Noun
self-play (uncountable)
(machine learning) A technique for improving the performance of reinforcement learning agents by having them play against themselves.
Self-play was used by the AlphaZero program to improve its performance in the games of chess, shogi and go.