|
gptkbp:instanceOf
|
gptkb:AI_research_project
|
|
gptkbp:conductedBy
|
gptkb:OpenAI
|
|
gptkbp:contributedTo
|
AI safety research
multi-agent systems research
reinforcement learning benchmarks
|
|
gptkbp:defeatedBy
|
semi-professional Dota 2 teams
professional Dota 2 team OG
|
|
gptkbp:demonstrates
|
gptkb:The_International_2018
gptkb:The_International_2019
real-time decision making
AI learning from scratch
AI surpassing human performance in esports
long-term planning in AI
team coordination in AI
|
|
gptkbp:endDate
|
2019
|
|
gptkbp:focusesOn
|
gptkb:Dota_2
|
|
gptkbp:goal
|
advance AI in complex environments
|
|
gptkbp:inspiredBy
|
further AI research in games
|
|
gptkbp:mediaCoverage
|
gptkb:BBC
gptkb:The_New_York_Times
gptkb:The_Verge
gptkb:Wired
gptkb:TechCrunch
gptkb:VentureBeat
|
|
gptkbp:notableAchievement
|
gptkb:OpenAI_Five
|
|
gptkbp:notableEvent
|
gptkb:OpenAI_Five_Finals
gptkb:OpenAI_Five_Arena
gptkb:OpenAI_Five_vs_OG
OpenAI Five vs top human players
|
|
gptkbp:openSource
|
OpenAI Five code (partially)
|
|
gptkbp:publishedIn
|
gptkb:arXiv
|
|
gptkbp:relatedTo
|
gptkb:OpenAI_Gym
gptkb:OpenAI_Five
|
|
gptkbp:resultedIn
|
gptkb:OpenAI_Five_Arena
public matches against human players
|
|
gptkbp:startYear
|
2017
|
|
gptkbp:usedFor
|
gptkb:Proximal_Policy_Optimization_(PPO)
cloud computing
self-play
LSTM neural networks
hundreds of GPUs
massive parallel training
tens of thousands of CPUs
|
|
gptkbp:uses
|
deep reinforcement learning
|
|
gptkbp:website
|
https://openai.com/research/openai-five
|
|
gptkbp:bfsParent
|
gptkb:OpenAI,_Inc.
|
|
gptkbp:bfsLayer
|
7
|
|
https://www.w3.org/2000/01/rdf-schema#label
|
OpenAI Dota 2 Research
|