GSE

This repository contains the code for the paper "Group-wise Sparse and Explainable Adversarial Attacks" (https://arxiv.org/abs/2311.17434).

Setup

Dependencies: numpy, torch, torchvision, natsort, pandas, matplotlib, skimage

The NIPS2017 data set can be found at https://www.kaggle.com/competitions/nips-2017-defense-against-adversarial-attack/data.

Run main

The main.py file contains the code for the main experiments. It can split the data set into chunks for 'embarrassingly parallel' execution. For example to run a targeted test for GSE and a ResNet20 on images 1000-1999 of the CIFAR10 test set with a batch size of 500, execute python main.py --dataset 'CIFAR10' --model 'ResNet20' --numchunks 10 --chunk 1 --batchsize 500 --attack 'GSE' --targeted 1

When all experiments are finished, execute process_results.py to combine the results corresponding to the same experiment.