Research map / Architectures and efficiency / Pruning and efficient tuning
SNIP: SINGLE-SHOT NETWORK PRUNING BASED ON CONNECTION SENSITIVITY
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 |
|---|---|---|---|
| Apply SNIP pruning at 95.0% sparsity | LeNet-300-100 · MNIST | error (%): 1.7 -> 1.6 | Helped |
| Apply SNIP pruning at 98.0% sparsity | LeNet-300-100 · MNIST | error (%): 1.7 -> 2.4 | Hurt |
| Apply SNIP pruning at 98.0% sparsity | LeNet-5-Caffe · MNIST | error (%): 0.9 -> 0.8 | Helped |
| Apply SNIP pruning at 99.0% sparsity | LeNet-5-Caffe · MNIST | error (%): 0.9 -> 1.1 | Hurt |
| Prune with SNIP at the reported sparsities | CNN and residual models · CIFAR-10 | error (%): +0.87; +0.58; +0.45; +0.33; -0.26; +0.42; +0.52; -0.29 | Mixed |
| Prune with SNIP at 95.0% sparsity | LSTM and GRU models · Sequential MNIST | error (%): -0.31; +0.20; +0.54; -0.19 | Mixed |
| Prune with SNIP at the reported sparsities | AlexNet and VGG models · Tiny-ImageNet | error (%): +2.75; +2.78; +0.99; +0.15; +0.87 | Hurt |
| Use variance scaling initialization rather than random normal initialization (ablation) | SNIP on LSTM-small · Sequential MNIST | error: 2.93 ± (0.20) -> 1.48 ± (0.09) | Helped |
| Use variance scaling initialization rather than random normal initialization (ablation) | SNIP on GRU-small · Sequential MNIST | error: 47.61 ± (20.49) -> 1.80 ± (0.10) | Helped |
| Use variance scaling initialization rather than random normal initialization (ablation) | SNIP on LeNet-300-100 · MNIST | error: 1.90 ± (0.09); 1.91 ± (0.10); 1.88 ± (0.10) | No clear effect |
| Increase the mini-batch used to compute saliency (ablation) | SNIP on LeNet-300-100 · MNIST | error | Helped |
| Prune using a single-example saliency batch (ablation) | SNIP on LeNet-300-100 · MNIST | connection relevance | Helped |
Closest papers by meaning
- Pruning from Scratch (AAAI 2020 · similarity 0.91)
- A Signal Propagation Perspective for Pruning Neural Networks at Initialization (ICLR 2020 · similarity 0.91)
- The Unreasonable Effectiveness of Random Pruning: Return of the Most Naive Baseline for Sparse Training (ICLR 2022 · similarity 0.90)
- A Probabilistic Approach to Neural Network Pruning (ICML 2021 · similarity 0.90)
- Prospect Pruning: Finding Trainable Weights at Initialization using Meta-Gradients (ICLR 2022 · similarity 0.90)
- Learning Pruning-Friendly Networks via Frank-Wolfe: One-Shot, Any-Sparsity, And No Retraining (ICLR 2022 · similarity 0.89)
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.