English

PAC-Bayesian Generalization Bounds for Adversarial Generative Models

Machine Learning 2023-11-15 v4 Artificial Intelligence Machine Learning

Abstract

We extend PAC-Bayesian theory to generative models and develop generalization bounds for models based on the Wasserstein distance and the total variation distance. Our first result on the Wasserstein distance assumes the instance space is bounded, while our second result takes advantage of dimensionality reduction. Our results naturally apply to Wasserstein GANs and Energy-Based GANs, and our bounds provide new training objectives for these two. Although our work is mainly theoretical, we perform numerical experiments showing non-vacuous generalization bounds for Wasserstein GANs on synthetic datasets.

Keywords

Cite

@article{arxiv.2302.08942,
  title  = {PAC-Bayesian Generalization Bounds for Adversarial Generative Models},
  author = {Sokhna Diarra Mbacke and Florence Clerc and Pascal Germain},
  journal= {arXiv preprint arXiv:2302.08942},
  year   = {2023}
}

Comments

Published at ICML 2023

R2 v1 2026-06-28T08:42:51.466Z