Projection by Convolution: Optimal Sample Complexity for Reinforcement Learning in Continuous-Space MDPs.
Davide MaranAlberto Maria MetelliMatteo PapiniMarcello RestelliPublished in: COLT (2024)
Keyphrases
- reinforcement learning
- sample complexity
- continuous space
- sequential decision problems
- learning problems
- markov decision processes
- learning algorithm
- supervised learning
- dynamic programming
- state space
- optimal policy
- theoretical analysis
- active learning
- lower bound
- optimal control
- discrete space
- generalization error
- machine learning
- upper bound
- sample size
- special case
- mathematical morphology
- optimal solution
- training examples
- worst case
- similarity measure
- partially observable
- image processing
- labeled data
- learning tasks
- computational complexity
- reward function
- decision trees
- data sets