Nearly Minimax Optimal Reinforcement Learning for Linear Mixture Markov Decision Processes.
Dongruo ZhouQuanquan GuCsaba SzepesváriPublished in: COLT (2021)
Keyphrases
- markov decision processes
- reinforcement learning
- dynamic programming
- optimal policy
- average cost
- average reward
- action sets
- state space
- finite state
- reinforcement learning algorithms
- policy iteration
- finite horizon
- planning under uncertainty
- total reward
- state and action spaces
- discounted reward
- action space
- model based reinforcement learning
- transition matrices
- stationary policies
- partially observable
- factored mdps
- infinite horizon
- function approximators
- optimal control
- decision theoretic planning
- markov decision process
- reachability analysis
- state abstraction
- continuous state spaces
- policy iteration algorithm
- approximate dynamic programming
- control policies
- stochastic games
- control problems
- function approximation
- learning algorithm
- model free
- decentralized control
- long run
- fixed point
- least squares
- np hard
- optimal solution
- machine learning