%0 Generic %1 parRL %A Nair, Arun %A Srinivasan, Praveen %A Blackwell, Sam %A Alcicek, Cagdas %A Fearon, Rory %A De Maria, Alessandro %A Panneershelvam, Vedavyas %A Suleyman, Mustafa %A Beattie, Charles %A Petersen, Stig %A Legg, Shane %A Mnih, Volodymyr %A Kavukcuoglu, Koray %A Silver, David %D 2015 %K thesis %T Massively Parallel Methods for Deep Reinforcement Learning %U http://arxiv.org/abs/1507.04296 %X We present the first massively distributed architecture for deep reinforcement learning. This architecture uses four main components: parallel actors that generate new behaviour; parallel learners that are trained from stored experience; a distributed neural network to represent the value function or behaviour policy; and a distributed store of experience. We used our architecture to implement the Deep Q-Network algorithm (DQN). Our distributed algorithm was applied to 49 games from Atari 2600 games from the Arcade Learning Environment, using identical hyperparameters. Our performance surpassed non-distributed DQN in 41 of the 49 games and also reduced the wall-time required to achieve these results by an order of magnitude on most games.

@misc{parRL, abstract = {We present the first massively distributed architecture for deep reinforcement learning. This architecture uses four main components: parallel actors that generate new behaviour; parallel learners that are trained from stored experience; a distributed neural network to represent the value function or behaviour policy; and a distributed store of experience. We used our architecture to implement the Deep Q-Network algorithm (DQN). Our distributed algorithm was applied to 49 games from Atari 2600 games from the Arcade Learning Environment, using identical hyperparameters. Our performance surpassed non-distributed DQN in 41 of the 49 games and also reduced the wall-time required to achieve these results by an order of magnitude on most games.}, added-at = {2020-02-06T11:17:11.000+0100}, author = {Nair, Arun and Srinivasan, Praveen and Blackwell, Sam and Alcicek, Cagdas and Fearon, Rory and De Maria, Alessandro and Panneershelvam, Vedavyas and Suleyman, Mustafa and Beattie, Charles and Petersen, Stig and Legg, Shane and Mnih, Volodymyr and Kavukcuoglu, Koray and Silver, David}, biburl = {https://www.bibsonomy.org/bibtex/277bc9a757483e0aa656cde0ef873ef40/ross_mck}, description = {[1507.04296] Massively Parallel Methods for Deep Reinforcement Learning}, interhash = {b5c30e0d79c5eae46ca695518f58a614}, intrahash = {77bc9a757483e0aa656cde0ef873ef40}, keywords = {thesis}, note = {cite arxiv:1507.04296Comment: Presented at the Deep Learning Workshop, International Conference on Machine Learning, Lille, France, 2015}, timestamp = {2020-02-06T11:17:11.000+0100}, title = {Massively Parallel Methods for Deep Reinforcement Learning}, url = {http://arxiv.org/abs/1507.04296}, year = 2015 }