Horizon-Free Regret for Linear Markov Decision Processes.
Zihan ZhangJason D. LeeYuxin ChenSimon Shaolei DuPublished in: ICLR (2024)
Keyphrases
- markov decision processes
- total reward
- reward function
- finite state
- transition matrices
- optimal policy
- state space
- reinforcement learning
- dynamic programming
- average reward
- decision theoretic planning
- reachability analysis
- planning under uncertainty
- reinforcement learning algorithms
- partially observable
- decision processes
- risk sensitive
- markov decision process
- discount factor
- expected reward
- factored mdps
- policy iteration
- infinite horizon
- average cost
- finite horizon
- np hard
- action sets
- state abstraction
- model based reinforcement learning
- lower bound