PAC-Bayes Compression Bounds So Tight That They Can Explain Generalization.
Sanae LotfiMarc FinziSanyam KapoorAndres PotapczynskiMicah GoldblumAndrew Gordon WilsonPublished in: NeurIPS (2022)
Keyphrases
- pac bayes
- generalization bounds
- risk bounds
- linear classifiers
- data dependent
- lower bound
- upper bound
- data compression
- compression ratio
- learning theory
- image compression
- compression scheme
- compression algorithm
- vc dimension
- generalization ability
- worst case
- model selection
- empirical risk minimization
- statistical learning theory
- large deviations
- ranking algorithm
- kernel function