- This event has passed.
Alessandro LAZARIC (Facebook AI) – "Exploration-exploitation in reinforcement learning with function approximation"
Time: 2:00 pm – 3:15 pm
Date: 15th of March 2021
Alessandro LAZARIC (Facebook AI) – “Exploration-exploitation in reinforcement learning with function approximation”
Abstract: Most of recent reinforcement learning algorithms combine standard dynamic programming algorithms (e.g., value iteration) with advanced function approximation techniques (notably, adapted from deep learning). Nonetheless, the effect of function approximation on the online performance of RL algorithms is still relatively poorly understood.
In this seminar, we will review how exploration-exploitation techniques can be paired with RL function approximation methods in environments with continuous state-action spaces. In particular, we will focus on linear approximations and show how they impact the learning performance. We will also review the different assumptions that have been used in this space and what are the major open questions.
Cristina BUTUCEA (CREST), Alexandre TSYBAKOV (CREST), Karim LOUNICI (CMAP) , Zoltan SZABO (CMAP)