Model-Based Reinforcement Learning Is Minimax-Optimal for Offline Zero-Sum Markov Games.
Yuling YanGen LiYuxin ChenJianqing FanPublished in: CoRR (2022)
Keyphrases
- markov games
- markov decision processes
- model based reinforcement learning
- dynamic programming
- average cost
- average reward
- reinforcement learning algorithms
- multiagent reinforcement learning
- finite state
- reinforcement learning
- worst case
- state space
- optimal policy
- policy iteration
- stochastic games
- markov decision process
- decision processes
- optimal solution
- infinite horizon
- optimal control
- supervised learning
- hidden markov models