Online Learning in Markov Decision Processes with Changing Cost Sequences.
Travis DickAndrás GyörgyCsaba SzepesváriPublished in: ICML (2014)
Keyphrases
- markov decision processes
- online learning
- average cost
- state space
- optimal policy
- finite state
- reinforcement learning
- transition matrices
- hidden markov models
- policy iteration
- factored mdps
- risk sensitive
- e learning
- reinforcement learning algorithms
- active learning
- reachability analysis
- dynamic programming
- finite horizon
- decision processes
- partially observable
- model based reinforcement learning
- semi markov decision processes
- state and action spaces
- markov decision process
- decision theoretic planning
- infinite horizon
- state abstraction
- average reward
- action space
- total cost
- long run
- np hard
- markov chain
- expected cost
- data mining