Research map / Optimization / Distributed and compressed
ScaleCom: Scalable Sparsified Gradient Compression for Communication-Efficient Distributed Training
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 |
|---|---|---|---|
| Apply ScaleCom at 92X compression | ResNet34 · CIFAR10, standard batch size | accuracy: 93.78 -> 93.98 | No clear effect |
| Apply ScaleCom at 112X compression | ResNet18 · ImageNet, standard batch size | accuracy: 70.482 -> 70.172 | No clear effect |
| Apply ScaleCom at 96X compression | ResNet50 · ImageNet, standard batch size | accuracy: 76.442 -> 75.988 | No clear effect |
| Apply ScaleCom at 155X compression | MobileNetV2 · ImageNet, standard batch size | accuracy: 71.644 -> 71.524 | No clear effect |
| Apply ScaleCom at 47X compression | Transformer-base · WMT14 En-De, standard batch size | BLEU: 27.64 -> 27.27 | No clear effect |
| Apply ScaleCom at 400X compression | 4-bidirectional-LSTM Speech · SWB300, standard batch size | WER: 10.4 -> 10.1 | Helped |
| Apply ScaleCom at 92X compression | ResNet34 · CIFAR10, large batch size | accuracy: 93.75 -> 93.36 | No clear effect |
| Apply ScaleCom at 112X compression | ResNet18 · ImageNet, large batch size | accuracy: 70.285 -> 69.879 | No clear effect |
| Apply ScaleCom at 96X compression | ResNet50 · ImageNet, large batch size | accuracy: 76.473 -> 75.895 | No clear effect |
| Apply ScaleCom at 155X compression | MobileNetV2 · ImageNet, large batch size | accuracy: 71.487 -> 71.014 | No clear effect |
| Apply ScaleCom at 47X compression | Transformer-base · WMT14 En-De, large batch size | BLEU: 27.79 -> 28.03 | Helped |
| Apply ScaleCom at 100X compression | 4-bidirectional-LSTM Speech · SWB300, large batch size | WER: 9.9 -> 10.0 | No clear effect |
Closest papers by meaning
- Deep Gradient Compression: Reducing the Communication Bandwidth for Distributed Training (ICLR 2018 · similarity 0.92)
- On the Discrepancy between the Theoretical Analysis and Practical Implementations of Compressed Communication for Distributed Deep Learning (AAAI 2020 · similarity 0.91)
- LEGACY: A Lightweight Dynamic Gradient Compression Strategy for Distributed Deep Learning (ICLR 2026 · similarity 0.90)
- AdaComp : Adaptive Residual Gradient Compression for Data-Parallel Distributed Training (AAAI 2018 · similarity 0.90)
- Subspace Networks: Scaling Decentralized Training with Communication-Efficient Model Parallelism (NeurIPS 2025 · similarity 0.89)
- Towards a Better Theoretical Understanding of Independent Subnetwork Training (ICML 2024 · 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.