Yann Ollivier

Papers

Total Citations

H-Index

About

Yann Ollivier is a leading researcher in machine learning and theoretical computer science, best known for his groundbreaking work on representation learning and reinforcement learning. His most cited paper, "Learning One Representation to Optimize All Rewards" (2021, 5 citations), introduces the forward-backward (FB) representation of dynamics in reward-free Markov decision processes. This innovative framework allows agents to learn a single, unsupervised representation from reward-free interactions, which can then be used to derive near-optimal policies for any reward specified after learning. Ollivier's contributions bridge the gap between unsupervised learning and sequential decision-making, offering a principled approach to generalization across tasks. His work has been highly influential in advancing the theoretical foundations of reinforcement learning, particularly in multi-task and transfer learning settings. Beyond this, Ollivier is recognized for his research on curvature in optimization and information geometry, which has shaped modern understanding of deep learning dynamics. His ability to combine rigorous theory with practical algorithms makes his research essential reading for students and researchers aiming to build more flexible, reward-agnostic AI systems.