Research map / Architectures and efficiency / Pruning and efficient tuning
A Layer Decomposition-Recomposition Framework for Neuron Pruning towards Accurate Lightweight 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 |
|---|---|---|---|
| Use LDRF instead of layer-by-layer pruning and reconstruction | VGG-16 · ImageNet, 5× speed-up, before fine-tuning | Error increase: 22.0% -> 2.1% | Helped |
| Use LDRF instead of layer-by-layer pruning and reconstruction | VGG-16 · ImageNet, 5× speed-up, after fine-tuning | Error increase: 1.7% -> 0.5% | Helped |
| Use LDRF instead of layer-by-layer pruning and reconstruction | VGG-9 · CIFAR-10, 5× speed-up, before fine-tuning | Error increase: 10.0% (Our impl.) -> 2.6% | Helped |
| Use LDRF with different neuron-selection heuristics rather than only Top-k (ablation) | LDRF on VGG-9 · CIFAR-10, 5× speed-up, before fine-tuning | Error increase: 2.61%, 2.81%, 2.72%, 2.57%, 2.76% | No clear effect |
| Retain information in an embedding space and propagate it when reconstructing pruned layers | LDRF · CIFAR-10, VGG-9, 5× speed-up, before fine-tuning | Error increase: 10.0% (Our impl.) -> 2.6% | Helped |
| Use LDRF rather than the listed pruning baselines | VGG-16 · ImageNet-2012 validation | Top-5 accuracy (1-view): 88.2% -> 89.4% | Helped |
| Use LDRF rather than Luo et al.-2 | ResNet-50 · ImageNet-2012 validation | Top-5 accuracy (1-view): 90.02% -> 91.55% | Helped |
Closest papers by meaning
- Neuron Merging: Compensating for Pruned Neurons (NeurIPS 2020 · similarity 0.91)
- Learning to Prune Deep Neural Networks via Layer-wise Optimal Brain Surgeon (NeurIPS 2017 · similarity 0.90)
- Pruning neural networks without any data by iteratively conserving synaptic flow (NeurIPS 2020 · similarity 0.89)
- NISP: Pruning Networks Using Neuron Importance Score Propagation (CVPR 2018 · similarity 0.89)
- Towards Meta-Pruning via Optimal Transport (ICLR 2024 · similarity 0.88)
- Sparse Training via Boosting Pruning Plasticity with Neuroregeneration (NeurIPS 2021 · 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.