arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2006.07549cs.LGstat.ML
收藏

目标条件强化学习的后见期望最大化

Hindsight Expectation Maximization for Goal-conditioned Reinforcement Learning

Columbia University(哥伦比亚大学);Fero Labs(费罗实验室)

Yunhao Tang, Alp Kucukelbir

摘要:

我们提出了一个图形化的模型框架目标条件强化学习,与EM算法,操作的RL目标的下限。E步骤提供了一个自然的解释如何“事后学习”的技术,如HER,处理极其稀疏的目标条件奖励。M-step将策略优化简化为监督学习更新,这极大地稳定了对图像等高维输入的端到端训练。我们表明,组合算法,hEM显着优于无模型的基线上广泛的目标条件的基准稀疏奖励。

英文摘要:

We propose a graphical model framework for goal-conditioned RL, with an EM algorithm that operates on the lower bound of the RL objective. The E-step provides a natural interpretation of how 'learning in hindsight' techniques, such as HER, to handle extremely sparse goal-conditioned rewards. The M-step reduces policy optimization to supervised learning updates, which greatly stabilizes end-to-end training on high-dimensional inputs such as images. We show that the combined algorithm, hEM significantly outperforms model-free baselines on a wide range of goal-conditioned benchmarks with sparse rewards.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑