Near Optimal Exploration-Exploitation in Non-Communicating Markov Decision Processes.
Ronan FruitMatteo PirottaAlessandro LazaricPublished in: NeurIPS (2018)
Keyphrases
- markov decision processes
- exploration exploitation
- reinforcement learning
- active learning
- optimal policy
- state space
- bandit problems
- finite state
- transition matrices
- policy iteration
- function approximation
- action space
- action sets
- planning under uncertainty
- decision problems
- dynamic programming
- partially observable
- decision theoretic planning
- infinite horizon
- transfer learning
- machine learning
- temporal difference
- markov decision process
- average reward
- model based reinforcement learning
- relevance feedback
- reward function
- average cost
- multiple features
- reachability analysis
- supervised learning
- model free
- finite horizon
- real time dynamic programming
- state and action spaces
- markov decision problems
- optimal control
- graphical models
- learning algorithm