阿里巴巴认知计算实验室:星际争霸与人工智能


星际争霸与人工智能
阿里巴巴认知计算实验室龙海涛

 


Challenge Problems for Artificial Intelligence

 

  • Imperfect Information
  • Huge State and Action Space
  • Adversarial Real-time Strategy
  • Long-Term Planning
  • Temporal and Spatial Reasoning
  • Multiagent Cooperation

 


StarCraft AI Research and Competitions

 


Reinforcement Learning

 


Deep Reinforcement Learning

 


What is next?

 

•All above are single AI agent
•But, true human intelligence embraces social and collective wisdom
•How large-scale multiple AI agents could learn human-level collaborations (or competitions) from their experiences?

 


Cooperative Hunting

 


Artificial Collective Intelligence

 

 


Multiagent Bidirectionally-Coordinated Net (BiCNet)

 


Neuroscience Hypothesis

 


Architecture Overview

 


Coordinated Moves without Collision

 


Hit and Run Tactics

 


Coordinated Cover Attack

 


Focus Fire without Overkill

 


Coordinated Heterogonous Agents

 


Hierarchical Reinforcement Learning

 


Imitation Learning

 


Continual Learning

 


Memory-Augmented Neural Networks

 

推荐文章

沪公网安备 31010702002009号