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 2102.10264cs.LGcs.ROstat.ML
收藏

关于近端策略优化的重尾分布

On Proximal Policy Optimization's Heavy-tailed Gradients

Carnegie Mellon University(卡内基梅隆大学)

Saurabh Garg, Joshua Zhanson, Emilio Parisotto, Adarsh Prasad, J. Zico Kolter, Zachary C. Lipton, Sivaraman Balakrishnan, Ruslan Salakhutdinov, Pradeep Ravikumar

摘要:

诸如近端策略优化(PPO)的现代策略梯度算法依赖于包括损失裁剪和梯度裁剪的一系列算法来确保成功的学习。这些统计学让人想起稳健统计学的技术,通常用于异常值丰富(“重尾”)制度的估计。在本文中,我们提出了一个详细的实证研究,以表征PPO代理奖励函数的梯度的重尾性质。我们证明了梯度,特别是演员网络,表现出明显的重尾性,并且随着代理的策略偏离行为策略(即,随着代理进一步偏离策略)。进一步的研究表明,似然比和替代奖励的优势是观察到的重尾的主要来源。然后,我们强调由于梯度的重尾性质而产生的问题。在这种情况下,我们研究了标准PPO裁剪算法的效果,证明这些技巧主要用于抵消梯度中的重尾。因此,出于动机,我们建议将Gestival,一个高维的鲁棒估计,PPO作为替代三个裁剪技巧。尽管需要更少的超参数调整,我们的方法匹配的性能PPO(与所有的动力学启用)电池的MuJoCo连续控制任务。

英文摘要:

Modern policy gradient algorithms such as Proximal Policy Optimization (PPO) rely on an arsenal of heuristics, including loss clipping and gradient clipping, to ensure successful learning. These heuristics are reminiscent of techniques from robust statistics, commonly used for estimation in outlier-rich (``heavy-tailed'') regimes. In this paper, we present a detailed empirical study to characterize the heavy-tailed nature of the gradients of the PPO surrogate reward function. We demonstrate that the gradients, especially for the actor network, exhibit pronounced heavy-tailedness and that it increases as the agent's policy diverges from the behavioral policy (i.e., as the agent goes further off policy). Further examination implicates the likelihood ratios and advantages in the surrogate reward as the main sources of the observed heavy-tailedness. We then highlight issues arising due to the heavy-tailed nature of the gradients. In this light, we study the effects of the standard PPO clipping heuristics, demonstrating that these tricks primarily serve to offset heavy-tailedness in gradients. Thus motivated, we propose incorporating GMOM, a high-dimensional robust estimator, into PPO as a substitute for three clipping tricks. Despite requiring less hyperparameter tuning, our method matches the performance of PPO (with all heuristics enabled) on a battery of MuJoCo continuous control tasks.

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。

↑