AI News, Deep Successor Reinforcement Learning (DSR)

Deep Successor Reinforcement Learning (DSR)

DSR is a hybrid model-free and model-based deep RL algorithm to learn robust value functions.

DSR has several appealing properties including: increased sensitivity to distal reward changes due to factorization of reward and world dynamics, and the ability to extract bottleneck states (subgoals) given successor maps trained under a random policy.