VisionDraughts - um sistema de aprendizagem de jogos de damas baseado em redes neurais, diferenças temporais, algoritmos eficientes de busca em árvores e informações perfeitas contidas em bases de dados

• Ano de defesa: 2008
• Autor(a) principal: Caexêta, Gutierrez Soares
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Uberlândia; BR; Programa de Pós-graduação em Ciência da Computação; Ciências Exatas e da Terra; UFU
• Palavras-chave em Português: Aprendizagem automática; Aprendizagem de máquina; Aprendizagem incremental; Aprendizagem por reforço; Redes neurais; Aprendizagem por diferenças temporais; Teoria dos jogos; Damas; Busca em árvores de jogos; MiniMax; Tabelas de transposição; Aprofundamento iterativo; Bases de dados de finais de jogos; Chaves Zobrist; Tabela Hash; Tratamento de colisões; Aprendizado do computador; Redes neurais (Computação); Alfa-Beta; Automatic Learning; Machine Learning; Incremental Learning; Reinforcement Learning; Neural network; Temporal diference learning; Game theory; Draughts; Checkers; Minimax; Transposition table; Iterative deepening; Endgame databases; Zobrist key; Hash table; Collisions; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: https://repositorio.ufu.br/handle/123456789/12460

The objective of this work is to propose a draughts learning system, VisionDraughts, based on works of Neto and Julia (LS-Draughts) and Mark Lynch (NeuroDraughts). The NeuroDraughts is a good automatic draughts player which uses temporal diference learning to adjust the weights of an articial neural network whose role is to estimate how much the board state represented in its input layer by NET-FEATUREMAP is favorable to the player agent. The set of features is manually dened. The search for the best action corresponding to a current board state is performed by minimax algorithm. The LS-Draughts expands the NeuroDraughts, through the genetic algorithms, generating automatically a set of minimal features which are necessary and essential to a game of draughts and optimizing, successfully, the training of the apprentice player. The Vision- Draughts adds two modules to the former architectures: an eficient tree-search module with alfa-beta, iterative deepening and transposition table, providing the player agent larger capacity to analyse future moves (board states more distant from the current board) and a module to access endgame databases, allowing to acquire perfect information to positions with less than 8 pieces on the board. Some tournaments were promoted between the best players obtained by NeuroDraughts, LS-Draughts and VisionDraughts. The tournament's results, all won by the VisionDraughts, show the importance of the new two modules in the building of good automatic draughts players: the runtime required for training the new player was drastically reduced and its performance was significantly improved. Furthermore, the VisionDraughts name was just chosen to emphasize the great importance of analysing future moves in order to the success of this work.