On April 13th, 2019, OpenAI Five became the first AI system to defeat the
world champions at an esports game. The game of Dota 2 presents novel
challenges for AI systems such as long time horizons, imperfect information,
and complex, continuous state-action spaces, all challenges which will become
increasingly central to more capable AI systems. OpenAI Five leveraged existing
reinforcement learning techniques, scaled to learn from batches of
approximately 2 million frames every 2 seconds. We developed a distributed
training system and tools for continual training which allowed us to train
OpenAI Five for 10 months. By defeating the Dota 2 world champion (Team OG),
OpenAI Five demonstrates that self-play reinforcement learning can achieve
superhuman performance on a difficult task.
Description
Dota 2 with Large Scale Deep Reinforcement Learning
%0 Generic
%1 openai2019large
%A OpenAI,
%A :,
%A Berner, Christopher
%A Brockman, Greg
%A Chan, Brooke
%A Cheung, Vicki
%A Dębiak, Przemysław
%A Dennison, Christy
%A Farhi, David
%A Fischer, Quirin
%A Hashme, Shariq
%A Hesse, Chris
%A Józefowicz, Rafal
%A Gray, Scott
%A Olsson, Catherine
%A Pachocki, Jakub
%A Petrov, Michael
%A Pinto, Henrique P. d. O.
%A Raiman, Jonathan
%A Salimans, Tim
%A Schlatter, Jeremy
%A Schneider, Jonas
%A Sidor, Szymon
%A Sutskever, Ilya
%A Tang, Jie
%A Wolski, Filip
%A Zhang, Susan
%D 2019
%K RL
%T Dota 2 with Large Scale Deep Reinforcement Learning
%U http://arxiv.org/abs/1912.06680
%X On April 13th, 2019, OpenAI Five became the first AI system to defeat the
world champions at an esports game. The game of Dota 2 presents novel
challenges for AI systems such as long time horizons, imperfect information,
and complex, continuous state-action spaces, all challenges which will become
increasingly central to more capable AI systems. OpenAI Five leveraged existing
reinforcement learning techniques, scaled to learn from batches of
approximately 2 million frames every 2 seconds. We developed a distributed
training system and tools for continual training which allowed us to train
OpenAI Five for 10 months. By defeating the Dota 2 world champion (Team OG),
OpenAI Five demonstrates that self-play reinforcement learning can achieve
superhuman performance on a difficult task.
@misc{openai2019large,
abstract = {On April 13th, 2019, OpenAI Five became the first AI system to defeat the
world champions at an esports game. The game of Dota 2 presents novel
challenges for AI systems such as long time horizons, imperfect information,
and complex, continuous state-action spaces, all challenges which will become
increasingly central to more capable AI systems. OpenAI Five leveraged existing
reinforcement learning techniques, scaled to learn from batches of
approximately 2 million frames every 2 seconds. We developed a distributed
training system and tools for continual training which allowed us to train
OpenAI Five for 10 months. By defeating the Dota 2 world champion (Team OG),
OpenAI Five demonstrates that self-play reinforcement learning can achieve
superhuman performance on a difficult task.},
added-at = {2023-08-25T05:52:51.000+0200},
author = {OpenAI and : and Berner, Christopher and Brockman, Greg and Chan, Brooke and Cheung, Vicki and Dębiak, Przemysław and Dennison, Christy and Farhi, David and Fischer, Quirin and Hashme, Shariq and Hesse, Chris and Józefowicz, Rafal and Gray, Scott and Olsson, Catherine and Pachocki, Jakub and Petrov, Michael and Pinto, Henrique P. d. O. and Raiman, Jonathan and Salimans, Tim and Schlatter, Jeremy and Schneider, Jonas and Sidor, Szymon and Sutskever, Ilya and Tang, Jie and Wolski, Filip and Zhang, Susan},
biburl = {https://www.bibsonomy.org/bibtex/2b945e7512766e1ca33664248d4596896/desplode},
description = {Dota 2 with Large Scale Deep Reinforcement Learning},
interhash = {d176f9c00b2ce62b3e1d8398e18187bf},
intrahash = {b945e7512766e1ca33664248d4596896},
keywords = {RL},
note = {cite arxiv:1912.06680},
timestamp = {2023-08-25T05:52:51.000+0200},
title = {Dota 2 with Large Scale Deep Reinforcement Learning},
url = {http://arxiv.org/abs/1912.06680},
year = 2019
}