Research map / Architectures and efficiency / CNN design
Fast ConvNets Using Group-Wise Brain Damage
Check 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 gradual group-wise sparsification instead of tensor decomposition | AlexNet conv2 and conv3 · AlexNet conv2 and conv3 | speed-up: 6.6x -> 8.5x | Helped |
| Replace gradual sparsification with group-wise sparsification followed by fine-tuning (ablation) | AlexNet conv2 and conv3 · AlexNet conv2 and conv3 | speed-up: 8.5x -> 5x | Hurt |
| Sparsify all five convolutional layers gradually rather than with fixed patterns (ablation) | AlexNet · AlexNet all five convolutional layers | speed-up: 3.0x -> 3.2x | Helped |
| Add group-sparsity regularization during training before group-wise pruning (ablation) | LeNet · MNIST | error rate: 2.1% -> 1.71% | Helped |
| Use an l2,1 group regularizer rather than an l1 regularizer before pruning (ablation) | LeNet · MNIST | error rate | Helped |
| Fine-tune after fixing the sparsity patterns (ablation) | LeNet with group-sparse convolutions · MNIST | error rate: 2.1% -> 1.71% | Helped |
| Replace learned sparsity patterns with predefined fixed patterns (ablation) | LeNet · MNIST | accuracy | Hurt |
| Use gradual group-wise sparsification rather than sparsification with fine-tuning (ablation) | AlexNet conv2 · AlexNet conv2 | accuracy drop: 1.13% -> 0.28% | Helped |
| Increase pruning by using gradual group-wise sparsification | AlexNet conv2 · AlexNet conv2 | speed-up: 2.7x -> 9.0x | Helped |
| Group-wise sparsify all convolutional layers | AlexNet feature extractor · INRIA Holidays | mAP: 0.783 -> 0.780 | Hurt |
| Group-wise sparsify all convolutional layers | AlexNet feature extractor · Oxford Building | mAP: 0.45 to 0.41 | Hurt |
Closest papers by meaning
- Optimal Brain Apoptosis (ICLR 2025 · similarity 0.86)
- Faster CNNs with Direct Sparse Convolutions and Guided Pruning (ICLR 2017 · similarity 0.85)
- Pruning Filters for Efficient ConvNets (ICLR 2017 · similarity 0.85)
- On the VC dimension of deep group convolutional neural networks (NeurIPS 2025 · similarity 0.84)
- GSE: Group-wise Sparse and Explainable Adversarial Attacks (ICLR 2025 · similarity 0.83)
- Benchmarking Out-of-Distribution Generalization Capabilities of DNN-based Encoding Models for the Ventral Visual Cortex. (NeurIPS 2024 · similarity 0.83)
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.