Fast rates for online learning in Linearly Solvable Markov Decision Processes.
Gergely NeuVicenç GómezPublished in: CoRR (2017)
Keyphrases
- markov decision processes
- online learning
- optimal policy
- state space
- finite state
- decision theoretic planning
- dynamic programming
- reinforcement learning
- e learning
- model based reinforcement learning
- special case
- np complete
- transition matrices
- action space
- average cost
- active learning
- reinforcement learning algorithms
- policy iteration
- finite horizon
- risk sensitive
- partially observable
- computational complexity
- reachability analysis
- np hard
- planning under uncertainty
- average reward
- action sets
- markov decision process
- decision processes
- semi markov decision processes
- probabilistic planning
- stationary policies
- linear programming
- factored mdps
- discounted reward
- partially observable markov decision processes
- decision problems