Research map / Architectures and efficiency / Pruning and efficient tuning
“Learning-Compression” Algorithms for Neural Net 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 |
|---|---|---|---|
| Use LC ℓ0-constraint pruning rather than magnitude pruning | ResNet-32 · CIFAR10 test, 15% surviving weights | test error (%): 8.85 -> 7.32 | Helped |
| Use LC ℓ0-constraint pruning rather than magnitude pruning | ResNet-56 · CIFAR10 test, 10% surviving weights | test error (%): 8.66 -> 6.67 | Helped |
| Use LC ℓ0-constraint pruning rather than magnitude pruning | ResNet-110 · CIFAR10 test, 10% surviving weights | test error (%): 7.20 -> 6.50 | Helped |
| Use LC ℓ0-constraint pruning rather than magnitude pruning | ResNet-110 · CIFAR10 test, 3% surviving weights | test error (%): 9.32 -> 7.61 | Helped |
| Use the constraint form rather than the penalty form (ablation) | LC pruning on LeNet300 · MNIST | test error (%) | Mixed |
| Use ℓ0 cost rather than ℓ1 cost (ablation) | LC pruning on LeNet300 · MNIST | test error (%) | Mixed |
| Retrain surviving weights after pruning (ablation) | LC ℓ1 pruning on LeNet300 · MNIST | test error (%): 10.16 -> 2.49 | Helped |
| Retrain surviving weights after pruning (ablation) | LC ℓ0 pruning on LeNet300 · MNIST | test error (%): 2.27 -> 2.13 | No clear effect |
| Use soft iterative pruning that can revisit which weights are marked | LC pruning · LeNet and ResNets | test error | Helped |
Closest papers by meaning
- Pruning from Scratch (AAAI 2020 · similarity 0.92)
- Computation and Memory-Efficient Model Compression with Gradient Reweighting (NeurIPS 2025 · similarity 0.92)
- Pruning Neural Networks at Initialization: Why Are We Missing the Mark? (ICLR 2021 · similarity 0.92)
- Pruning Neural Networks via Coresets and Convex Geometry: Towards No Assumptions (NeurIPS 2022 · similarity 0.91)
- Data-Independent Neural Pruning via Coresets (ICLR 2020 · similarity 0.91)
- EPSD: Early Pruning with Self-Distillation for Efficient Model Compression (AAAI 2024 · similarity 0.90)
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.