Research map / Architectures and efficiency / Pruning and efficient tuning
DSD: Dense-Sparse-Dense Training for Deep Neural Networks
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 dense-sparse-dense training | GoogLeNet · ImageNet | Top-1 error: 31.14% -> 30.02% | Helped |
| Apply dense-sparse-dense training | VGG-16 · ImageNet | Top-1 error: 31.50% -> 27.19% | Helped |
| Apply dense-sparse-dense training | ResNet-18 · ImageNet | Top-1 error: 30.43% -> 29.17% | Helped |
| Apply dense-sparse-dense training | ResNet-50 · ImageNet | Top-1 error: 24.01% -> 22.89% | Helped |
| Apply dense-sparse-dense training | NeuralTalk · Flickr-8K | BLEU-4: 16.8 -> 18.5 | Helped |
| Apply dense-sparse-dense training | DeepSpeech 1 · WSJ '93 | WER: 33.55 -> 31.59 | Helped |
| Apply dense-sparse-dense training | DeepSpeech 2 · WSJ '93 | WER: 14.52 -> 13.44 | Helped |
| Add the final dense retraining phase after sparse training (ablation) | GoogLeNet · ImageNet | Top-1 error: 30.58% -> 30.02% | Helped |
| Add the final dense retraining phase after sparse training (ablation) | VGG-16 · ImageNet | Top-1 error: 28.19% -> 27.19% | Helped |
| Add the final dense retraining phase after sparse training (ablation) | NeuralTalk · Flickr-8K | BLEU-4: 17.5 -> 18.5 | Helped |
| Add a second DSD iteration (ablation) | DeepSpeech 1 · WSJ '93 | WER: 32.20 -> 31.59 | Helped |
| Add a second DSD iteration (ablation) | DeepSpeech 2 · WSJ '93 | WER: 13.96 -> 13.44 | Helped |
Closest papers by meaning
- The Unreasonable Effectiveness of Random Pruning: Return of the Most Naive Baseline for Sparse Training (ICLR 2022 · similarity 0.87)
- Learning Pruning-Friendly Networks via Frank-Wolfe: One-Shot, Any-Sparsity, And No Retraining (ICLR 2022 · similarity 0.87)
- On the Hardness of Training Deep Neural Networks Discretely (AAAI 2025 · similarity 0.87)
- Sparse maximal update parameterization: A holistic approach to sparse training dynamics (NeurIPS 2024 · similarity 0.86)
- Navigating Extremes: Dynamic Sparsity in Large Output Spaces (NeurIPS 2024 · similarity 0.85)
- Dynamic Sparse Training versus Dense Training: The Unexpected Winner in Image Corruption Robustness (ICLR 2025 · similarity 0.85)
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.