Research map / Federated learning and privacy / Communication-efficient FL
Federated Learning with Client Subsampling, Data Heterogeneity, and Unbounded Smoothness: A New Algorithm and Lower Bounds
Michael Crawshaw, Yajie Bao, Mingrui Liu
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 EPISODE++ instead of CELGC | federated RNN · SNLI | test accuracy: 77.45 -> 83.89 | Improved |
| use EPISODE++ instead of Clipped Minibatch SGD | federated RNN · SNLI | test accuracy: 72.74 -> 83.89 | Improved |
| use EPISODE++ instead of NaiveParallelClip (with less communication) | federated RNN · SNLI | test accuracy: 81.74 -> 83.89 | Improved |
| use EPISODE++ instead of CELGC | federated RNN · Sentiment140 | test accuracy: 76.36 -> 77.97 | Improved |
| use EPISODE++ instead of NaiveParallelClip | federated RNN · Sentiment140 | train loss: 0.365 -> 0.258 | Improved |
| replace episodic clipping with per-step clipping (SCAFFOLDClip) (ablation) | EPISODE++ · SNLI | test accuracy: 83.89 -> 81.11 | Hurt |
| replace episodic clipping with per-step clipping (SCAFFOLDClip) (ablation) | EPISODE++ · Sentiment140 | test accuracy: 77.97 -> 77.38 | Hurt |
| ignore information from previous rounds (EPISODE) (ablation) | EPISODE++ · SNLI | test accuracy: 83.89 -> 80.24 | Hurt |
| ignore information from previous rounds (EPISODE) (ablation) | EPISODE++ · Sentiment140 | train loss: 0.258 -> 0.466 | Hurt |
| decrease number of participating clients S | EPISODE++ · SNLI/Sentiment140 | test accuracy | No clear effect |
| decrease number of participating clients S | Clipped minibatch SGD, NaiveParallelClip, EPISODE · SNLI/Sentiment140 | test accuracy | Hurt |
| decrease data similarity s | CELGC, Clipped minibatch SGD, NaiveParallelClip · SNLI/Sentiment140 | test accuracy | Hurt |
Closest papers by meaning
- Federated Learning under Periodic Client Participation and Heterogeneous Data: A New Communication-Efficient Algorithm and Analysis (NeurIPS 2024 · similarity 0.91)
- Byzantine-Robust Federated Learning: Impact of Client Subsampling and Local Updates (ICML 2024 · similarity 0.90)
- Federated Learning with Sample-level Client Drift Mitigation (AAAI 2025 · similarity 0.89)
- A New Theoretical Perspective on Data Heterogeneity in Federated Optimization (ICML 2024 · similarity 0.89)
- Exact and Linear Convergence for Federated Learning under Arbitrary Client Participation is Attainable (NeurIPS 2025 · similarity 0.89)
- Efficient Federated Learning against Heterogeneous and Non-stationary Client Unavailability (NeurIPS 2024 · 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.
