MCTS with UCT is the base for a series of breakthroughs developed by DeepMind. These include the Go-playing AlphaGo, its improved version AlphaGo Zero, and finally AlphaZero (https://arxiv.org/abs/1712.01815), an improvement on AlphaGo Zero for playing multiple games, such as Chess and Shogi. Let's discuss AlphaZero. For the sake of simplicity, we'll assume that we want to teach the agent to play chess. Each state of the environment will be one configuration of the board (the positions of the pieces). By taking a turn (moving a piece), the players transition the environment from one state to another.
At the center of the algorithm is a neural network, which takes as input the current state of the board and ...
