D-VisionDraughts: a draughts player neural network that learns by reinforcement in a high performance environment.
Ayres Roberto Araújo Barcelos, Rita Maria da Silva Julia, Rivalino Matias · The European Symposium on Artificial Neural Networks · 2011
This paper describes D-VisionDraughts, a distributed player agent for draughts which is based on Neural Networks trained by Temporal Differences. DVisionDraughs is trained in a high performance environment and achieves a high level of play without expert game analysis and with minimum human intervention. D-VisionDraughts corresponds to a distributed version of the efficient agent player VisionDraughts. In this way, the main contributions of this paper consist on substituting the distributed Young Brothers Wait Concept algorithm (YBWC) for the serial alpha-beta search algorithm used in VisionDraughts and on measuring the impact of a high performance environment into the non-supervised learning abilities of the player. Evaluative tests proved that even a modest distributed version counting just on ten processors is able to reduce from about 83% the search runtime and to increase from 15% its capacity of winning.