Research map / Optimization / Distributed and compressed
On the Convergence of FedAvg on Non-IID Data
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 |
|---|---|---|---|
| Increase local steps E (ablation) | FedAvg · Unbalanced federated datasets | communication rounds to reach a fixed accuracy | Mixed |
| Reduce the number of active devices per round (ablation) | FedAvg · synthetic(0,0) | convergence | No clear effect |
| Use Scheme I rather than Scheme II (ablation) | FedAvg · mnist balanced | convergence | Helped |
| Use Scheme I rather than the original scheme (ablation) | FedAvg · mnist balanced | convergence | Helped |
| Use Scheme I rather than Scheme II (ablation) | FedAvg · mnist unbalanced | convergence | Helped |
| Use Scheme I rather than the original scheme (ablation) | FedAvg · mnist unbalanced | convergence | Helped |
| Use Scheme II on unbalanced data (ablation) | FedAvg · mnist unbalanced | convergence stability | Hurt |
| Transform Scheme II by scaling local objectives (ablation) | FedAvg · mnist unbalanced | convergence rate | Hurt |
| Use the original sampling and averaging scheme rather than Scheme I (ablation) | FedAvg · Federated datasets | convergence speed | Hurt |
| Increase the number of participating devices K (ablation) | FedAvg · synthetic(0,0) | convergence | Helped |
| Use a fixed learning rate instead of a decaying learning rate (ablation) | FedAvg with E>1 and full batch size · Constructed strongly convex and smooth optimization problem | distance to optimum: Ω((E-1)η) | Hurt |
Closest papers by meaning
- Widening the Network Mitigates the Impact of Data Heterogeneity on FedAvg (ICML 2025 · similarity 0.90)
- Exact and Linear Convergence for Federated Learning under Arbitrary Client Participation is Attainable (NeurIPS 2025 · similarity 0.90)
- A New Theoretical Perspective on Data Heterogeneity in Federated Optimization (ICML 2024 · similarity 0.88)
- Scaffold with Stochastic Gradients: New Analysis with Linear Speed-Up (ICML 2025 · similarity 0.88)
- Momentum Benefits Non-iid Federated Learning Simply and Provably (ICLR 2024 · similarity 0.88)
- Decentralized Sporadic Federated Learning: A Unified Algorithmic Framework with Convergence Guarantees (ICLR 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.