Reproduce “SAFL: Structure-Aware Personalized Federated Learning via Client-Specific Clustering and SCSI-Guided Model Pruning” 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

Nan Li, Xiaolu Wang, Xiao Du, Puyu Cai, Ting Wang

SAFL is a federated learning framework that integrates client-specific clustering with a Similar Client Structure Information (SCSI)-guided model pruning mechanism to create personalized and efficient models. The framework achieves superior inference accuracy on non-IID datasets, like 93.29% on CIFAR-10 with a 0.3 pruning rate (20.12% higher than FedAvg), while reducing communication overhead by up to 67%.

Key insights

  • Leveraging structural information from similar clients (SCSI) for guiding model pruning is more effective in non-IID federated settings than relying solely on individual client data.
  • Dynamic client clustering combined with a model recovery mechanism enables adaptive and robust model pruning without sacrificing personalization or incurring permanent damage.
  • Structured pruning (e.g., Network Slimming on BN layers) provides computational efficiency for resource-constrained clients while effectively reducing model size and communication.

Summary by alphaXiv.

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