Abalearn: a risk-sensitive approach to self-play learning in Abalone

• Main Author: Campos, Pedro
• Publication Date: 2003
• Other Authors: Langlois, Thibault
• Language: eng
• Source: Repositórios Científicos de Acesso Aberto de Portugal (RCAAP)
• Download full: http://hdl.handle.net/10400.13/4603
• topic: Abalearn; Self-play learning; Abalone; .; Faculdade de Ciências Exatas e da Engenharia

This paper presents Abalearn, a self-teaching Abalone pro gram capable of automatically reaching an intermediate level of play without needing expert-labeled training examples, deep searches or ex posure to competent play. Our approach is based on a reinforcement learning algorithm that is risk seeking, since defensive players in Abalone tend to never end a game. We show that it is the risk-sensitivity that allows a successful self-play training. We also propose a set of features that seem relevant for achiev ing a good level of play. We evaluate our approach using a fixed heuristic opponent as a bench mark, pitting our agents against human players online and comparing samples of our agents at different times of training.