English

Diagnosing Vulnerability of Variational Auto-Encoders to Adversarial Attacks

Cryptography and Security 2021-05-07 v3 Machine Learning Machine Learning

Abstract

In this work, we explore adversarial attacks on the Variational Autoencoders (VAE). We show how to modify data point to obtain a prescribed latent code (supervised attack) or just get a drastically different code (unsupervised attack). We examine the influence of model modifications (β\beta-VAE, NVAE) on the robustness of VAEs and suggest metrics to quantify it.

Keywords

Cite

@article{arxiv.2103.06701,
  title  = {Diagnosing Vulnerability of Variational Auto-Encoders to Adversarial Attacks},
  author = {Anna Kuzina and Max Welling and Jakub M. Tomczak},
  journal= {arXiv preprint arXiv:2103.06701},
  year   = {2021}
}