Reproduce “Heterogeneous Federated Learning with Convolutional and 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
Yingchao Yu, Yuping Yan, Jisong Cai, Yaochu Jin
A study by researchers from Westlake University and collaborating institutions introduces a heterogeneous federated learning framework that successfully integrates Convolutional Neural Networks (CNNs) and Spiking Neural Networks (SNNs). This framework demonstrates convergence and outperforms one-way model transfer methods, while also identifying a 'competitive suppression' phenomenon between modalities that is mitigated by non-IID data.
Key insights
- Federated learning can successfully integrate fundamentally different neural network architectures like CNNs and SNNs, demonstrating the feasibility of heterogeneous model aggregation.
- A notable 'competitive suppression' phenomenon can occur in heterogeneous federated learning, where the performance of one model modality may be hindered by another during aggregation.
- Non-IID data distribution can surprisingly help mitigate the competitive suppression problem between different modalities, and increasing client participation or upload frequency also reduces this discrepancy.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.