Seminar: Causal Decision-Making, Why and How?

Abstract

In this talk, I'll present some of my past and ongoing works, and lay down some ideas for future research in the direction of causal decision making. In particular, I'll motivate the idea that reinforcement learning (RL) is a fundamentally causal problem, and I'll show how merging the tools and ideas from causality with those from statistical learning widens the potential of RL algorithms, in a principled way.

Bio

Maxime Gasse is a researcher at Polytechnique Montréal, an associate academic member of MILA, and a member of the Data Science for Real-Time Decision Making CERC. The fundamental question that drives his research is: can machines think ? He has been working on a variety of fundamental machine learning problems such as probabilistic graphical model structure learning (PhD thesis), and on interdisciplinary topics such as machine learning applied to medical imaging, or machine learning applied to combinatorial optimization (Ecole library). Today he is mostly interested in investigating if and how the tools from causality can help in the design of autonomous agents. He is proud to serve the scientific community, and he has been rewarded NeurIPS 2019 best reviewer (top 40%), ICML 2020 best reviewer (top 30%), NeurIPS 2020 top reviewer (top 10%), ICLR 2021 outstanding reviewer (top 10%), and ICML 2021 expert reviewer.