Reproduce “Protecting the Neural Networks against FGSM Attack Using Machine Unlearning” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Amir Hossein Khorasani, Ali Jahanian, Maryam Rastgarpour

The authors propose a workflow to improve LeNet accuracy on adversarial images by generating FGSM examples, selecting high-loss cases, removing them from training data, and retraining. In an MNIST experiment, FGSM with ε = 0.1 reduced clean-test accuracy from 99.32% to 7.96% on attacked images; the updated model reached 96.69% on attacked images, but its clean-test accuracy was not reported.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.