Research map / Federated learning and privacy / Communication-efficient FL
FetchSGD: Communication-Efficient Federated Learning with Sketching
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 FetchSGD with a 12.4M-column sketch instead of uncompressed SGD | GPT2-small · PersonaChat validation | perplexity: 14.9 ± 0.02 -> 14.8 ± 0.002 | Helped |
| Use FetchSGD with a 1.24M-column sketch instead of uncompressed SGD | GPT2-small · PersonaChat validation | perplexity: 14.9 ± 0.02 -> 15.8 ± 0.007 | Hurt |
| Use FetchSGD rather than local top-k at k=50,000 | GPT2-small · PersonaChat validation | perplexity: 19.3 ± 0.05 -> 15.8 ± 0.007 | Helped |
| Use FetchSGD rather than local top-k at k=500,000 | GPT2-small · PersonaChat validation | perplexity: 17.1 ± 0.02 -> 14.8 ± 0.002 | Helped |
| Use FetchSGD rather than FedAvg with 2 local iterations | GPT2-small · PersonaChat validation | perplexity: 16.3 ± 0.2 -> 14.8 ± 0.002 | Helped |
| Use FetchSGD rather than FedAvg with 5 local iterations | GPT2-small · PersonaChat validation | perplexity: 20.1 ± 0.02 -> 14.8 ± 0.002 | Helped |
| Use FetchSGD sketching rather than local top-k compression | GPT2-small · PersonaChat | perplexity | Mixed |
| Add global momentum to local top-k (ablation) | Local top-k · PersonaChat | perplexity | Hurt |
| Use true top-k with intermediate k instead of uncompressed updates (ablation) | FetchSGD-style true top-k · PersonaChat validation | perplexity | Helped |
| Use true top-k with large k and momentum factor masking (ablation) | FetchSGD-style true top-k · PersonaChat validation | perplexity | Hurt |
| Use a single error-accumulation sketch instead of sliding-window error accumulation (ablation) | FetchSGD · Federated learning experiments | convergence | No clear effect |
Closest papers by meaning
- FedNS: A Fast Sketching Newton-Type Algorithm for Federated Learning (AAAI 2024 · similarity 0.88)
- Communication-Efficient Federated Learning with Accelerated Client Gradient (CVPR 2024 · similarity 0.87)
- FedSMU: Communication-Efficient and Generalization-Enhanced Federated Learning through Symbolic Model Updates (ICML 2025 · similarity 0.87)
- LoCoDL: Communication-Efficient Distributed Learning with Local Training and Compression (ICLR 2025 · similarity 0.87)
- The Panaceas for Improving Low-Rank Decomposition in Communication-Efficient Federated Learning (ICML 2025 · similarity 0.86)
- Revisiting Ensembling in One-Shot Federated Learning (NeurIPS 2024 · similarity 0.86)
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.