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

Venue:
Advances in Computer Games (ACG), 2021

Topics:
Monte Carlo Tree Search, playout optimisation, general game playing, Ludii, AI performance

Links: PDF · Springer · ACM · arXiv

This paper introduces several optimised implementations of playouts for the Ludii general game system, targeting the efficiency of Monte Carlo Tree Search and related algorithms.

The proposed methods exploit structural properties of game rules to reduce computational overhead during simulations, including specialised strategies for different classes of games.

Experimental results demonstrate significant improvements in performance, with a median speedup of over five times across a large set of games, enabling stronger general game-playing agents.

This work addresses a key bottleneck in general game playing: the efficiency of playout-based search algorithms such as Monte Carlo Tree Search.

By leveraging the structured descriptions of games in Ludii, the system can automatically detect when optimisations are applicable, enabling adaptive performance improvements across a wide range of games.

These contributions are central to scaling general AI systems and improving their practical performance without sacrificing generality.

Soemers, D. J. N. J., Piette, E., Stephenson, M., Browne, C. (2021). Optimised Playout Implementations for the Ludii General Game System. In Advances in Computer Games (ACG).

@inproceedings{soemers2021playouts,
  author    = {Soemers, Dennis J. N. J. and Piette, Eric and Stephenson, Matthew and Browne, Cameron},
  title     = {Optimised Playout Implementations for the Ludii General Game System},
  booktitle = {Advances in Computer Games (ACG)},
  year      = {2021},
  url       = {https://arxiv.org/abs/2111.02839}
}