"Near-optimal Regret in Online MDPs with Aggregate Bandit Feedback" - Tal Lancewicki, Research TTIC

Near-optimal Regret in Online MDPs with Aggregate Bandit Feedback

Description:
Tal Lancewicki explores the challenge of learning online Markov decision processes (MDPs) with aggregate bandit feedback, where agents receive only total trajectory loss instead of step-by-step feedback. He reviews existing approaches and introduces a new policy optimization algorithm to improve learning efficiency in this setting, with applications in AI, robotics, and reinforcement learning.

Tal Lancewicki, Tel Aviv University

Originally recorded on March 10, 2025.

Timestamps:
00:00 – Intro
00:55 – Talk
48:48 – Conclusion and Q&A

Tags:
#AI #MachineLearning #ReinforcementLearning #MDP #BanditAlgorithms #PolicyOptimization #Robotics #ComputationalTheory #ResearchTalk