Nearly Minimax Optimal Reinforcement Learning for Linear Mixture Markov Decision Processes.
Dongruo ZhouQuanquan GuCsaba SzepesváriPublished in: CoRR (2020)
Keyphrases
- markov decision processes
- reinforcement learning
- dynamic programming
- average cost
- optimal policy
- average reward
- state space
- action sets
- reinforcement learning algorithms
- finite horizon
- policy iteration
- finite state
- total reward
- discounted reward
- state and action spaces
- stationary policies
- partially observable
- transition matrices
- model based reinforcement learning
- reachability analysis
- state abstraction
- reward function
- markov decision process
- optimal control
- approximate dynamic programming
- policy iteration algorithm
- factored mdps
- control policies
- function approximation
- planning under uncertainty
- action space
- decision theoretic planning
- infinite horizon
- optimality criterion
- policy evaluation
- stochastic games
- objective function
- markov decision problems
- optimal solution
- decentralized control
- monte carlo
- machine learning
- partially observable markov decision processes
- np hard
- multistage
- continuous state
- long run