Research map / Federated learning and privacy / Communication-efficient FL
SplitFed: When Federated Learning Meets Split Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use parallel client processing with split networks (SFLV1/SFLV2) instead of relay-based SL | SFL · Multiple clients; ResNet18 on HAM10000 and AlexNet on MNIST | training time: four to six times | Helped |
| Use SFL rather than SL | ResNet18 on HAM10000, five clients · HAM10000 | test accuracy: 79.1% -> 79%; 79.1% -> 79.2% | No clear effect |
| Use SFL rather than FL | AlexNet on HAM10000, five clients · HAM10000 | test accuracy: 75 % -> 70.5%; 75 % -> 74.9% | Mixed |
| Use SFL rather than FL | LeNet on FMNIST, five clients · FMNIST | test accuracy: 91.9 % -> 89.6%; 91.9 % -> 90.4% | Hurt |
| Use SFL rather than SL | AlexNet on FMNIST, five clients · FMNIST | test accuracy: 84.7% -> 86%; 84.7% -> 81% | Mixed |
| Use SFL rather than FL | LeNet on CIFAR10, five clients · CIFAR10 | test accuracy: 69.4 % -> 62.6%; 69.4 % -> 63.8% | Hurt |
| Use SFL rather than SL | AlexNet on MNIST, five clients · MNIST | test accuracy: 95.1% -> 96.9%; 95.1% -> 92% | Mixed |
| Use SFL rather than SL | ResNet18 on MNIST, five clients · MNIST | test accuracy: 99.2% -> 99%; 99.2% -> 99.2% | No clear effect |
| Remove server-side FedAvg and update the server model sequentially (SFLV2 rather than SFLV1) (ablation) | SFL · HAM10000 with AlexNet | test accuracy: 70.5% -> 74.9% | Helped |
| Remove server-side FedAvg and update the server model sequentially (SFLV2 rather than SFLV1) (ablation) | SFL · FMNIST with AlexNet | test accuracy: 86% -> 81% | Hurt |
| Remove server-side FedAvg and update the server model sequentially (SFLV2 rather than SFLV1) (ablation) | SFL · MNIST with AlexNet | test accuracy: 96.9% -> 92% | Hurt |
| Increase the PixelDP privacy budget ε' from 0.5 to 1, 2, or 5 (ablation) | SFLV1 with DP client training · MNIST | test accuracy: 40%, 64%, 73%, 77% | Helped |
Closest papers by meaning
- Convergence Analysis of Split Federated Learning on Heterogeneous Data (NeurIPS 2024 · similarity 0.88)
- FSL-SAGE: Accelerating Federated Split Learning via Smashed Activation Gradient Estimation (ICML 2025 · similarity 0.87)
- Little Is Enough: Boosting Privacy by Sharing Only Hard Labels in Federated Semi-Supervised Learning (AAAI 2025 · similarity 0.87)
- PubSub-VFL: Towards Efficient Two-Party Split Learning in Heterogeneous Environments via Publisher/Subscriber Architecture (NeurIPS 2025 · similarity 0.87)
- Clients Collaborate: Flexible Differentially Private Federated Learning with Guaranteed Improvement of Utility-Privacy Trade-off (ICML 2025 · similarity 0.87)
- FedLMT: Tackling System Heterogeneity of Federated Learning via Low-Rank Model Training with Theoretical Guarantees (ICML 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.