Comparing the Performance of Stochastic Algorithms in Generating Adversarial Attacks on Image Classifiers
This project aims to compare the performance of three stochastic algorithms in generating adversarial attacks on image classifiers:
- DeepSearch, from the paper Deepsearch: A Simple And Effective Blackbox Attack For Deep Neural Networks by F. Zhang, S. Chowdhury, and M. Christakis
- POBA-GA, from the paper POBA-GA: Perturbation Optimized Black-Box Adversarial Attacks Via Genetic Algorithm by J. Chen, M. Su, S. Shen, H. Xiong, and H. Zheng
- PSO, from the paper Attacking Black-Box Image Classifiers With Particle Swarm Optimization by Q. Zhang, K. Wang, W. Zhang and J. Hu
We originally collaborated using Google Drive and Google Colab. Feel free to check this link (https://drive.google.com/drive/folders/1zniHYkyOdiINbFyRr9wpVaCbK4UZ6rov?usp=sharing) to see our group's shared drive.
This repository contains the following files and directories:
GA_results/ - some saved adversarial attack examples from POBA-GA trials on the same input image with different perturbations
ckpts/ - the saved checkpoint file for POBA-GA (checkpoint is updated every generation in the algorithm)
sample_images/ - a sample image used for initial algorithm testing
Batch_Test_GA.ipynb - POBA-GA implementation
Deep_Search_Final.ipynb - DeepSearch implementation
PSO.ipynb - PSO implementation