Research map / Federated learning and privacy / Personalized and heterogeneous FL
Understanding How Consistency Works in Federated Learning via Stage-wise Relaxed Initialization
Yan Sun, Li Shen, Dacheng Tao
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 relaxed initialization (beta=0.1) to FedAvg, i.e. FedInit | FedAvg · CIFAR-10 ResNet-18-GN Dir-0.1 10%-100 | test accuracy (%): 72.53 -> 75.95 | Improved |
| reduce participation from 10% to 5% clients | FedInit · CIFAR-10 ResNet-18-GN Dir-0.1 | test accuracy (%): 75.95 -> 74.92 | No clear effect |
| reduce participation from 10% to 5% clients | FedAvg · CIFAR-10 ResNet-18-GN Dir-0.1 | test accuracy (%): 72.53 -> 70.65 | Hurt |
| add relaxed initialization to FedAdam (ablation) | FedAdam · CIFAR-10 ResNet-18-GN 10%-100 Dir-0.6 | test accuracy (%): 76.52 -> 78.33 | Improved |
| add relaxed initialization to FedSAM (ablation) | FedSAM · CIFAR-10 ResNet-18-GN 10%-100 Dir-0.6 | test accuracy (%): 79.23 -> 83.36 | Improved |
| add relaxed initialization to SCAFFOLD (ablation) | SCAFFOLD · CIFAR-10 ResNet-18-GN 10%-100 Dir-0.1 | test accuracy (%): 75.06 -> 77.30 | Improved |
| add relaxed initialization to FedDyn (ablation) | FedDyn · CIFAR-10 ResNet-18-GN 10%-100 Dir-0.6 | test accuracy (%): 82.43 -> 81.91 | No clear effect |
| add relaxed initialization to FedCM (ablation) | FedCM · CIFAR-10 ResNet-18-GN 10%-100 Dir-0.1 | test accuracy (%): 73.93 -> 73.71 | No clear effect |
| negative beta (-0.2) instead of beta=0 (ablation) | FedInit · CIFAR-10 ResNet-18-GN Dir-0.1 10%-100 | test accuracy (%): 72.53 -> 64.70 | Hurt |
| beta 0.1 instead of 0 (ablation) | FedInit · CIFAR-10 ResNet-18-GN Dir-0.1 10%-100 | test accuracy (%): 72.53 -> 75.95 | Improved |
| beta 0.15 (too large) instead of 0.1 (ablation) | FedInit · CIFAR-10 ResNet-18-GN Dir-0.1 10%-100 | test accuracy (%): 75.95 -> 44.47 | Hurt |
| increase local interval K (ablation) | FedInit · CIFAR-10 | test accuracy (%) | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SCAFFOLD — SCAFFOLD: Stochastic Controlled Averaging for Federated Learning (ICML 2020) · better on 8 of 8 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- FedImpro: Measuring and Improving Client Update in Federated Learning (ICLR 2024 · similarity 0.91)
- Federated Learning with Sample-level Client Drift Mitigation (AAAI 2025 · similarity 0.91)
- Exact and Linear Convergence for Federated Learning under Arbitrary Client Participation is Attainable (NeurIPS 2025 · similarity 0.89)
- FedAS: Bridging Inconsistency in Personalized Federated Learning (CVPR 2024 · similarity 0.89)
- Tackling Intertwined Data and Device Heterogeneities in Federated Learning with Unlimited Staleness (AAAI 2025 · similarity 0.89)
- FLHetBench: Benchmarking Device and State Heterogeneity in Federated Learning (CVPR 2024 · similarity 0.89)
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.
