Reproduce “The Robustness of Spiking Neural Networks in Federated Learning with Compression Against Non-omniscient Byzantine Attacks” 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
Manh V. Nguyen, Liang Zhao, Bobin Deng, Shaoen Wu
Researchers at Kennesaw State University demonstrate that integrating Top-κ sparsification into Federated Learning with Spiking Neural Networks (FL-SNNs) improves communication efficiency and significantly enhances robustness against non-omniscient Byzantine attacks, achieving approximately 40% higher accuracy against MinMax attacks compared to uncompressed FL-SNNs.
Key insights
- SNNs in Federated Learning initially show higher intrinsic robustness to Noise and IPM attacks but are more susceptible to MinMax attacks compared to ANNs when no compression is applied.
- Integrating Top-κ sparsification with FL-SNNs leverages their unique gradient statistical properties, which fundamentally transforms their resilience to sophisticated non-omniscient Byzantine attacks.
- This compression method provides a dual benefit, offering both substantial communication efficiency and enhanced security against non-omniscient Byzantine adversaries in FL-SNNs.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.