Defending Adversaries Using Unsupervised Feature Clustering VAE
Abstract
We propose a modified VAE (variational autoencoder) as a denoiser to remove adversarial perturbations for image classification. Vanilla VAE's purpose is to make latent variables approximating normal distribution, which reduces the latent inter-class distance of data points. Our proposed VAE modifies this problem by adding a latent variable cluster. So the VAE can guarantee inter-class distance of latent variables and learn class-wised features. Our Feature Clustering VAE performs better on removing perturbations and reconstructing the image to defend adversarial attacks.
Cite
Text
Zhang and Gao. "Defending Adversaries Using Unsupervised Feature Clustering VAE." ICML 2021 Workshops: AML, 2021.Markdown
[Zhang and Gao. "Defending Adversaries Using Unsupervised Feature Clustering VAE." ICML 2021 Workshops: AML, 2021.](https://mlanthology.org/icmlw/2021/zhang2021icmlw-defending/)BibTeX
@inproceedings{zhang2021icmlw-defending,
title = {{Defending Adversaries Using Unsupervised Feature Clustering VAE}},
author = {Zhang, Cheng and Gao, Pan},
booktitle = {ICML 2021 Workshops: AML},
year = {2021},
url = {https://mlanthology.org/icmlw/2021/zhang2021icmlw-defending/}
}