Horizon-Free and Variance-Dependent Reinforcement Learning for Latent Markov Decision Processes.
Runlong ZhouRuosong WangSimon Shaolei DuPublished in: ICML (2023)
Keyphrases
- markov decision processes
- reinforcement learning
- optimal policy
- reinforcement learning algorithms
- state space
- finite state
- policy iteration
- model based reinforcement learning
- partially observable
- discount factor
- dynamic programming
- function approximation
- markov decision process
- finite horizon
- state and action spaces
- factored mdps
- action space
- transition matrices
- planning under uncertainty
- policy evaluation
- action sets
- decision theoretic planning
- average cost
- average reward
- variance reduction
- decision processes
- state abstraction
- model free
- reachability analysis
- infinite horizon
- continuous state
- latent variables
- semi markov decision processes
- control problems
- stochastic games
- decentralized control
- temporal difference
- reward function
- linear programming