Research map / Federated learning and privacy / Communication-efficient FL
A Non-parametric View of FedAvg and FedProx:Beyond Stationary Points
Lili Su, Jiaming Xu, Pengkun Yang
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 more local steps s (s>=2) or FedProx instead of s=1 | FedAvg · Federated linear regression (M=25, d=100) | gradient norm at global risk | Hurt |
| use local steps s=5,10 or FedProx instead of s=1 | FedAvg · Federated linear regression (M=25, d=100) | estimation error ||theta_t - theta*|| | No clear effect |
| use minibatches (B=20,50,100) instead of full batch | FedAvg s=5,10 · Federated linear regression (M=25, d=100) | gradient magnitude | No clear effect |
| use minibatches (B=20,50,100) instead of full batch | FedAvg s=1 · Federated linear regression (M=25, d=100) | gradient magnitude | Hurt |
| use minibatches (B=20,50,100) instead of full batch | FedAvg and FedProx · Federated linear regression (M=25, d=100) | estimation error | No clear effect |
| vary local steps s (1,5,10) or use FedProx | FedAvg/FedProx · Federated linear regression (M=20, d=100) | federation gain | No clear effect |
| increase model heterogeneity Gamma | FedAvg/FedProx · Federated linear regression (M=20, d=100) | federation gain: drops below 1 at Gamma 7.5 (scarce), 0.3 (rich) | Hurt |
| client has scarce local data (n_i=50) instead of rich (n_i=500) | FedAvg/FedProx · Federated linear regression (M=20, d=100) | federation gain: 7.5 vs 0.3 Gamma threshold | Improved |
| increase subspace dimension r of covariates to about 16-23 | FedAvg/FedProx · Federated linear regression (subspace covariates) | federation gain: jumps from r about 16 to 23 | Mixed |
Closest papers by meaning
- Widening the Network Mitigates the Impact of Data Heterogeneity on FedAvg (ICML 2025 · similarity 0.91)
- FedGPS: Statistical Rectification Against Data Heterogeneity in Federated Learning (NeurIPS 2025 · similarity 0.90)
- Efficient Federated Learning against Heterogeneous and Non-stationary Client Unavailability (NeurIPS 2024 · similarity 0.90)
- A New Theoretical Perspective on Data Heterogeneity in Federated Optimization (ICML 2024 · similarity 0.90)
- Exact and Linear Convergence for Federated Learning under Arbitrary Client Participation is Attainable (NeurIPS 2025 · similarity 0.90)
- Momentum Benefits Non-iid Federated Learning Simply and Provably (ICLR 2024 · similarity 0.90)
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.
