Alpha Zero vs. Alpha Zero

Videoperformance (8:00 min), 2020-2021

Alpha Zero is trained through self-play reinforcement learning. It combines a neural network and a Monte Carlo tree search in a policy iteration framework. AI systems have achieved superhuman performance in games. All of these games involve only two players and are zero-sum games (meaning that whatever one player wins, the other loses). To verify robustness, it plays additional matches starting from common human openings.