[NIPS2017] A Unified Game-Theoretic Approach to Multiagent Reinforcement Learning 笔记

文章目录

• 前言
• Background and Related Work
• • Neural Fictitious Self-Play
• Policy-Space Response Oracles
• • Meta-Strategy Solvers
  • Deep Cognitive Hierarchies
  • • Decoupled Meta-Strategy Solvers
• Experiments
• • Joint Policy Correlation in Independent Reinforcement Learning
  • Learning to Safely Exploit and Indirectly Model Opponents in Leduc Poker