Authors:
Dennis J. N. J. Soemers, Eric Piette, Cameron Browne

Venue:
2019 IEEE Congress on Evolutionary Computation (CEC), 2019

DOI:
10.1109/CEC.2019.8790141

Topics:
General Game Playing, Monte Carlo Tree Search, feature learning, self-play, interpretable AI

Links: PDF · ACM / IEEE entry · arXiv

This paper proposes using a linear function approximator, rather than a deep neural network, to bias a Monte Carlo Tree Search (MCTS) player for general games.

The approach uses interpretable features describing local patterns, learned through self-play, with the aim of improving playing strength while keeping the method general, lightweight, and easier to analyse than deep learning alternatives.

Experiments across a variety of board games show that an MCTS player biased by learned features can significantly outperform a standard UCT player in the majority of tested games after only a limited amount of self-play training.

This paper extends earlier work on general strategic features by integrating them directly into the MCTS learning loop for general games.

Instead of relying on large game-specific neural architectures, it explores a more lightweight and interpretable alternative based on feature growth and linear policies, which fits well with the broader goals of the Digital Ludeme Project.

The work is particularly important for large-scale experimentation across many games and variants, where computational efficiency, transferability, and explainability are essential.

Soemers, D. J. N. J., Piette, E., Browne, C. (2019). Biasing MCTS with Features for General Games. In 2019 IEEE Congress on Evolutionary Computation (CEC). DOI: 10.1109/CEC.2019.8790141

@inproceedings{soemers2019biasing,
  author    = {Soemers, Dennis J. N. J. and Piette, Eric and Browne, Cameron},
  title     = {Biasing MCTS with Features for General Games},
  booktitle = {2019 IEEE Congress on Evolutionary Computation (CEC)},
  year      = {2019},
  doi       = {10.1109/CEC.2019.8790141},
  url       = {https://dl.acm.org/doi/10.1109/CEC.2019.8790141}
}