Research map / Federated learning and privacy / Personalized and heterogeneous FL
SphereFed: Hyperspherical Federated Learning
Xin Dong, Sai Qian Zhang, Ang Li, H.T. Kung
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 SphereFed (fixed classifier, feature norm, MSE, FFC) to FedAvg | FedAvg MobileNetV2 · CIFAR-100 (alpha=0.1) | accuracy (%): 63.90 -> 66.52 (up 2.62) | Improved |
| add SphereFed to FedNova | FedNova VGG13 · TinyImageNet | accuracy (%): 39.63 -> 45.21 (up 5.58) | Improved |
| add SphereFed to FedOpt | FedOpt SENet · CIFAR-100 (alpha=0.5) | accuracy (%): 59.60 -> 65.69 (up 6.09) | Improved |
| add CCVR to FedAvg | FedAvg MobileNetV2 · CIFAR-100 (alpha=0.5) | accuracy (%): 68.78 -> 69.14 (up 0.36) | Improved |
| add BABU to FedAvg | FedAvg MobileNetV2 · TinyImageNet | accuracy (%): 29.95 -> 28.38 (down 1.57) | Hurt |
| add fixed random classifier (ablation) | FedAvg MobileNetV2 with CE · CIFAR-100 (alpha=0.5) | accuracy (%): 68.78 -> 69.35 (up 0.57) | Improved |
| fix classifier with orthogonal unit-norm init instead of random (ablation) | FedAvg MobileNetV2 with CE · CIFAR-100 (alpha=0.5) | accuracy (%): 69.35 -> 69.65 | Improved |
| add feature normalization to Fix (OU) (ablation) | FedAvg MobileNetV2 with CE · CIFAR-100 (alpha=0.5) | accuracy (%): 69.65 -> 70.87 | Improved |
| replace CE with MSE loss on the full hyperspherical setup (ablation) | FedAvg + Fix (OU) + Norm · CIFAR-100 (alpha=0.5) | accuracy (%): 70.87 -> 71.85 | Improved |
| add feature normalization alone with MSE loss (no fixed classifier) (ablation) | FedAvg MobileNetV2 with MSE · CIFAR-100 (alpha=0.5) | accuracy (%): 66.42 -> Diverging | Hurt |
| replace CE with MSE without fixed classifier or norm (ablation) | FedAvg MobileNetV2 · CIFAR-100 (alpha=0.5) | accuracy (%): 68.78 -> 66.42 | Hurt |
| add l2 penalty on calibrated classifier weights in FFC (ablation) | SphereFed FFC · CIFAR-100 (alpha=0.5) | accuracy (%): 71.85 -> 71.96 | No clear effect |
Closest papers by meaning
- DynaFed: Tackling Client Data Heterogeneity With Global Dynamics (CVPR 2023 · similarity 0.88)
- Federated Learning with Manifold Regularization and Normalized Update Reaggregation (NeurIPS 2023 · similarity 0.87)
- Federated Dataset Simulation: Inducing Label-Free Heterogeneity Across Tasks (NeurIPS 2026 · similarity 0.87)
- FLUX: Efficient Descriptor-Driven Clustered Federated Learning under Arbitrary Distribution Shifts (NeurIPS 2025 · similarity 0.87)
- SparsyFed: Sparse Adaptive Federated Learning (ICLR 2025 · similarity 0.87)
- Beyond the Federation: Topology-aware Federated Learning for Generalization to Unseen Clients (ICML 2024 · similarity 0.86)
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.
