arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 1901.09330cs.LGstat.ML
收藏

通过元学习的奖励塑造

Reward Shaping via Meta-Learning

Tsinghua University(清华大学)

Haosheng Zou, Tongzheng Ren, Dong Yan, Hang Su, Jun Zhu

摘要:

奖励塑造是解决强化学习(RL)中信用分配这一关键而又具有挑战性的问题的最有效方法之一。然而,设计成形函数通常需要大量的专业知识和手工工程,并且考虑到要解决的多个类似任务,困难进一步加剧。在本文中,我们考虑奖励成形的任务分布,并提出了一个通用的元学习框架,自动学习有效的奖励塑造新采样的任务,假设只有共享的状态空间,但不一定行动空间。首先,我们推导出理论上最优的奖励塑造的信用分配无模型RL。然后,我们提出了一个基于值的元学习算法,以提取一个有效的先验的最佳奖励塑造。先验可以直接应用于新的任务,或可证明地适应任务后验,同时在几个梯度更新内解决任务。我们通过显着提高学习效率和跨各种设置的可解释的可视化来展示我们的塑造的有效性,特别是从DQN到DDPG的成功转移。

英文摘要:

Reward shaping is one of the most effective methods to tackle the crucial yet challenging problem of credit assignment in Reinforcement Learning (RL). However, designing shaping functions usually requires much expert knowledge and hand-engineering, and the difficulties are further exacerbated given multiple similar tasks to solve. In this paper, we consider reward shaping on a distribution of tasks, and propose a general meta-learning framework to automatically learn the efficient reward shaping on newly sampled tasks, assuming only shared state space but not necessarily action space. We first derive the theoretically optimal reward shaping in terms of credit assignment in model-free RL. We then propose a value-based meta-learning algorithm to extract an effective prior over the optimal reward shaping. The prior can be applied directly to new tasks, or provably adapted to the task-posterior while solving the task within few gradient updates. We demonstrate the effectiveness of our shaping through significantly improved learning efficiency and interpretable visualizations across various settings, including notably a successful transfer from DQN to DDPG.

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。

↑