PAC-Bayesian Generalization Bounds for Adversarial Generative Models - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2023

PAC-Bayesian Generalization Bounds for Adversarial Generative Models

Résumé

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.
Fichier principal
Vignette du fichier
PAC_Bayes_OT-15.pdf (3.43 Mo) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-04204203 , version 1 (11-09-2023)

Identifiants

  • HAL Id : hal-04204203 , version 1

Citer

Sokhna Diarra Mbacke, Florence Clerc, Pascal Germain. PAC-Bayesian Generalization Bounds for Adversarial Generative Models. International Conference on Machine Learning, Jul 2023, Honololu, Hawaii, United States. ⟨hal-04204203⟩
17 Consultations
15 Téléchargements

Partager

Gmail Mastodon Facebook X LinkedIn More