Research map / Optimization / Distributed and compressed
AdaComp : Adaptive Residual Gradient Compression for Data-Parallel 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 |
|---|---|---|---|
| Compress gradients with AdaComp | MNIST-CNN · MNIST | top-1 error: 0.88% -> 0.85% | No clear effect |
| Compress gradients with AdaComp | CIFAR10-CNN · CIFAR10 | top-1 error: 18% -> 18.4% | No clear effect |
| Compress gradients with AdaComp | AlexNet · ImageNet | top-1 error: 42.7% -> 42.9% | No clear effect |
| Compress gradients with AdaComp | ResNet18 · ImageNet | top-1 error: 32.41% -> 32.87% | No clear effect |
| Compress gradients with AdaComp | ResNet50 · ImageNet | top-1 error: 28.91% -> 29.15% | No clear effect |
| Compress gradients with AdaComp | BN50-DNN · BN50 | top-1 error: 59.8% -> 59.8% | No clear effect |
| Compress gradients with AdaComp | LSTM · Shakespeare | top-1 error: 1.73% -> 1.75% | No clear effect |
| Compress gradients with AdaComp | CIFAR10-CNN with Adam · CIFAR10 | test error: 18.1% vs compression:18.3% | No clear effect |
| Use AdaComp local selection and adaptive threshold | CIFAR10-CNN · CIFAR10-CNN | compression rate: 2000× | Helped |
| Remove adaptive soft-threshold (ablation) | AdaComp local selection on CIFAR10-CNN · CIFAR10-CNN | test error: 17.84% | Hurt |
| Remove adaptive soft-threshold (ablation) | AdaComp local selection on CIFAR10-CNN at high compression · CIFAR10-CNN | test error: 22%; 2000×; 600× | Hurt |
| Use AdaComp instead of fixed-count Local Selection (ablation) | CIFAR10-CNN · CIFAR10-CNN | residual gradient magnitude: 240K to 239K | Helped |
Closest papers by meaning
- ScaleCom: Scalable Sparsified Gradient Compression for Communication-Efficient Distributed Training (NeurIPS 2020 · similarity 0.90)
- LEGACY: A Lightweight Dynamic Gradient Compression Strategy for Distributed Deep Learning (ICLR 2026 · similarity 0.89)
- Deep Gradient Compression: Reducing the Communication Bandwidth for Distributed Training (ICLR 2018 · similarity 0.89)
- On the Discrepancy between the Theoretical Analysis and Practical Implementations of Compressed Communication for Distributed Deep Learning (AAAI 2020 · similarity 0.89)
- Sketched Adaptive Distributed Deep Learning: A Sharp Convergence Analysis (NeurIPS 2025 · similarity 0.88)
- Subspace Networks: Scaling Decentralized Training with Communication-Efficient Model Parallelism (NeurIPS 2025 · 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.