Research map / Federated learning and privacy / Communication-efficient FL
H-SFP: Hierarchical Federated Learning with Decoupled Split-Model Prototyping
Trung-Dung Tran, Nguyen Ha, Minh-Duong Nguyen, Van-Dinh Nguyen, Kok-Seng Wong
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add contrastive loss (ablation) | HSFL baseline · CIFAR-100 IID | accuracy (%): 35.47 -> 41.20 | Improved |
| add synthetic data (ablation) | HSFL + contrastive loss · CIFAR-100 IID | accuracy (%): 41.20 -> 47.60 | Improved |
| add dimension-wise variance to mean-only statistics (ablation) | H-SFP · CIFAR-100 IID | accuracy (%): 52.35 -> 55.10 | Improved |
| add dimension-wise variance to mean-only statistics (ablation) | H-SFP · ImageNet-1K IID | accuracy (%): 26.8 -> 27.9 | Improved |
| add model averaging to purely generative variant (ablation) | H-SFP · CIFAR-100 IID | accuracy (%): 48.20 -> 55.10 | Improved |
| set contrastive temperature tau 0.5 instead of 0.2 (ablation) | H-SFP · CIFAR-100 IID | accuracy (%): 48.15 -> 55.10 | Improved |
| set contrastive temperature tau 0.5 instead of 2.0 (ablation) | H-SFP · CIFAR-100 IID | accuracy (%): 48.65 -> 55.10 | Improved |
| use less frequent aggregation intervals (C instead of A) (ablation) | H-SFP · CIFAR-10 IID | accuracy (%): 67.44 -> 65.44 | Hurt |
| increase clients from 20 to 200 (data sparsity) | FedAvg · CIFAR-100 IID | accuracy (%): FedAvg 60.50 -> 48.85 | Hurt |
| replace SplitFed with H-SFP | SplitFed · ISIC-2018 | training time (h): 72.4 -> 6.8 | Improved |
| replace FedAvg with H-SFP (A) | FedAvg · CIFAR-10 dir(0.3) | accuracy (%): 12.15 -> 40.15 | Improved |
| replace FedAvg with H-SFP (C) in communication | FedAvg · CIFAR-100 communication | total GB: 353x reduction vs FedAvg; 10.94 GB | Improved |
Closest papers by meaning
- Convergence Analysis of Split Federated Learning on Heterogeneous Data (NeurIPS 2024 · similarity 0.90)
- FSL-SAGE: Accelerating Federated Split Learning via Smashed Activation Gradient Estimation (ICML 2025 · similarity 0.88)
- GAS: Generative Activation-Aided Asynchronous Split Federated Learning (AAAI 2025 · similarity 0.88)
- Low Precision Local Training is Enough for Federated Learning (NeurIPS 2024 · similarity 0.87)
- Towards Straggler-Resilient Split Federated Learning: An Unbalanced Update Approach (NeurIPS 2025 · similarity 0.87)
- Hierarchical Federated Learning with Multi-Timescale Gradient Correction (NeurIPS 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
