Additional information
by Dominik Klein (Author)
Deep Neural Networks have revolutionized computer engines for Go, Shogi and chess. Finally, computers are able to evaluate a game position similar to the way human experts do it. By that, computers are able to identify long-term strategic advantages and disadvantages. But how do chess engines based on neural networks such as AlphaZero, Leela Chess Zero actually work? This book gives an answer to that question. With lots of practical examples and illustrations, all basic building blocks that are required to understand modern computer chess are introduced. Based on that, the concepts of both classic and modern chess engines are explained. Finally, a miniature version of AlphaZero to play the game Hexapawn is implemented in Python.
Chapters include: Single-Layer and Multilayer Perceptrons, Back-Propagation and Gradient Descent, Classification and Regression, Network Vectorization, Convolutional Layers, Squeeze and Excitation Networks, Fully Connected Layers, Batch Normalization, Rectified Linear Unit (ReLU), Residual Layers, Minimax, Alpha-Beta Search, Monte-Carlo Tree Search, AlphaGo, AlphaGo Zero, AlphaZero, Leela Chess Zero (Lc0), Fat Fritz, Efficiently Updateable Neural Networks (NNUE), Fat Fritz 2, Maia, Supervised Learning Hexapawn, Reinforcement Learning of Hexapawn (Hexapawn Zero) The latest update from May 19th, 2023 adds an implementation example of gradient descent.
