Waris Radji

Information-Based Exploration via Random Features for Reinforcement Learning

Waris Radji, Odalric-Ambrym Maillard

Representation learning has enabled classical exploration strategies to be extended to deep Reinforcement Learning (RL), but often makes algorithms more complex and theoretical guarantees harder to establish. We introduce Random Feature Information Gain (RFIG), grounded in Bayesi…Read more