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

Venue:
IEEE Conference on Games (CoG), 2021

Topics:
general game playing, heuristics, machine learning, ludemes, supervised learning

Links: PDF · IEEE · arXiv

This paper investigates the performance of different general-game-playing heuristics within the Ludii system and explores whether their effectiveness can be predicted from game descriptions.

Using ludemes as features, several regression models are trained to estimate heuristic performance across a large set of games, enabling the prediction of suitable heuristics without exhaustive evaluation.

Results show that ludeme-based representations provide useful signals for predicting heuristic effectiveness, supporting the development of general AI agents capable of adapting to new games.

This work contributes to the development of general game playing systems by addressing the challenge of selecting effective heuristics across diverse games.

It leverages the Ludii framework and its ludeme-based representation to bridge symbolic game descriptions and machine learning models.

The approach is particularly relevant for building adaptive and general AI agents, aligning with broader research on general and human-like intelligence.

Stephenson, M., Soemers, D. J. N. J., Piette, E., Browne, C. (2021). General Game Heuristic Prediction Based on Ludeme Descriptions. IEEE Conference on Games (CoG).

@inproceedings{stephenson2021heuristic,
  author    = {Stephenson, Matthew and Soemers, Dennis J. N. J. and Piette, Eric and Browne, Cameron},
  title     = {General Game Heuristic Prediction Based on Ludeme Descriptions},
  booktitle = {IEEE Conference on Games (CoG)},
  year      = {2021},
  url       = {https://ieeexplore.ieee.org/document/9619052}
}