Twice regularized MDPs and the equivalence between robustness and regularization.
Esther DermanMatthieu GeistShie MannorPublished in: CoRR (2021)
Keyphrases
- regularization method
- markov decision processes
- risk minimization
- regularization framework
- regularization methods
- trace norm
- reinforcement learning
- half quadratic
- decision diagrams
- state space
- equivalence relationship
- solution path
- regularized least squares
- markov decision process
- mixed norm
- factored mdps
- regularization parameter
- total least squares
- loss function
- image restoration
- least squares
- objective function
- prior information
- tikhonov regularization
- semi markov decision processes