Integral Probability Metrics PAC-Bayes Bounds
Preprint
Abstract
We present a PAC-Bayes-style generalization bound which enables the replacement of the KL-divergence with a variety of Integral Probability Metrics (IPM). We provide instances of this bound with the IPM being the total variation metric and the Wasserstein distance. A notable feature of the obtained bounds is that they naturally interpolate between classical uniform convergence bounds in the worst case (when the prior and posterior are far away from each other), and improved bounds in favorable cases (when the posterior and prior are close). This illustrates the possibility of reinforcing classical generalization bounds with algorithm- and data-dependent components, thus making them more suitable to analyze algorithms that use a large hypothesis space.Keywords
All Related Versions
This publication has 13 references indexed in Scilit:
- Tighter Expected Generalization Error Bounds via Convexity of Information Measures2022
- Meta Dynamic Pricing: Transfer Learning Across ExperimentsManagement Science, 2022
- Concentration inequalities for the empirical distribution of discrete distributions: beyond the method of typesInformation and Inference: A Journal of the IMA, 2019
- Simpler PAC-Bayesian bounds for hostile dataMachine Learning, 2017
- Tighter PAC-Bayes bounds through distribution-dependent priorsTheoretical Computer Science, 2013
- On the empirical estimation of integral probability metricsElectronic Journal of Statistics, 2012
- A tail inequality for quadratic forms of subgaussian random vectorsElectronic Communications in Probability, 2012
- An elementary proof of the triangle inequality for the Wasserstein metricProceedings of the American Mathematical Society, 2007
- Local Rademacher complexitiesThe Annals of Statistics, 2005
- Rademacher Processes and Bounding the Risk of Function Learning2000