Agnostic Reinforcement Learning with Low-Rank MDPs and Rich Observations.
Christoph DannYishay MansourMehryar MohriAyush SekhariKarthik SridharanPublished in: CoRR (2021)
Keyphrases
- low rank
- reinforcement learning
- markov decision processes
- convex optimization
- missing data
- state space
- singular value decomposition
- matrix factorization
- linear combination
- low rank matrix
- matrix completion
- optimal policy
- high dimensional data
- function approximation
- semi supervised
- high order
- rank minimization
- reinforcement learning algorithms
- matrix decomposition
- policy iteration
- partially observable
- model free
- markov decision process
- kernel matrix
- learning algorithm
- machine learning
- learning process
- function approximators
- robust principal component analysis
- low rank matrices
- image processing
- minimization problems
- temporal difference
- action space
- dynamic programming
- reward function
- singular values
- learning problems
- markov decision problems
- higher order
- trace norm
- transfer learning
- input image
- denoising
- data mining