Saved in:
Bibliographic Details
Main Authors: Branch, Alexander, Pooladzandi, Omead, Khosraviani, Radin, Bhat, Sunay Gajanan, Jiang, Jeffrey, Pottie, Gregory
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2509.25792
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918151486177280
author Branch, Alexander
Pooladzandi, Omead
Khosraviani, Radin
Bhat, Sunay Gajanan
Jiang, Jeffrey
Pottie, Gregory
author_facet Branch, Alexander
Pooladzandi, Omead
Khosraviani, Radin
Bhat, Sunay Gajanan
Jiang, Jeffrey
Pottie, Gregory
contents We introduce PureVQ-GAN, a defense against data poisoning that forces backdoor triggers through a discrete bottleneck using Vector-Quantized VAE with GAN discriminator. By quantizing poisoned images through a learned codebook, PureVQ-GAN destroys fine-grained trigger patterns while preserving semantic content. A GAN discriminator ensures outputs match the natural image distribution, preventing reconstruction of out-of-distribution perturbations. On CIFAR-10, PureVQ-GAN achieves 0% poison success rate (PSR) against Gradient Matching and Bullseye Polytope attacks, and 1.64% against Narcissus while maintaining 91-95% clean accuracy. Unlike diffusion-based defenses requiring hundreds of iterative refinement steps, PureVQ-GAN is over 50x faster, making it practical for real training pipelines.
format Preprint
id arxiv_https___arxiv_org_abs_2509_25792
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle PUREVQ-GAN: Defending Data Poisoning Attacks through Vector-Quantized Bottlenecks
Branch, Alexander
Pooladzandi, Omead
Khosraviani, Radin
Bhat, Sunay Gajanan
Jiang, Jeffrey
Pottie, Gregory
Artificial Intelligence
Computer Vision and Pattern Recognition
We introduce PureVQ-GAN, a defense against data poisoning that forces backdoor triggers through a discrete bottleneck using Vector-Quantized VAE with GAN discriminator. By quantizing poisoned images through a learned codebook, PureVQ-GAN destroys fine-grained trigger patterns while preserving semantic content. A GAN discriminator ensures outputs match the natural image distribution, preventing reconstruction of out-of-distribution perturbations. On CIFAR-10, PureVQ-GAN achieves 0% poison success rate (PSR) against Gradient Matching and Bullseye Polytope attacks, and 1.64% against Narcissus while maintaining 91-95% clean accuracy. Unlike diffusion-based defenses requiring hundreds of iterative refinement steps, PureVQ-GAN is over 50x faster, making it practical for real training pipelines.
title PUREVQ-GAN: Defending Data Poisoning Attacks through Vector-Quantized Bottlenecks
topic Artificial Intelligence
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2509.25792