AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

Pruning at Initialization – A Sketching Perspective

Noga Bar, Raja Giryes

TPAMI 2025 · Architectures and efficiency › Pruning and efficient tuning · approach: Pruning at initialization

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (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 changedWhereMeasuredEffect
replace threshold pruning with random mask sampled from SynFlow scoresSynFlow · Tiny-ImageNet ResNet-18 10% densityaccuracy: 58.3 -> 60.64Improved
replace threshold pruning with random mask sampled from SynFlow scoresSynFlow · ImageNet ResNet-50 10% densityaccuracy: 62.76 -> 63.05Improved
replace threshold pruning with random maskSynFlow · CIFAR-100 ResNet-20 2% densityaccuracy: 23.39 -> 24.89Improved
replace threshold pruning with random maskSynFlow · CIFAR-10 VGG-19 5% densityaccuracy: 92.68 -> 92.44No clear effect
replace the all-ones SynFlow input with chi-distributed random inputSynFlow · CIFAR-10 VGG-19 2% densityaccuracy: 91.68 -> 91.32No clear effect
use sparse random data and random mask in data-free SNIPSNIP + normal data · CIFAR-10 VGG-19 2% densityaccuracy: 90.35 -> 92.03Improved
use sparse random data instead of normal random data in SNIPSNIP · CIFAR-10 VGG-19 10% densityaccuracy: 92.15 -> 92.85Improved
add random mask to ProsPr with random dataProsPr + rand. data · CIFAR-100 VGG-19 10% densityaccuracy: 11.55 ± 2.11 -> 72.68 ± 0.16Improved
add random mask to IMP with random dataIMP + rand. data · CIFAR-100 VGG-19 2% densityaccuracy: 54.81 ± 6.87 -> 59.05 ± 0.14Improved
add random mask to GraSP with normal dataGraSP + normal data · CIFAR-10 ResNet-20 2% densityaccuracy: 79.59 ± 0.1 -> 78.36 ± 0.16Hurt
replace Kaiming with Xavier initialization when adding random mask to SynFlowSynFlow · CIFAR-10 VGG-19 10% densityaccuracy: 91.5 ± 0.17 -> 93.33 ± 0.18Improved
add random mask to layerwise-sparsity SNIPSNIP · CIFAR-10 VGG-19 2% densityaccuracy: 64.64 ± 10.0 -> 87.97 ± 0.12Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy