Hier das englischsprachige Buch „Neural Networks for Chess“:
https://github.com/asdfjkl/neural_network_chesshttps://github.com/asdfjkl/neural_network_chess/releases/download/v1.5/Neural_Networks_For_Chess.pdfInhalt:
Contents
1 Introduction 11
2 A Crash Course into Neural Networks 19
2.1 ThePerceptron ............................. 20
2.2 Back-PropagationandGradientDescent . . . . . . . . . . . . . . .
24 2.3 ClassificationvsRegression ...................... 35
2.4 PuttingitintoPractice ......................... 38
2.5 InherentLimitsofaSinglePerceptron ................ 40
2.6 MultilayerPerceptrons ......................... 42
2.7 VectorizationandGPUs ........................ 48
2.8 ConvolutionalLayers.......................... 49
2.9 Squeeze-and-ExcitationNetworks................... 61
2.10FullyConnectedLayers......................... 62
2.11 Batch normalization and Rectified Linear Units . . . . . . . . . . . 63
2.12ResidualLayers ............................. 67
2.13OverfittingandUnderfitting...................... 69
2.14Summary................................. 73
3 Searching 75
3.1 Minimax................................. 80
3.2 Alpha-BetaSearch............................ 86
3.3 AdvancedTechniques ......................... 91
3.4 ANoteonDeepBlue.......................... 94
CONTENTS
3.5 Monte-CarloTreeSearch........................100
3.6 Summary.................................121
4 Modern AI Approaches - A Deep Dive 123
4.1 AlphaGo.................................126
4.2 AlphaGoZero..............................149
4.3 AlphaZero................................162
4.4 LeelaChessZero(Lc0) .........................174
4.5 FatFritz..................................182
4.6 Efficiently Updateable Neural Networks (NNUE) . . . . . . . . . . 191
4.7 FatFritz2.................................216
4.8 Maia ...................................217
4.9 Conclusion................................227
5 HexapawnZero 231
5.1 TheNetwork...............................233
5.2 GameLogic ...............................236
5.3 SupervisedLearning ..........................242
5.4 “Zero-like”ReinforcementLearning . . . . . . . . . . . . . . . . .249
5.5 Summary.................................259
6 Conclusion and Outlook