# Adversarial Defense by Suppressing High-frequency Components

Source: [Adversarial Defense by Suppressing High-frequency Components](<https://www.sota2.com/research/paper/68fa66541e209131cea0f842>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2025-10-25T21:47:30.843Z

Authors: Zhendong Zhang, Cheolkon Jung, Xiaolong Liang

Published: 2019-08-19T02:43:35.000Z

## Abstract

Recent works show that deep neural networks trained on image classification dataset bias towards textures. Those models are easily fooled by applying small high-frequency perturbations to clean images. In this paper, we learn robust image classification models by removing high-frequency components. Specifically, we develop a differentiable high-frequency suppression module based on discrete Fourier transform (DFT). Combining with adversarial training, we won the 5th place in the IJCAI-2019 Alibaba Adversarial AI Challenge. Our code is available online.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/1908.06566>)

[PDF](<https://arxiv.org/pdf/1908.06566.pdf>)