We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.LG

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo

Computer Science > Machine Learning

Title: Policy Optimization Reinforcement Learning with Entropy Regularization

Abstract: Entropy regularization is an important idea in reinforcement learning, with great success in recent algorithms like Soft Q Network (SQN) and Soft Actor-Critic (SAC1). In this work, we extend this idea into the on-policy realm. We propose the soft policy gradient theorem (SPGT) for on-policy maximum entropy reinforcement learning. With SPGT, a series of new policy optimization algorithms are derived, such as SPG, SA2C, SA3C, SDDPG, STRPO, SPPO, SIMPALA and so on. We find that SDDPG is equivalent to SAC1. For policy gradient, the policy network is often represented as a Gaussian distribution with a global action variance, which damages the representation capacity. We introduce a local action variance for policy network and find it can work collaboratively with the idea of entropy regularization. Our method outperforms prior works on a range of benchmark tasks. Furthermore, our method can be easily extended to large scale experiment with great stability and parallelism.
Subjects: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Machine Learning (stat.ML)
Cite as: arXiv:1912.01557 [cs.LG]
  (or arXiv:1912.01557v3 [cs.LG] for this version)

Submission history

From: Xinyang Gu [view email]
[v1] Mon, 2 Dec 2019 11:01:32 GMT (7099kb,D)
[v2] Fri, 20 Dec 2019 03:28:45 GMT (7276kb,D)
[v3] Fri, 16 Oct 2020 05:51:08 GMT (7975kb,D)

Link back to: arXiv, form interface, contact.