Reinforcement Learning for Board Games: The Temporal Dierence Algorithm
Wolfgang Konen · 2015
This technical report shows how the ideas of reinforcement learning (RL) and temporal dierence (TD) learning can be applied to board games. This report collects the main ideas from Sutton and Barto [1998], Tesauro [1992] and Sutton and Bonde [1992] in a compact form and gives hints for the practical application. It contains a section on ’Self-Play’ TD algorithms, a section on game-learning applications for TD( ), furthermore appendices on eligibility traces and typical function approximators.