Reinforcement Learning through Asynchronous Advantage Actor-Critic on a GPU

Mohammad Babaeizadeh, Iuri Frosio, Stephen Tyree, Jason Clemons, Jan Kautz
11/18/2016
cs.LG

Official code: NVlabs/GA3C (661 stars, BSD-3-Clause, last commit 2020-02-25).

Abstract

We introduce a hybrid CPU/GPU version of the Asynchronous Advantage Actor-Critic (A3C) algorithm, currently the state-of-the-art method in reinforcement learning for various gaming tasks. We analyze its computational traits and concentrate on aspects critical to leveraging the GPU's computational power. We introduce a system of queues and a dynamic scheduling strategy, potentially helpful for other asynchronous algorithms as well. Our hybrid CPU/GPU version of A3C, based on TensorFlow, achieves a significant speed up compared to a CPU implementation; we make it publicly available to other researchers at https://github.com/NVlabs/GA3C .

View on arXivView PDF

Code Implementations(1)

NVlabs/GA3COfficial100%
BSD-3-Clause

Hybrid CPU/GPU implementation of the A3C algorithm for deep reinforcement learning.

661193Shell, PythonNov 19, 20166 years agoBSD-3-Clause

Similar Papers

ELF: An Extensive, Lightweight and Flexible Research Platform for Real-time Strategy Games

84% match1 repo2.1k2017

Dominion: A New Frontier for AI Research

83% match2024

Toward Computationally Efficient Inverse Reinforcement Learning via Reward Shaping

82% match2023

Soft Actor-Critic for Discrete Action Settings

82% match1 repo5.9k2019

Advances in Preference-based Reinforcement Learning: A Review

82% match2024

Subgoal-based Hierarchical Reinforcement Learning for Multi-Agent Collaboration

82% match1 repo102024

Cite this paper

@article{babaeizadeh2016reinforcement,
  title  = {Reinforcement Learning through Asynchronous Advantage Actor-Critic on a GPU},
  author = {Mohammad Babaeizadeh and Iuri Frosio and Stephen Tyree and Jason Clemons and Jan Kautz},
  year   = {2016},
  eprint = {1611.06256},
  archivePrefix = {arXiv},
  url    = {https://arxiv.org/abs/1611.06256}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/1611.06256.svg)](https://codeofpaper.com/paper/1611.06256)

Discussion