this process remains a significant challenge. We introduce Catalyst GFlowNet, a generative model that leverages machine learning-based predictors of formation and adsorption energy to design crystal surfaces that act as efficient catalysts. We demonstrate the performance of the model through a proof-of-concept application to the hydrogen evolution reaction, a key reaction in HES, for which we successfully identified platinum as the most efficient known catalyst. In future work, we aim to extend this approach to the oxygen evolution reaction, where current optimal catalysts are expensive metal oxides, and open the search space to discover new materials. This generative modeling framework offers a promising pathway for accelerating the search for novel and efficient catalysts.
Recursive Self-Aggregation Unlocks Deep Thinking in Large Language Models
Test-time scaling methods improve the capabilities of large language models (LLMs) by increasing the amount of compute used during inference… (voir plus) to make a prediction. Inference-time compute can be scaled in parallel by choosing among multiple independent solutions or sequentially through self-refinement. We propose Recursive Self-Aggregation (RSA), a test-time scaling method inspired by evolutionary methods that combines the benefits of both parallel and sequential scaling. Each step of RSA refines a population of candidate reasoning chains through aggregation of subsets to yield a population of improved solutions, which are then used as the candidate pool for the next iteration. RSA exploits the rich information embedded in the reasoning chains -- not just the final answers -- and enables bootstrapping from partially correct intermediate steps within different chains of thought. Empirically, RSA delivers substantial performance gains with increasing compute budgets across diverse tasks, model families and sizes. Notably, RSA enables Qwen3-4B-Instruct-2507 to achieve competitive performance with larger reasoning models, including DeepSeek-R1 and o3-mini (high), while outperforming purely parallel and sequential scaling strategies across AIME-25, HMMT-25, Reasoning Gym, LiveCodeBench-v6, and SuperGPQA. We further demonstrate that training the model to combine solutions via a novel aggregation-aware reinforcement learning approach yields significant performance gains. Code available at https://github.com/HyperPotatoNeo/RSA.
Active Attacks: Red-teaming LLMs via Adaptive Environments
We address the challenge of generating diverse attack prompts for large language models (LLMs) that elicit harmful behaviors (e.g., insults,… (voir plus) sexual content) and are used for safety fine-tuning. Rather than relying on manual prompt engineering, attacker LLMs can be trained with reinforcement learning (RL) to automatically generate such prompts using only a toxicity classifier as a reward. However, capturing a wide range of harmful behaviors is a significant challenge that requires explicit diversity objectives. Existing diversity-seeking RL methods often collapse to limited modes: once high-reward prompts are found, exploration of new regions is discouraged. Inspired by the active learning paradigm that encourages adaptive exploration, we introduce \textit{Active Attacks}, a novel RL-based red-teaming algorithm that adapts its attacks as the victim evolves. By periodically safety fine-tuning the victim LLM with collected attack prompts, rewards in exploited regions diminish, which forces the attacker to seek unexplored vulnerabilities. This process naturally induces an easy-to-hard exploration curriculum, where the attacker progresses beyond easy modes toward increasingly difficult ones. As a result, Active Attacks uncovers a wide range of local attack modes step by step, and their combination achieves wide coverage of the multi-mode distribution. Active Attacks, a simple plug-and-play module that seamlessly integrates into existing RL objectives, unexpectedly outperformed prior RL-based methods -- including GFlowNets, PPO, and REINFORCE -- by improving cross-attack success rates against GFlowNets, the previous state-of-the-art, from 0.07% to 31.28% (a relative gain greater than
Graph Dreamer: Temporal Graph World Models for Sample-Efficient and Generalisable Reinforcement Learning
Low Compute Unlearning via Sparse Representations
Frederik Träuble
Ashish Malik
Michael Curtis Mozer
Sanjeev Arora
Anirudh Goyal
Machine unlearning, which involves erasing knowledge about a \emph{forget set} from a trained model, can prove to be costly and infeasible … (voir plus)using existing techniques. We propose a low-compute unlearning technique based on a discrete representational bottleneck. We show that the proposed technique efficiently unlearns the forget set and incurs negligible damage to the model's performance on the rest of the dataset. We evaluate the proposed technique on the problem of class unlearning using four datasets: CIFAR-10, CIFAR-100, LACUNA-100 and ImageNet-1k. We compare the proposed technique to SCRUB, a state-of-the-art approach which uses knowledge distillation for unlearning. Across all four datasets, the proposed technique performs as well as, if not better than SCRUB while incurring almost no computational cost.
Relative Trajectory Balance is equivalent to Trust-PCL
Torsional-GFN: a conditional conformation generator for small molecules
Lena Nehale Ezzine
Piotr Gai'nski
Prudencio Tossou
Generating stable molecular conformations is crucial in several drug discovery applications, such as estimating the binding affinity of a mo… (voir plus)lecule to a target. Recently, generative machine learning methods have emerged as a promising, more efficient method than molecular dynamics for sampling of conformations from the Boltzmann distribution. In this paper, we introduce Torsional-GFN, a conditional GFlowNet specifically designed to sample conformations of molecules proportionally to their Boltzmann distribution, using only a reward function as training signal. Conditioned on a molecular graph and its local structure (bond lengths and angles), Torsional-GFN samples rotations of its torsion angles. Our results demonstrate that Torsional-GFN is able to sample conformations approximately proportional to the Boltzmann distribution for multiple molecules with a single model, and allows for zero-shot generalization to unseen bond lengths and angles coming from the MD simulations for such molecules. Our work presents a promising avenue for scaling the proposed approach to larger molecular systems, achieving zero-shot generalization to unseen molecules, and including the generation of the local structure into the GFlowNet model.
RL, but don’t do anything I wouldn’t do
Michael K. Cohen
Marcus Hutter
Stuart Russell
In reinforcement learning (RL), if the agent’s reward differs from the designers’ true utility, even only rarely, the state distribution… (voir plus) resulting from the agent’s policy can be very bad, in theory and in practice. When RL policies would devolve into undesired behavior, a common countermeasure is KL regularization to a trusted policy ("Don’t do anything I wouldn’t do"). All current cutting-edge language models are RL agents that are KL-regularized to a "base policy" that is purely predictive. Unfortunately, we demonstrate that when this base policy is a Bayesian predictive model of a trusted policy, the KL constraint is no longer reliable for controlling the behavior of an advanced RL agent. We demonstrate this theoretically using algorithmic information theory, and while systems today are too weak to exhibit this theorized failure precisely, we RL-finetune a language model and find evidence that our formal results are plausibly relevant in practice. We also propose a theoretical alternative that avoids this problem by replacing the "Don’t do anything I wouldn’t do" principle with "Don’t do anything I mightn’t do".
RL, but don't do anything I wouldn't do
Michael K. Cohen
Marcus Hutter
Stuart Russell
In reinforcement learning (RL), if the agent's reward differs from the designers' true utility, even only rarely, the state distribution res… (voir plus)ulting from the agent's policy can be very bad, in theory and in practice. When RL policies would devolve into undesired behavior, a common countermeasure is KL regularization to a trusted policy ("Don't do anything I wouldn't do"). All current cutting-edge language models are RL agents that are KL-regularized to a "base policy" that is purely predictive. Unfortunately, we demonstrate that when this base policy is a Bayesian predictive model of a trusted policy, the KL constraint is no longer reliable for controlling the behavior of an advanced RL agent. We demonstrate this theoretically using algorithmic information theory, and while systems today are too weak to exhibit this theorized failure precisely, we RL-finetune a language model and find evidence that our formal results are plausibly relevant in practice. We also propose a theoretical alternative that avoids this problem by replacing the "Don't do anything I wouldn't do" principle with "Don't do anything I mightn't do".
Can a Bayesian Oracle Prevent Harm from an Agent?
Is there a way to design powerful AI systems based on machine learning methods that would satisfy probabilistic safety guarantees? With the … (voir plus)long-term goal of obtaining a probabilistic guarantee that would apply in every context, we consider estimating a context-dependent bound on the probability of violating a given safety specification. Such a risk evaluation would need to be performed at run-time to provide a guardrail against dangerous actions of an AI. Noting that different plausible hypotheses about the world could produce very different outcomes, and because we do not know which one is right, we derive bounds on the safety violation probability predicted under the true but unknown hypothesis. Such bounds could be used to reject potentially dangerous actions. Our main results involve searching for cautious but plausible hypotheses, obtained by a maximization that involves Bayesian posteriors over hypotheses. We consider two forms of this result, in the i.i.d. case and in the non-i.i.d. case, and conclude with open problems towards turning such theoretical results into practical AI guardrails.
Discrete Feynman-Kac Correctors
The performance of Large Language Models (LLMs) directly depends on the size of the context that the model was trained on. Despite significa… (voir plus)nt progress in increasing the context size of the current models, some applications remain bottlenecked by the number of processed tokens at inference time. A particular mathematical problem LLMs can be used for is inferring parameters in a statistical model, given data-points as input. Here we make a case demonstrating that discrete diffusion models offer a promising avenue for scaling such parameter prediction tasks, by combining the outputs of the same model evaluated on different parts of the training data. We propose Discrete Fenyman-Kac Correctors --- a framework that allows for controlling the generated distribution of discrete masked diffusion models at inference time. We derive Sequential Monte Carlo (SMC) algorithms that, given a trained discrete diffusion model, sample from its annealed distribution or the product of distributions with different conditions. Notably, our framework does not require any training, finetuning and external reward functions. Finally, we apply our framework to amortized linear regression using LLaDA and demonstrate that it drastically outperforms the standard inference procedure in terms of accuracy and adherence to prompt format.
HVAC-GRACE: Transferable Building Control via Heterogeneous Graph Neural Network Policies
Buildings consume 40% of global energy, with HVAC systems responsible for up to half of that demand. As energy use grows, optimizing HVAC ef… (voir plus)ficiency is critical to meeting climate goals. While reinforcement learning (RL) offers a promising alternative to rule-based control, real-world adoption is limited by poor sample efficiency and generalisation. We introduce HVAC-GRACE, a graph-based RL framework that models buildings as heterogeneous graphs and integrates spatial message passing directly into temporal GRU gates. This enables each zone to learn control actions informed by both its own history and its structural context. Our architecture supports zero-shot transfer by learning topology-agnostic functions—but initial experiments reveal that this benefit depends on sufficient conditioned zone connectivity to maintain gradient flow. These findings highlight both the promise and the architectural requirements of scalable, transferable RL for building control