Fast rates for online learning in Linearly Solvable Markov Decision Processes.
Gergely NeuVicenç GómezPublished in: COLT (2017)
Keyphrases
- markov decision processes
- online learning
- state space
- optimal policy
- reinforcement learning
- dynamic programming
- finite state
- special case
- transition matrices
- np complete
- e learning
- decision theoretic planning
- risk sensitive
- reinforcement learning algorithms
- model based reinforcement learning
- finite horizon
- policy iteration
- reachability analysis
- average cost
- planning under uncertainty
- active learning
- factored mdps
- decision processes
- infinite horizon
- np hard
- partially observable
- average reward
- action space
- state and action spaces
- data mining
- machine learning
- heuristic search
- multistage
- multi agent
- optimal solution
- action sets
- computational complexity