Learning: An Effective Approach in Endgame Chess Board Evaluation

Classical chess engines exhaustively explore moving possibilities from a chess board position to decide what the next best move to play is. The main component of a chess engine is board evaluation function. In this article we present a new method to solve chess endgames optimally without using brute-force algorithms or endgame tables. We propose to use artificial neural network to obtain better evaluation function for endgame positions. This method is specifically applied to three classical endgames: king-bishop-bishop-king, king-rook-king, and king-queen-king. The empirical results show that the proposed learning strategy is effective in wining against an opponent who offers its best survival defense using Nalimov database of best endgame moves.