Information-theoretic generalization bounds for black-box learning algorithms.
Hrayr HarutyunyanMaxim RaginskyGreg Ver SteegAram GalstyanPublished in: CoRR (2021)
Keyphrases
- information theoretic
- black box
- generalization bounds
- generalization ability
- learning algorithm
- algorithmic stability
- learning problems
- mutual information
- information theory
- data dependent
- learning theory
- kernel machines
- learning machines
- vc dimension
- machine learning algorithms
- ranking algorithm
- machine learning
- generalization error
- model selection
- learning tasks
- linear classifiers
- sample complexity
- statistical learning theory
- test cases
- kernel methods
- support vector machine
- feature selection
- data sets
- supervised learning
- multi task
- active learning
- image processing
- ranking functions
- support vector machine svm
- pairwise
- support vector
- reinforcement learning
- training data
- similarity measure
- neural network