Reproduce “Adversarially Robust Spiking Neural Networks with Sparse Connectivity” 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

Mathias Schmolli, Maximilian Baronig, Robert Legenstein, Ozan Özdenizci

The method trains and prunes an adversarially robust artificial neural network, transfers its weights and sparse connectivity to a spiking neural network, then adversarially fine-tunes it. On CIFAR-10 with VGG-16 at 90% sparsity, it achieves 89.5% clean accuracy and 26.1% accuracy under a 10-step PGD ensemble attack at ε=8/255, versus 9.7% robust accuracy for end-to-end training with the same connectivity.

Summary by alphaXiv.

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