Probably Approximately Correct (PAC) Exploration in Reinforcement Learning.
Alexander L. StrehlPublished in: ISAIM (2008)
Keyphrases
- pac learning
- reinforcement learning
- sample complexity
- learning problems
- active exploration
- exploration strategy
- exploration exploitation
- reinforcement learning algorithms
- learning algorithm
- active learning
- uniform distribution
- concept class
- action selection
- vc dimension
- supervised learning
- computational learning theory
- upper bound
- pac model
- random sampling
- function approximation
- model based reinforcement learning
- learning theory
- markov decision processes
- concept classes
- sample size
- theoretical analysis
- model free
- target concept
- mistake bound
- machine learning
- lower bound
- optimal policy
- balancing exploration and exploitation
- exploration exploitation tradeoff
- multi agent
- pac learning model
- sample complexity bounds
- state space
- temporal difference
- generalization error
- membership queries
- autonomous learning
- dynamic programming
- learning tasks
- training examples
- classification noise
- boolean functions
- kernel methods
- machine learning algorithms