Unverified90% confidenceFactTime unknown
AlphaStar在2019年在《星际争霸II》欧服达到大师级水平（前0.15%），通过模仿学习与多智能体强化学习的结合实现
1
Sources
90%
Confidence
Long-term
Relevance
6/1/2026
First Seen
Sources
DeepMind联手EVE Online：为何选择这款20年老游戏做AI研究？
twitterdemishassabis
Related Entities
Related Claims
UnverifiedAlphaStar采用多智能体自我博弈和联盟训练机制，其训练量相当于人类玩家200年的游戏经验75% similarUnverifiedThe 2019 AlphaStar faced imperfect information, real-time decision-making, and an action space of up to 10^26, and its league training mechanism made the total training volume equivalent to 200 years of human gameplay experience.74% similarUnverifiedAlphaGo's defeat of the human Go champion is a landmark application of reinforcement learning.73% similarUnverifiedThe self-play mechanism of reinforcement learning for code tasks draws on the training paradigm of AlphaGo.60% similarUnverifiedThe AlphaZero reproduction used JAX and Flax to build a reinforcement learning training pipeline, conducting self-play training from scratch via multiple TPU Pods.57% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/26555API
curl https://kongchang.com/api/v1/knowledge/claims/26555MCP
get_claim(id=26555)