Non-vacuous Generalization Bounds at the ImageNet Scale: a PAC-Bayesian Compression Approach.
Wenda ZhouVictor VeitchMorgane AusternRyan P. AdamsPeter OrbanzPublished in: ICLR (Poster) (2019)
Keyphrases
- pac bayesian
- generalization bounds
- linear classifiers
- data dependent
- learning theory
- generalization ability
- model selection
- ranking algorithm
- vc dimension
- statistical learning theory
- large deviations
- learning problems
- data compression
- image compression
- empirical risk minimization
- uniform convergence
- compression scheme
- distribution free
- compression ratio
- ranking functions
- hyperplane
- kernel machines
- svm classifier
- learning machines
- machine learning
- sample size
- ensemble learning
- cross validation
- semi supervised learning
- feature extraction
- learning algorithm