Research map / Federated learning and privacy / Communication-efficient FL
Compressed-VFL: Communication-Efficient Learning with Vertically Partitioned 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 |
|---|---|---|---|
| Compress embeddings with scalar quantization at 2 bits per component instead of no compression | C-VFL · MIMIC-III | communication cost to target F1-Score 0.4: 3830.0 ± 558.2 -> 233.1 ± 28.7 | Helped |
| Use vector quantization at 3 bits per component instead of no compression | C-VFL · MIMIC-III | maximum test F1-Score: 0.448 ± 0.010 -> 0.455 ± 0.020 | No clear effect |
| Use vector quantization at 4 bits per component instead of no compression | C-VFL · CIFAR-10 | maximum test accuracy: 73.18% ± 0.44% -> 73.17% ± 0.39% | No clear effect |
| Use vector quantization at 2 bits per component instead of no compression | C-VFL · CIFAR-10 | communication cost to target accuracy 70%: 7.69 ± 0.35 -> 0.68 ± 0.06 | Helped |
| Use vector quantization at 4 bits per component instead of no compression | C-VFL · ModelNet10 | maximum test accuracy: 85.68% ± 1.57% -> 85.77% ± 1.69% | No clear effect |
| Use vector quantization at 2 bits per component instead of no compression | C-VFL · ModelNet10 | communication cost to target accuracy 75%: 9604.80 ± 2933.40 -> 593.40 ± 170.98 | Helped |
| Use top-k sparsification instead of scalar or vector quantization (ablation) | C-VFL · MIMIC-III | maximum test F1-Score: b=2: 0.431 ± 0.016; b=3: 0.435 ± 0.030; b=4: 0.453 ± 0.014 | Mixed |
| Use vector quantization instead of scalar quantization or top-k sparsification (ablation) | C-VFL · CIFAR-10 and ModelNet10 | maximum test accuracy: CIFAR-10 b=2: 71.43% ± 0.47%; ModelNet10 b=2: 84.80% ± 2.58% | Helped |
| Increase local iterations from Q=1 to Q=25 (ablation) | C-VFL with vector quantization b=3 · MIMIC-III, communication latency t_c=50 | time to target F1-Score 0.45: 3788.32 ± 822.30 -> 532.12 ± 61.49 | Helped |
| Increase local iterations from Q=1 to Q=25 (ablation) | C-VFL with vector quantization b=3 · MIMIC-III, communication latency t_c=200 | time to target F1-Score 0.45: 13259.14 ± 2878.04 -> 798.19 ± 92.23 | Helped |
Closest papers by meaning
- Stochastic Controlled Averaging for Federated Learning with Communication Compression (ICLR 2024 · similarity 0.89)
- LoCoDL: Communication-Efficient Distributed Learning with Local Training and Compression (ICLR 2025 · similarity 0.89)
- EFSkip: A New Error Feedback with Linear Speedup for Compressed Federated Learning with Arbitrary Data Heterogeneity (AAAI 2025 · similarity 0.88)
- Accelerated Methods with Compressed Communications for Distributed Optimization Problems Under Data Similarity (AAAI 2025 · similarity 0.88)
- Achieving Lossless Gradient Sparsification via Mapping to Alternative Space in Federated Learning (ICML 2024 · similarity 0.87)
- Towards Faster Decentralized Stochastic Optimization with Communication Compression (ICLR 2025 · similarity 0.87)
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.