Research map / Federated learning and privacy / Communication-efficient FL
Toward Understanding Generalization and Stability Gaps Between Centralized and Decentralized Federated Learning
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 |
|---|---|---|---|
| use centralized FedAvg with 2log m active clients instead of ring-topology decentralized FedAvg | FedAvg · CIFAR-10 Dir-0.1, m=100, E=5 | test accuracy (%): 59.20 -> 77.19 | Improved |
| use centralized FedAvg with 2log m clients instead of ring topology | FedAvg · CIFAR-10 Dir-0.1, m=500, E=20 | test accuracy (%): 45.20 -> 71.49 | Improved |
| use centralized FedAvg-6 instead of grid D-FedAvg at larger scale | FedAvg · CIFAR-10 Dir-0.1, m=500, E=5 | test accuracy (%): 58.20 -> 67.10 | Improved |
| use D-FedAvg on exp topology instead of FedAvg-2log m | D-FedAvg · CIFAR-10 Dir-0.1, m=100, E=20 | test accuracy (%): 76.12 -> 76.09 | Mixed |
| DFL with ring topology at 1x bandwidth vs CFL FedAvg-3 | D-FedAvg · CIFAR-10 Dir-0.1, m=100, E=5 | test accuracy (%): 59.20 -> 67.71 | Hurt |
| increase local epochs from 5 to 20 for D-FedAvg-ring | D-FedAvg-ring · CIFAR-10 Dir-0.1, m=100 | test accuracy (%): 59.20 -> 46.65 | Hurt |
| increase total clients m from 300 to 500 with fixed S=100 on full topology | D-FedAvg-full · CIFAR-10 fixed S=100 | test accuracy (%): 67.86 -> 73.12 | Improved |
| increase total clients m from 300 to 500 with fixed S=100 on exp topology | D-FedAvg-exp · CIFAR-10 fixed S=100 | test accuracy (%): 64.64 -> 67.28 | Improved |
| increase total clients m from 300 to 500 with fixed S=100 on ring topology | D-FedAvg-ring · CIFAR-10 fixed S=100 | test accuracy (%): 43.68 -> 41.19 | Hurt |
| increase total clients m from 300 to 500 with fixed S=100 on grid topology | D-FedAvg-grid · CIFAR-10 fixed S=100 | test accuracy (%): 57.28 -> 58.20 | Hurt |
| increase m from 30 to 50 with fixed S=1000 on ring topology | D-FedAvg-ring · CIFAR-10 fixed S=1000 | test accuracy (%): 63.50 -> 58.30 | Hurt |
| increase m from 30 to 50 with fixed S=1000 on full topology | D-FedAvg-full · CIFAR-10 fixed S=1000 | test accuracy (%): 71.36 -> 75.69 | Improved |
Closest papers by meaning
- Decentralized Sporadic Federated Learning: A Unified Algorithmic Framework with Convergence Guarantees (ICLR 2025 · similarity 0.90)
- SemiDFL: A Semi-Supervised Paradigm for Decentralized Federated Learning (AAAI 2025 · similarity 0.89)
- Overcoming Data and Model heterogeneities in Decentralized Federated Learning via Synthetic Anchors (ICML 2024 · similarity 0.88)
- Unveiling the Power of Multiple Gossip Steps: A Stability-Based Generalization Analysis in Decentralized Training (NeurIPS 2025 · similarity 0.88)
- Rethinking the Flat Minima Searching in Federated Learning (ICML 2024 · similarity 0.88)
- NTK-DFL: Enhancing Decentralized Federated Learning in Heterogeneous Settings via Neural Tangent Kernel (ICML 2025 · 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.
