Portrait de Francois Rivest

Francois Rivest

Membre affilié
Royal Military College of Canada
Sujets de recherche
Apprentissage continu
Apprentissage de représentations
Apprentissage en ligne
Apprentissage par renforcement
IA inspirée du cerveau
Neurosciences computationnelles
Réseaux de neurones récurrents
Site web
Google Scholar
LinkedIn

Publications

Detecting Naturally Emerging Temporal Abstractions
Jayesh Khullar
Francois Rivest
Breaking down large tasks into smaller sub-tasks, either to accelerate learning or enable transfer across related environments, remains a ce… (voir plus)ntral challenge in reinforcement learning (RL). One way Hierarchical Reinforcement Learning addresses this is by introducing temporal abstractions, often instantiated as options: temporally extended action sequences directed toward sub-goals. While prior work largely focuses on algorithms that explicitly learn such options, we ask a different question: can temporal abstractions emerge naturally within general deep reinforcement learning agents? To this end, we introduce Decorrelate Cluster Temporal Activation (DCTA) Analysis, a tool for detecting temporal abstractions in agents that do not explicitly model them. We validate this approach on a custom Four-Room environment and Atari benchmarks. Most importantly, we show that DQN and PPO naturally develop internal representations with semi-Markov consistent temporal structure — the defining statistical property of temporal abstractions — without explicit option learning objective.
2026-06-15
ContinualRL @ Reinforcement Learning Conference (publié)
openreview.net
openreview.net
Squeezing More from the Stream : Learning Representation Online for Streaming Reinforcement Learning
Nilaksh
Antoine Clavaud
Mathieu Reymond
Francois Rivest
Sarath Chandar
2026-04-29
International Conference on Machine Learning (accepté)
doi.org
openreview.net
Squeezing More from the Stream : Learning Representation Online for Streaming Reinforcement Learning
Nilaksh
Antoine Clavaud
Mathieu Reymond
Francois Rivest
Sarath Chandar
In streaming Reinforcement Learning (RL), transitions are observed and discarded immediately after a single update. While this minimizes res… (voir plus)ource usage for on-device applications, it makes agents notoriously sample-inefficient, since value-based losses alone struggle to extract meaningful representations from transient data. We propose extending Self-Predictive Representations (SPR) to the streaming pipeline to maximize the utility of every observed frame. However, due to the highly correlated samples induced by the streaming regime, naively applying this auxiliary loss results in training instabilities. Thus, we introduce orthogonal gradient updates relative to the momentum target and resolve gradient conflicts arising from streaming-specific optimizers. Validated across the Atari, MinAtar, and Octax suites, our approach systematically outperforms existing streaming baselines. Latent-space analysis, including t-SNE visualizations and effective-rank measurements, confirms that our method learns significantly richer representations, bridging the performance gap caused by the absence of a replay buffer, while remaining efficient enough to train on just a few CPU cores.
2025-12-31
International Conference on Machine Learning (Accept (regular))
doi.org
proceedings.mlr.press