A Minimum Relative Entropy Controller for Undiscounted Markov Decision Processes
Pedro A. OrtegaDaniel A. BraunPublished in: CoRR (2010)
Keyphrases
- markov decision processes
- relative entropy
- policy iteration
- information theoretic
- information theory
- optimal policy
- state space
- mutual information
- finite state
- log likelihood
- infinite horizon
- average reward
- reinforcement learning
- dynamic programming
- covariance matrix
- stochastic games
- partially observable
- mahalanobis distance
- optimal control
- reinforcement learning algorithms
- maximum entropy
- average cost
- bregman divergences
- reward function
- action space
- markov decision process
- partially observable markov decision processes
- control strategy
- markov decision problems