Theoretical considerations for practical meta reinforcement learning
26 May 2025, 14:00 — Room 322, UniGe DIBRIS/DIMA, Via Dodecaneso 35
![Mirco.Mutti - [Occhiali da vista Mento Guancia]](/media/images/Mirco_waiKOcG.width-800.jpg)
Speaker:
Mirco Mutti — Technion
Mirco Mutti — Technion
Abstract:
Reinforcement Learning (RL) is a powerful framework to solve sequential decision-making problems from sampled interactions. Meta RL aims at exploiting the experience gathered from solving some RL problems to efficiently solve more and more RL problems. This talk will touch upon some theoretical aspects of meta RL that may shed light on its potential and pitfalls in practice, addressing questions of the likes of: When does meta RL improves the efficiency of vanilla RL? How can we meta learn interpretable RL algorithms? How can we meta learn from human demonstrations?
Reinforcement Learning (RL) is a powerful framework to solve sequential decision-making problems from sampled interactions. Meta RL aims at exploiting the experience gathered from solving some RL problems to efficiently solve more and more RL problems. This talk will touch upon some theoretical aspects of meta RL that may shed light on its potential and pitfalls in practice, addressing questions of the likes of: When does meta RL improves the efficiency of vanilla RL? How can we meta learn interpretable RL algorithms? How can we meta learn from human demonstrations?
Bio:
Mirco Mutti is a postdoctoral researcher working with Aviv Tamar in the Robots Learning Lab at the Technion. Formerly, he was a PhD student at Politecnico di Milano (jointly with Università di Bologna) under the supervision of Marcello Restelli. His research focuses on the theoretical and methodological foundations of reinforcement learning, including unsupervised pre-training, partial observability, and meta reinforcement learning.
Mirco Mutti is a postdoctoral researcher working with Aviv Tamar in the Robots Learning Lab at the Technion. Formerly, he was a PhD student at Politecnico di Milano (jointly with Università di Bologna) under the supervision of Marcello Restelli. His research focuses on the theoretical and methodological foundations of reinforcement learning, including unsupervised pre-training, partial observability, and meta reinforcement learning.