1316 Evolutionary Acquisition a Frame of Reinforce Learning in Multi-agent System
Hideaki YAJIMA, Kazuhiro Ohkura, Kanji Ueda · The Proceedings of Conference of Kansai Branch · 2000
We propose an approach to acquire frames of Reinforcement Learning (RL) by an evolutionary method for multi-agents systems. It parts the subspace in which important information is seen, because RL does not work efficiently in huge search spaces. It is difficult to design frames which induce RL performs well. Besides it may make flexibility of the systems down to fix frames in advance because they depend on problems. Based on this motivation, evolutionary method is applied to acquire the RL frame autonomically.