Asynchronous Methods for Deep Reinforcement Learning

Volodymyr Mnih, Adria Puigdomenech Badia, Mehdi Mirza, Alex Graves, Timothy Lillicrap, Tim Harley, David Silver, Koray Kavukcuoglu
Proceedings of The 33rd International Conference on Machine Learning, PMLR 48:1928-1937, 2016.

Abstract

We propose a conceptually simple and lightweight framework for deep reinforcement learning that uses asynchronous gradient descent for optimization of deep neural network controllers. We present asynchronous variants of four standard reinforcement learning algorithms and show that parallel actor-learners have a stabilizing effect on training allowing all four methods to successfully train neural network controllers. The best performing method, an asynchronous variant of actor-critic, surpasses the current state-of-the-art on the Atari domain while training for half the time on a single multi-core CPU instead of a GPU. Furthermore, we show that asynchronous actor-critic succeeds on a wide variety of continuous motor control problems as well as on a new task of navigating random 3D mazes using a visual input.

Cite this Paper


BibTeX
@InProceedings{pmlr-v48-mniha16, title = {Asynchronous Methods for Deep Reinforcement Learning}, author = {Mnih, Volodymyr and Badia, Adria Puigdomenech and Mirza, Mehdi and Graves, Alex and Lillicrap, Timothy and Harley, Tim and Silver, David and Kavukcuoglu, Koray}, booktitle = {Proceedings of The 33rd International Conference on Machine Learning}, pages = {1928--1937}, year = {2016}, editor = {Balcan, Maria Florina and Weinberger, Kilian Q.}, volume = {48}, series = {Proceedings of Machine Learning Research}, address = {New York, New York, USA}, month = {20--22 Jun}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v48/mniha16.pdf}, url = {http://proceedings.mlr.press/v48/mniha16.html}, abstract = {We propose a conceptually simple and lightweight framework for deep reinforcement learning that uses asynchronous gradient descent for optimization of deep neural network controllers. We present asynchronous variants of four standard reinforcement learning algorithms and show that parallel actor-learners have a stabilizing effect on training allowing all four methods to successfully train neural network controllers. The best performing method, an asynchronous variant of actor-critic, surpasses the current state-of-the-art on the Atari domain while training for half the time on a single multi-core CPU instead of a GPU. Furthermore, we show that asynchronous actor-critic succeeds on a wide variety of continuous motor control problems as well as on a new task of navigating random 3D mazes using a visual input.} }
Endnote
%0 Conference Paper %T Asynchronous Methods for Deep Reinforcement Learning %A Volodymyr Mnih %A Adria Puigdomenech Badia %A Mehdi Mirza %A Alex Graves %A Timothy Lillicrap %A Tim Harley %A David Silver %A Koray Kavukcuoglu %B Proceedings of The 33rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2016 %E Maria Florina Balcan %E Kilian Q. Weinberger %F pmlr-v48-mniha16 %I PMLR %P 1928--1937 %U http://proceedings.mlr.press/v48/mniha16.html %V 48 %X We propose a conceptually simple and lightweight framework for deep reinforcement learning that uses asynchronous gradient descent for optimization of deep neural network controllers. We present asynchronous variants of four standard reinforcement learning algorithms and show that parallel actor-learners have a stabilizing effect on training allowing all four methods to successfully train neural network controllers. The best performing method, an asynchronous variant of actor-critic, surpasses the current state-of-the-art on the Atari domain while training for half the time on a single multi-core CPU instead of a GPU. Furthermore, we show that asynchronous actor-critic succeeds on a wide variety of continuous motor control problems as well as on a new task of navigating random 3D mazes using a visual input.
RIS
TY - CPAPER TI - Asynchronous Methods for Deep Reinforcement Learning AU - Volodymyr Mnih AU - Adria Puigdomenech Badia AU - Mehdi Mirza AU - Alex Graves AU - Timothy Lillicrap AU - Tim Harley AU - David Silver AU - Koray Kavukcuoglu BT - Proceedings of The 33rd International Conference on Machine Learning DA - 2016/06/11 ED - Maria Florina Balcan ED - Kilian Q. Weinberger ID - pmlr-v48-mniha16 PB - PMLR DP - Proceedings of Machine Learning Research VL - 48 SP - 1928 EP - 1937 L1 - http://proceedings.mlr.press/v48/mniha16.pdf UR - http://proceedings.mlr.press/v48/mniha16.html AB - We propose a conceptually simple and lightweight framework for deep reinforcement learning that uses asynchronous gradient descent for optimization of deep neural network controllers. We present asynchronous variants of four standard reinforcement learning algorithms and show that parallel actor-learners have a stabilizing effect on training allowing all four methods to successfully train neural network controllers. The best performing method, an asynchronous variant of actor-critic, surpasses the current state-of-the-art on the Atari domain while training for half the time on a single multi-core CPU instead of a GPU. Furthermore, we show that asynchronous actor-critic succeeds on a wide variety of continuous motor control problems as well as on a new task of navigating random 3D mazes using a visual input. ER -
APA
Mnih, V., Badia, A.P., Mirza, M., Graves, A., Lillicrap, T., Harley, T., Silver, D. & Kavukcuoglu, K.. (2016). Asynchronous Methods for Deep Reinforcement Learning. Proceedings of The 33rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 48:1928-1937 Available from http://proceedings.mlr.press/v48/mniha16.html.

Related Material