Deep learning and computer chess (part 1)

Abstract

This project explores advanced neural network architectures for chess position evaluation and move prediction, extending the foundational ideas introduced by the Giraffe model. We propose a novel supervised multi-head evaluation network that combines structured feature decomposition with modular design to capture complementary aspects of strategic and tactical reasoning in chess.

This work advances the intersection of classical chess knowledge and modern deep learning, offering a principled framework for strategic reasoning in complex decision-making domains.