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


Current browse context:


Change to browse by:


References & Citations

DBLP - CS Bibliography


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

Computer Science > Machine Learning

Title: ACERAC: Efficient reinforcement learning in fine time discretization

Abstract: One of the main goals of reinforcement learning (RL) is to provide a~way for physical machines to learn optimal behavior instead of being programmed. However, effective control of the machines usually requires fine time discretization. The most common RL methods apply independent random elements to each action, which is not suitable in that setting. It is not feasible because it causes the controlled system to jerk, and does not ensure sufficient exploration since a~single action is not long enough to create a~significant experience that could be translated into policy improvement. In our view these are the main obstacles that prevent application of RL in contemporary control systems. To address these pitfalls, in this paper we introduce an RL framework and adequate analytical tools for actions that may be stochastically dependent in subsequent time instances. We also introduce an RL algorithm that approximately optimizes a~policy that produces such actions. It applies experience replay to adjust likelihood of sequences of previous actions to optimize expected $n$-step returns the policy yields. The efficiency of this algorithm is verified against four other RL methods (CDAU, PPO, SAC, ACER) in four simulated learning control problems (Ant, HalfCheetah, Hopper, and Walker2D) in diverse time discretization. The algorithm introduced here outperforms the competitors in most cases considered.
Comments: Submitted to IEEE Transactions on Neural Networks and Learning Systems. arXiv admin note: text overlap with arXiv:2009.04777
Subjects: Machine Learning (cs.LG)
ACM classes: I.2.6
Cite as: arXiv:2104.04004 [cs.LG]
  (or arXiv:2104.04004v4 [cs.LG] for this version)

Submission history

From: Paweł Wawrzyński [view email]
[v1] Thu, 8 Apr 2021 18:40:20 GMT (1087kb,D)
[v2] Thu, 25 Nov 2021 21:13:15 GMT (1205kb,D)
[v3] Mon, 29 Nov 2021 08:40:46 GMT (1205kb,D)
[v4] Mon, 11 Jul 2022 10:41:39 GMT (2807kb,D)

Link back to: arXiv, form interface, contact.