Generalization performance of learning agents depends on the training experience to which they have been exposed. In game-playing domains, that experience is determined by the opponents faced during learning. This analytical study investigates two characteristics of opponents in competitive coevolutionary learning: behavioral diversity and difficulty (performance against other players). To assess diversity, we propose a generic intra-game behavioral distance measure, that could be adopted to other sequential decision problems. We monitor both characteristics in two-population coevolutionary learning of Othello strategies, attempting to explain their relationship with the generalization performance achieved by the evolved solutions. The main observation is the existence of a non-obvious trade-off between difficulty and diversity, with the latter being essential for obtaining high generalization performance.
@INPROCEEDINGS { SzubertEvoGames2015,
ABSTRACT = { Generalization performance of learning agents depends on the training experience to which they have been exposed. In game-playing domains, that experience is determined by the opponents faced during learning. This analytical study investigates two characteristics of opponents in competitive coevolutionary learning: behavioral diversity and difficulty (performance against other players). To assess diversity, we propose a generic intra-game behavioral distance measure, that could be adopted to other sequential decision problems. We monitor both characteristics in two-population coevolutionary learning of Othello strategies, attempting to explain their relationship with the generalization performance achieved by the evolved solutions. The main observation is the existence of a non-obvious trade-off between difficulty and diversity, with the latter being essential for obtaining high generalization performance. },
AUTHOR = { Szubert, Marcin and Ja{\'s}kowski, Wojciech and Liskowski, Pawe{\l} and Krawiec, Krzysztof },
BOOKTITLE = { Applications of Evolutionary Computation },
DOI = { 10.1007/978-3-319-16549-3_32 },
EDITOR = { Mora, Antonio M. and Squillero, Giovanni },
ISBN = { 978-3-319-16548-6 },
KEYWORDS = { Behavioral diversity; Diversity maintenance; Test difficulty; Competitive coevolution; Generalization performance; Games; Othello },
PAGES = { 394-405 },
PUBLISHER = { Springer International Publishing },
SERIES = { Lecture Notes in Computer Science },
TITLE = { The Role of Behavioral Diversity and Difficulty of Opponents in Coevolving Game-Playing Agents },
URL = { http://dx.doi.org/10.1007/978-3-319-16549-3_32 },
VOLUME = { 9028 },
YEAR = { 2015 },
1 = { http://dx.doi.org/10.1007/978-3-319-16549-3_32 },
}