Reproduce “HUS-SNN: Homeostatic Unbiased Sparsity for N:M Sparse 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
Xianzhe Meng, Zhenyu Zhang, Zihe Dong, Xueyi Wang, Yiming Guo
HUS addresses the mismatch between sparse spiking activity and dense weight computation by training changing N:M masks and regulating neuron firing rates. It uses gradient-informed connection selection and multiplicative weight updates based on activity. On MNIST, a 2:4 MLP-SNN's inference time fell from 1.10 seconds to 0.58 seconds, and energy from 13.07 joules to 8.54 joules; accuracy is not always preserved.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.