Research map / Federated learning and privacy / Communication-efficient FL
SCAFFOLD: Stochastic Controlled Averaging for Federated 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 SCAFFOLD instead of FedAvg | Federated learning on EMNIST · EMNIST, 2-layer fully connected network, 0% similarity | test accuracy: 0.787 -> 0.801 | Helped |
| Use SCAFFOLD instead of FedAvg | Federated learning on EMNIST · EMNIST, 2-layer fully connected network, 10% similarity | test accuracy: 0.828 -> 0.842 | Helped |
| Increase client similarity from 0% to 10% (ablation) | FedAvg · EMNIST, 2-layer fully connected network | test accuracy: 0.787 -> 0.828 | Helped |
| Increase client similarity from 0% to 10% (ablation) | SCAFFOLD · EMNIST, 2-layer fully connected network | test accuracy: 0.801 -> 0.842 | Helped |
| Increase client similarity from 0% to 10% (ablation) | SGD · EMNIST, 2-layer fully connected network | test accuracy: 0.766 -> 0.764 | No clear effect |
| Use SCAFFOLD instead of FedAvg | EMNIST logistic regression, 0% similarity, 20% clients · EMNIST logistic regression, 0% similarity, 20% clients | communication rounds to 0.45 test accuracy: 179 -> 143 | Helped |
| Use SCAFFOLD instead of FedAvg | EMNIST logistic regression, 0% similarity, 5% clients · EMNIST logistic regression, 0% similarity, 5% clients | communication rounds to 0.45 test accuracy: 334 -> 290 | Helped |
| Use SCAFFOLD instead of FedAvg | EMNIST logistic regression, 0% similarity, 1% clients · EMNIST logistic regression, 0% similarity, 1% clients | communication rounds to 0.45 test accuracy: 1k+ vs 790 | Helped |
| Use SCAFFOLD instead of FedAvg | EMNIST logistic regression, 10% similarity, 20% clients · EMNIST logistic regression, 10% similarity, 20% clients | communication rounds to 0.45 test accuracy: 12 -> 9 | Helped |
| Use SCAFFOLD instead of FedAvg | EMNIST logistic regression, 10% similarity, 5% clients · EMNIST logistic regression, 10% similarity, 5% clients | communication rounds to 0.45 test accuracy: 17 -> 13 | Helped |
| Use SCAFFOLD instead of FedAvg | EMNIST logistic regression, 10% similarity, 1% clients · EMNIST logistic regression, 10% similarity, 1% clients | communication rounds to 0.45 test accuracy: 35 -> 28 | Helped |
| Increase local steps from K=2 to K=10 (ablation) | SCAFFOLD · Simulated quadratic functions, G in [1, 10, 100] | convergence speed: about 2 times fast -> about 5 times faster | Helped |
Closest papers by meaning
- Scaffold with Stochastic Gradients: New Analysis with Linear Speed-Up (ICML 2025 · similarity 0.95)
- Stochastic Controlled Averaging for Federated Learning with Communication Compression (ICLR 2024 · similarity 0.91)
- SCAFFLSA: Taming Heterogeneity in Federated Linear Stochastic Approximation and TD Learning (NeurIPS 2024 · similarity 0.90)
- Improving Generalization in Federated Learning with Highly Heterogeneous Data via Momentum-Based Stochastic Controlled Weight Averaging (ICML 2025 · similarity 0.90)
- Exact and Linear Convergence for Federated Learning under Arbitrary Client Participation is Attainable (NeurIPS 2025 · similarity 0.90)
- Decentralized Sporadic Federated Learning: A Unified Algorithmic Framework with Convergence Guarantees (ICLR 2025 · 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.