Research map / Federated learning and privacy / Differential privacy
Toward the Flatter Landscape and Better Generalization in Federated Learning Under Client-Level Differential Privacy
Yifan Shi, Kang Wei, Li Shen, Yingqi Liu, Xueqian Wang, Bo Yuan, 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 |
|---|---|---|---|
| replace local SGD with SAM optimizer | DP-FedAvg · CIFAR-10 (Dirichlet 0.6) | validation accuracy (%): 50.05 -> 54.14 | Improved |
| replace local SGD with SAM optimizer | DP-FedAvg · CIFAR-100 (Dirichlet 0.6) | validation accuracy (%): 15.92 -> 19.09 | Improved |
| replace local SGD with SAM optimizer | DP-FedAvg · EMNIST (Dirichlet 0.6) | validation accuracy (%): 82.20 -> 84.32 | Improved |
| add top-k local update sparsification (p=0.4) to DP-FedSAM (ablation) | DP-FedSAM · CIFAR-10 (Dirichlet 0.6) | validation accuracy (%): 54.14 -> 57.00 | Improved |
| add top-k local update sparsification (p=0.4) to DP-FedSAM (ablation) | DP-FedSAM · CIFAR-100 (Dirichlet 0.6) | validation accuracy (%): 19.09 -> 21.24 | Improved |
| lower sparsity ratio from p=0.4 to p=0.1 (ablation) | DP-FedSAM-topk · CIFAR-10 (Dirichlet 0.6) | validation accuracy (%): 57.00 -> 55.05 | Hurt |
| raise sparsity ratio from p=0.4 to p=0.8 (ablation) | DP-FedSAM-topk · CIFAR-100 (Dirichlet 0.6) | validation accuracy (%): 21.24 -> 19.79 | Hurt |
| replace Fed-SMP-topk with DP-FedSAM-topk (SAM added) (ablation) | Fed-SMP-topk · EMNIST (Dirichlet 0.6) | generalization gap (train-val, %): 16.31 -> 11.07 | Improved |
| replace DP-FedAvg with DP-FedSAM (ablation) | DP-FedAvg · EMNIST (Dirichlet 0.6) | generalization gap (train-val, %): 17.35 -> 10.75 | Improved |
| replace DP-FedAvg with DP-FedSAM-topk under privacy budget epsilon=4 | DP-FedAvg · CIFAR-10 (Dirichlet 0.6) | test accuracy (%): 38.23 -> 38.96 | Improved |
| replace DP-FedAvg with DP-FedSAM-topk under privacy budget epsilon=10 | DP-FedAvg · CIFAR-10 (Dirichlet 0.6) | test accuracy (%): 49.06 -> 56.36 | Improved |
| replace DP-FedAvg with DP-FedSAM-topk under privacy budget epsilon=10 | DP-FedAvg · CIFAR-100 (Dirichlet 0.6) | test accuracy (%): 15.23 -> 20.79 | Improved |
Closest papers by meaning
- Make Landscape Flatter in Differentially Private Federated Learning (CVPR 2023 · similarity 0.98)
- Towards the Robustness of Differentially Private Federated Learning (AAAI 2024 · similarity 0.92)
- Noise-Aware Algorithm for Heterogeneous Differentially Private Federated Learning (ICML 2024 · similarity 0.90)
- Mitigating the Privacy–Utility Trade-off in Decentralized Federated Learning via f-Differential Privacy (NeurIPS 2025 · similarity 0.89)
- Privacy-Preserving Federated Convex Optimization: Balancing Partial-Participation and Efficiency via Noise Cancellation (ICML 2025 · similarity 0.88)
- One-shot Empirical Privacy Estimation for Federated Learning (ICLR 2024 · similarity 0.88)
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.
