TD-Gammon

PureTD model achieves strong backgammon performance with no search

Researchers have developed PureTD, a reinforcement learning model for backgammon money games that achieves near state-of-the-art performance without requiring search during evaluation. Trained solely through self-play, …