後継表現を用いた経験リプレイの改善
Improving Experience Replay with Successor Representation
強化学習の経験リプレイにおいて、TD誤差による「利得」だけでなく、現在の状況との関連性を示す「必要性」も考慮して優先順位付けする新手法を提案し、Dyna-Q迷路やAtariゲームで性能向上を示した。
著者: Yizhi Yuan, Marcelo G Mattar
分類: cs.LG, cs.AI, cs.RO
原文アブストラクト
Prioritized experience replay is a reinforcement learning technique whereby agents speed up learning by replaying useful past experiences. This usefulness is quantified as the expected gain from replaying the experience, a quantity often approximated as the prediction error (TD-error). However, recent work in neuroscience suggests that, in biological organisms, replay is prioritized not only by gain, but also by "need" -- a quantity measuring the expected relevance of each experience with respect to the current situation. Importantly, this term is not currently considered in algorithms such as prioritized experience replay. In this paper we present a new approach for prioritizing experiences for replay that considers both gain and need. Our proposed algorithms show a significant increase in performance in benchmarks including the Dyna-Q maze and a selection of Atari games.
