ベイズ最適化を用いた深層強化学習のハイパーパラメータの最適化に関する検討

カテゴリ: 部門大会

論文No: TC15-1

グループ名: 【C】2023年電気学会電子・情報・システム部門大会

発行日: 2023/08/23

タイトル(英語): A Study on Optimization of Hyper-Parameters in Deep Reinforcement Learning by Bayesian Optimization

著者名: 曽田 涼介（松江工業高等専門学校）,西村 拓人（松江工業高等専門学校）,堀内 匡（松江工業高等専門学校）

著者名(英語): Ryosuke Sota (National Institute of Technology, Matsue College),Nishimura Takuto (National Institute of Technology, Matsue College),Tadashi Horiuchi (National Institute of Technology, Matsue College)

キーワード: 深層強化学習|ベイズ最適化|移動ロボット|行動獲得|Deep Reinforcement Learning|Bayesian Optimization|Mobile Robot|Behavior Acquisition

要約(日本語): Deep reinforcement learning is a machine learning method that combines deep learning and reinforcement learning. Deep Q-network (DQN) is one of the typical methods of deep reinforcement learning. We have applied DQN method to the robot navigation problem. The values of hyper-parameters including the network structure of DQN have been determined empirically. In this study, we attempt to optimize the values of hyper- parameters of deep reinforcement learning by using Bayesian optimization. We realized to optimize the values of hyper-parameters including the network structure of DQN by Optuna, a framework of Bayesian optimization. We confirmed that the values of hyper-parameters obtained by Optuna have higher learning performance than those by empirical method.