Diverse Policies Converge in Reward-Free Markov Decision Processes.
Fanqi LinShiyu HuangWei-Wei TuPublished in: PRICAI (1) (2022)
Keyphrases
- markov decision processes
- reward function
- expected reward
- total reward
- optimal policy
- discounted reward
- average reward
- stationary policies
- reinforcement learning
- reinforcement learning algorithms
- state space
- dynamic programming
- finite state
- markov decision process
- long run
- finite horizon
- policy iteration
- partially observable
- decision processes
- transition matrices
- hierarchical reinforcement learning
- decision theoretic planning
- average cost
- action sets
- model based reinforcement learning
- decentralized control
- infinite horizon
- stochastic games
- factored mdps
- planning under uncertainty
- partially observable markov decision processes
- state and action spaces
- reachability analysis
- semi markov decision processes
- multistage
- policy iteration algorithm
- optimality criterion
- control policies
- macro actions
- decision problems
- dynamical systems
- stochastic shortest path