A Gauss-Newton Method for Markov Decision Processes.
Thomas FurmstonGuy LeverPublished in: CoRR (2015)
Keyphrases
- markov decision processes
- newton method
- convergence analysis
- linear equations
- regularized least squares
- state space
- variational inequalities
- optimality conditions
- quadratic programming
- reinforcement learning
- optimal policy
- linear svm
- dynamic programming
- reinforcement learning algorithms
- global convergence
- markov decision process
- nonnegative matrix factorization
- reward function
- action space
- sensitivity analysis
- sufficient conditions