Improving generalization ability in a puzzle game using reinforcement learning
Hiroya Oonishi, Hitoshi Iima · 2017
Nowadays machine learning has attracted much attention. In order to apply it to various problems without relearning, its generalization ability is needed. Geometry Friends is a puzzle game where a player has to collect all targets in a two-dimensional world, and it is used in some artificial intelligence competitions. Although sufficient generalization ability is needed to apply the machine learning to this game, such machine learning methods are not proposed yet. In this paper, we propose a method based on reinforcement learning in which the generalization ability is improved for Geometry Friends.