A Tight Excess Risk Bound via a Unified PAC-Bayesian-Rademacher-Shtarkov-MDL Complexity.
Peter D. GrünwaldNishant A. MehtaPublished in: CoRR (2017)
Keyphrases
- pac bayesian
- generalization bounds
- worst case
- data dependent
- rademacher complexity
- model selection
- distribution free
- upper bound
- error bounds
- lower bound
- vc dimension
- risk bounds
- learning theory
- generalization error bounds
- generalization ability
- function classes
- large deviations
- ranking algorithm
- empirical risk minimization
- sample size
- generalization error
- loss function
- linear classifiers
- statistical learning theory
- learning problems
- uniform convergence
- empirical risk
- reproducing kernel hilbert space
- normal distribution
- np hard
- computational complexity