Scholarlink

The main challenge of chess position evaluation is the size and complexity of the game tree. Consequently, chess engines rely on heuristic functions to estimate the position’s evaluation using a predefined maximum search depth. However, even heuristic functions can be computationally and memory intensive, requiring the game tree to be pruned to remove unlikely moves. In this article, multiple convolutional neural networks (CNNs) are trained to generate a specific player’s likely moves in a given position using a top-k accuracy model to set the upper bound of the game tree’s branching factor.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Convolutional Neural Networks for Chess Move Prediction with an Emphasis on Top-k Accuracy

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • William Moss,
  • Lisa Gandy

摘要

The main challenge of chess position evaluation is the size and complexity of the game tree. Consequently, chess engines rely on heuristic functions to estimate the position’s evaluation using a predefined maximum search depth. However, even heuristic functions can be computationally and memory intensive, requiring the game tree to be pruned to remove unlikely moves. In this article, multiple convolutional neural networks (CNNs) are trained to generate a specific player’s likely moves in a given position using a top-k accuracy model to set the upper bound of the game tree’s branching factor.