“Learning to Play Chess With Minimal Lookahead and Deep Value Neural Networks”, 2017-10-30 ():
The game of chess has always been a very important testbed for the artificial intelligence community. Even though the goal of training a program to play as good as the strongest human players is not considered as a hard challenge anymore, so far no work has been done in creating a system that does not have to rely on expensive lookahead algorithms to play the game at a high level.
In this work we show how carefully trained Value Neural Networks are able to play high level chess without looking ahead more than one move [by imitation learning of Stockfish evaluations]. To achieve this, we have investigated the capabilities that Artificial Neural Networks (ANNs) have when it comes to pattern recognition, an ability that distinguishes chess Grandmasters from the more amateur players.
We firstly propose a novel training approach specifically designed for pursuing the previously mentioned goal.
Secondly, we investigate the performances of both Multilayer Perceptrons (MLPs) and Convolutional Neural Networks (CNNs) as optimal neural architecture in chess. After having assessed the superiority of the first architecture, we propose a novel input representation of the chess board that allows CNNs to outperform MLPs for the first time as chess evaluation functions.
We finally investigate the performances of our best ANNs on a state-of-the-art test, specifically designed to evaluate the strength of chess playing programs.
Our results show how it is possible to play high quality chess only with Value Neural Networks, without having to rely on techniques involving lookahead.
…, the level reached by the best performing ANNs is still remarkable since it reached an Elo rating of ~2000 on a reputable chess server [chess24]… The ANN played against opponents with an Elo rating 1741–2140 and obtained a final game playing performance corresponding to a strong Candidate Master-titled player.
…Is it possible to use Convolutional Neural Networks in chess?
We show that this is possible both in Chapter 6 and in Chapter 7. To do so, it is extremely important to design the ANN architecture in such a way that the input preserves as much geometrical information as possible during the training process, as has been highlighted in Chapter 4. Training this ANN results in lower performances when compared to the ones obtained by the MLP on standard state-of-the-art board representations. However, when combined with the novel representation presented in Chapter 7 its performances become even better than the ones obtained by MLPs. Furthermore, we also show how the training time required by this ANN is much more efficient when compared to the ones required by the MLP, if appropriate GPU support is provided. This is particularly the case for the CNNs trained on the Feature Input which converged in ~36 hours for the experiment performed on Dataset 4.