Research map / Architectures and efficiency / Pruning and efficient tuning
Neuron Merging: Compensating for Pruned Neurons
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 |
|---|---|---|---|
| Add neuron merging after structured pruning | LeNet-300-100 · FashionMNIST, 80% pruning with l2-norm selection | accuracy: 63.90% -> 77.11% | Helped |
| Add neuron merging after structured pruning | VGG-16 · CIFAR-10 | accuracy: 88.70% -> 93.16% | Helped |
| Add neuron merging after structured pruning | VGG-16 · CIFAR-100 | accuracy: 67.38% -> 71.95% | Helped |
| Add neuron merging after structured pruning | ResNet-56 · CIFAR-10, pruning ratios 20%, 30%, 40%, and 50% | accuracy | Helped |
| Add neuron merging after structured pruning | WideResNet-40-4 · CIFAR-10, 50% pruning | accuracy | Helped |
| Add neuron merging after structured pruning | VGG-16 · ImageNet, last convolution layer 70% pruning | Top 1 accuracy: 35.15% -> 43.99% | Helped |
| Add neuron merging after structured pruning | ResNet-34 · ImageNet, 30% pruning | Top 1 accuracy: 12.52% -> 35.56% | Helped |
| Add neuron merging after structured pruning | VGG-16, ResNet-56, and WideResNet-40-4 · CIFAR-10 | WARE: 4.285 -> 1.465; 12.095 -> 4.986; 3.502 -> 2.364 | Helped |
| Use l2-GM rather than l1-norm or l2-norm for filter selection | VGG-16 · CIFAR-100 | accuracy: 71.63% / 71.86% -> 71.95% | Helped |
| Set cosine threshold t to 0.15 to compensate filters above that similarity | ResNet-56 · CIFAR-10 | accuracy | Helped |
Closest papers by meaning
- A Layer Decomposition-Recomposition Framework for Neuron Pruning towards Accurate Lightweight Networks (AAAI 2019 · similarity 0.91)
- NISP: Pruning Networks Using Neuron Importance Score Propagation (CVPR 2018 · similarity 0.90)
- Importance Estimation for Neural Network Pruning (CVPR 2019 · similarity 0.89)
- Pruning from Scratch (AAAI 2020 · similarity 0.89)
- Towards Meta-Pruning via Optimal Transport (ICLR 2024 · similarity 0.89)
- $C^2M^3$: Cycle-Consistent Multi-Model Merging (NeurIPS 2024 · 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.