強化学習arXiv:1612.08967
効率的な反復方策最適化
Efficient iterative policy optimization
期待報酬を凹な下界の列で近似して効率的に最大化することで、少ない方策更新回数で高性能を得る手法を提案し、負の報酬にも拡張して制御変数を利用可能にした。
著者: Nicolas Le Roux
分類: cs.AI, cs.LG, cs.RO
原文アブストラクト
We tackle the issue of finding a good policy when the number of policy updates is limited. This is done by approximating the expected policy reward as a sequence of concave lower bounds which can be efficiently maximized, drastically reducing the number of policy updates required to achieve good performance. We also extend existing methods to negative rewards, enabling the use of control variates.
