Reproduce “Robust Stable Spiking Neural Networks” 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
Jianhao Ding, Zhiyu Pan, Yujia Liu, Zhaofei Yu, Tiejun Huang
The authors introduce a training penalty for spiking neural networks that measures membrane-potential perturbations over time. They simplify neuron reset dynamics to define this continuous proxy and prove an input-output stability bound for the simplified dynamics. On CIFAR-10, VGG11 DLIF adversarial training raises PGD-10 accuracy from 29.06% to 37.55%, while clean accuracy falls from 90.07% to 87.21% in tested setup.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.