No-Regret Reinforcement Learning with Value Function Approximation: a Kernel Embedding Approach.
Sayak Ray ChowdhuryRafael OliveiraPublished in: CoRR (2020)
Keyphrases
- reinforcement learning
- temporal difference
- state space
- temporal difference learning
- approximate dynamic programming
- state action
- laplacian eigenmaps
- function approximation
- reinforcement learning algorithms
- reward function
- kernel function
- total reward
- online learning
- function approximators
- evaluation function
- model free
- kernel methods
- loss function
- lower bound
- learning algorithm
- graph embedding
- kernel based learning
- machine learning
- feature space
- basis functions
- markov decision processes
- binary classification
- support vector
- nonlinear dimensionality reduction
- control problems
- step size
- optimal policy
- markov decision process
- action space
- action selection
- dynamic programming
- kernel matrix
- multi agent
- vector space
- policy iteration
- learning problems
- kernel pca
- worst case
- input space
- bandit problems
- confidence bounds
- reward signal