Imperfect Information Learning Team Seminar (Talk by Xinqiang Cai, The University of Tokyo).

This is an online seminar. Registration is required.
【 Imperfect Information Learning Team】
【Date】2024/February/27(Tue) 14:00-15:00(JST)
【Speaker】Xinqiang Cai, Complexity Science and Engineering, The University of Tokyo

Title: Reward Modeling in Reinforcement and Imitation Learning: Overcoming Challenges of Dimensionality, Imperfection, and Uncertainty

Abstract:
In the field of artificial intelligence, reinforcement and imitation learning stand out for their potential in enabling agents to learn complex behaviors through interaction with the environment or by imitating expert demonstrations. However, the effectiveness of these learning paradigms is significantly influenced by the design and implementation of reward models, which often face challenges such as high dimensionality, imperfection in demonstrations, and uncertainty in outcomes. This presentation delves into innovative approaches for reward modeling, addressing these challenges across various environments and scenarios , including the following topics: balancing multiple objectives ; dealing with bagged rewards; imitation in high
dimensional spaces; learning from imperfect demonstrations; adapting to heterogeneous observations; interpreting vague feedback.