Reproduce “Robust Representation Learning for Spiking Neural Networks based on Conditional Information Bottleneck” 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

Yuli Zha, Wang Fan, Yixiao Dai, Lingjie Wang, Bang Hu, Wei Cao, Fengzhe Zhang, Fan Zhang

SCIB improves classification robustness in spiking neural networks by penalizing input-representation information that remains after conditioning on the class label. It combines cross-entropy with a KL-divergence penalty to a learnable class-conditional prior. On CIFAR-10, with four timesteps and a 4/255 PGD attack, accuracy is 47.58%, versus 21.33% for FEEL-SNN; SCIB also estimates a 24.9% lower inference-energy proxy, not measured device energy.

Summary by alphaXiv.

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