Markov Abstractions for PAC Reinforcement Learning in Non-Markov Decision Processes.
Alessandro RoncaGabriel Paludo LicksGiuseppe De GiacomoPublished in: CoRR (2022)
Keyphrases
- markov decision processes
- reinforcement learning
- reinforcement learning algorithms
- state space
- optimal policy
- policy iteration
- finite state
- markov chain
- dynamic programming
- partially observable
- reachability analysis
- transition matrices
- markov model
- action space
- state and action spaces
- average reward
- model based reinforcement learning
- markov decision process
- decision theoretic planning
- average cost
- function approximation
- finite horizon
- factored mdps
- control problems
- state abstraction
- action sets
- reward function
- model free
- policy evaluation
- planning under uncertainty
- learning algorithm
- optimal control
- markov decision problems
- multi agent
- supervised learning
- infinite horizon
- stochastic games
- sample size
- action selection
- learning tasks
- lot sizing
- machine learning