Research map / Architectures and efficiency / Pruning and efficient tuning
Winning the Lottery Ahead of Time: Efficient Early Network Pruning
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 |
|---|---|---|---|
| Prune early in training rather than before training | CroP pruning on image-classification models · CIFAR10, CIFAR100, Tiny-ImageNet | test accuracy | Helped |
| Prune at epoch 1 rather than before training | ResNet50 pruned to 98% sparsity · CIFAR100 | final test accuracy: 3% | Helped |
| Prune at epoch 26 rather than before training | ResNet50 pruned to 98% sparsity · CIFAR100 | final test accuracy: 11% | Helped |
| Continue dense training before pruning after epoch 30 | ResNet50 pruned to 98% sparsity · CIFAR100 | final test accuracy | No clear effect |
| Use structured pruning early rather than the dense model | EarlyCroP-S · VGG16/ImageNet-2012 | Top-1 accuracy: 58.78% -> 61.43% | Helped |
| Use structured EarlyCroP instead of dense training | VGG16 · ImageNet-2012, same 18 epochs | training time (hours): 62 -> 51 | Helped |
| Use EarlyCroP-U instead of LTR | BERT · 5 of 8 language tasks | score: 10× faster | Mixed |
| Add iterative pruning to CroP-S before training instead of pruning early (ablation) | ResNet18 · CIFAR10, 95% sparsity | test accuracy: 87.8% -> 91.0% | Hurt |
| Add iterative pruning to CroP-U before training instead of pruning early (ablation) | ResNet18 · CIFAR10, 95% sparsity | test accuracy: 89.1% -> 91.1% | Hurt |
| Use structured rather than unstructured early pruning (ablation) | EarlyCroP · CartPole-v0 | average returned environment reward | Helped |
| Train unstructured models longer (ablation) | EarlyCroP-U · CartPole-v0 | average returned environment reward | No clear effect |
| Use early structured pruning instead of CroP-S before training (ablation) | VGG16 · CIFAR100, 98% sparsity | test accuracy: 57.4% -> 62.2% | Helped |
Closest papers by meaning
- SPDY: Accurate Pruning with Speedup Guarantees (ICML 2022 · similarity 0.90)
- The Unreasonable Effectiveness of Random Pruning: Return of the Most Naive Baseline for Sparse Training (ICLR 2022 · similarity 0.89)
- Sparsest Models Elude Pruning: An Exposé of Pruning’s Current Capabilities (ICML 2024 · similarity 0.89)
- How Sparse Can We Prune A Deep Network: A Fundamental Limit Perspective (NeurIPS 2024 · similarity 0.87)
- EPSD: Early Pruning with Self-Distillation for Efficient Model Compression (AAAI 2024 · similarity 0.87)
- SAFE: Finding Sparse and Flat Minima to Improve Pruning (ICML 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.