Yann Ollivier
After receiving a Ph.D. in probability and group theory, Yann joined the CNRS, the French national research institute. He initially worked on unveiling connections between probability, Markov chains, differential geometry, and discrete geometry. For this work, he was awarded the bronze medal of the CNRS. Due to a lifelong interest in artificial intelligence, Yann focused his research on machine learning and joined the computer science department at Paris-Sud University. Following the industrial development of deep learning, he joined Facebook AI Research. Yann works on understanding and improving the learning algorithms for neural networks. In the long term, he is interested in building general artificial intelligence systems. More specific fields of interest include the geometry of gradient descent algorithms, the dynamics of recurrent networks and online learning, better algorithms for reinforcement learning, and what "learning" means in terms of information theory.
Yann's Publications
RESEARCH
ML APPLICATIONS
Separating value functions across time-scales
In many finite horizon episodic reinforcement learning (RL) settings, it is desirable to optimize for the undiscounted return – in settings like Atari, for instance, the goal is to collect the most points while staying alive in the long run.…
Joshua Romoff, Peter Henderson, Ahmed Touati, Emma Brunskill, Joelle Pineau, Yann Ollivier,
RESEARCH
RANKING & RECOMMENDATIONS
Making Deep Q-learning Methods Robust to Time Discretization
Despite remarkable successes, Deep Reinforcement Learning (DRL) is not robust to hyperparameterization, implementation details, or small environment changes (Henderson et al. 2017, Zhang et al. 2018). Overcoming such sensitivity is key to…
Corentin Tallec, Léonard Blier, Yann Ollivier,
RESEARCH
THEORY
First-order Adversarial Vulnerability of Neural Networks and Input Dimension
Over the past few years, neural networks were proven vulnerable to adversarial images: targeted but imperceptible image perturbations lead to drastically different predictions. We show that adversarial vulnerability increases with the gradients…
Carl-Johann Simon-Gabriel, Yann Ollivier, Bernhard Scholkopf, Leon Bottou, David Lopez-Paz,
