Research map / Architectures and efficiency / Pruning and efficient tuning
Only Train Once: A One-Shot Neural Network Training And Pruning Framework
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 |
|---|---|---|---|
| Use zero-invariant groups and HSPG for one-shot structured pruning | VGG16 · CIFAR10 | Top-1 accuracy: 91.6% -> 91.0% | Helped |
| Use zero-invariant groups and HSPG for one-shot structured pruning | VGG16-BN · CIFAR10 | FLOPs remaining: 100% -> 26.8% | Helped |
| Use zero-invariant groups and HSPG for one-shot structured pruning | ResNet50 · CIFAR10 | FLOPs remaining: 100% -> 12.8% | Helped |
| Use zero-invariant groups and HSPG for one-shot structured pruning | ResNet50 · ImageNet | Top-1 accuracy: 76.1% -> 74.7% | Hurt |
| Increase training effort | OTO ResNet50 · ImageNet | Top-1 accuracy: 74.7% -> 75.1% | Helped |
| Use structured group pruning rather than unstructured parameter pruning | BERT · SQuAD | Inference speedup: 1× -> 1.8× | Helped |
| Delay the switch to group-sparsity stage from T to 2T or 3T (ablation) | OTO · CIFAR10 | Top-1 accuracy: 90.9% ± 0.3% across N=T, 2T, 3T | No clear effect |
| Include ZIG partition for ResConv-BN (ablation) | OTO ResNet50 · ImageNet | Top-1 accuracy: up to 71% Top-1 Accuracy | Hurt |
| Remove inference-optimized structured group pruning and use unstructured sparsity | BERT pruning · SQuAD | Inference speedup: 1× versus up to 1.8× | Hurt |
Closest papers by meaning
- Learning Pruning-Friendly Networks via Frank-Wolfe: One-Shot, Any-Sparsity, And No Retraining (ICLR 2022 · similarity 0.92)
- Auto-Train-Once: Controller Network Guided Automatic Network Pruning from Scratch (CVPR 2024 · similarity 0.91)
- Learning to Prune Deep Neural Networks via Layer-wise Optimal Brain Surgeon (NeurIPS 2017 · similarity 0.90)
- OSSCAR: One-Shot Structured Pruning in Vision and Language Models with Combinatorial Optimization (ICML 2024 · similarity 0.89)
- Automatic Joint Structured Pruning and Quantization for Efficient Neural Network Training and Compression (CVPR 2025 · similarity 0.88)
- Preserving Deep Representations in One-Shot Pruning: A Hessian-Free Second-Order Optimization Framework (ICLR 2025 · 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.