AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

Fantastic Weights and How to Find Them: Where to Prune in Dynamic Sparse Training

Aleksandra Nowak, Bram Grooten, Decebal Constantin Mocanu, Jacek Tabor

NeurIPS 2023 · Architectures and efficiency › Pruning and efficient tuning · approach: Sparse training from scratch · 1 citations (OpenAlex)

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
use SNIP (Taylor/gradient) instead of magnitude pruningDST · CIFAR10/CIFAR100/Tiny-ImageNet models at low densitytest accuracyHurt
use SNIP instead of magnitude pruning at density 0.9 (random growth)DST on RoBERTa Large · CommonsenseQAaccuracy: 73.80 -> 56.57Hurt
use RSensitivity instead of magnitude pruning at high densityDST · CIFAR10 large-MLP and ResNet models, density > 0.15test accuracyImproved
use RSensitivity instead of magnitude at low density (random growth, density 0.9)DST on RoBERTa Large · CommonsenseQAaccuracy: 73.80 -> 25.87Hurt
use MEST instead of magnitude pruningDST · vision modelstest accuracyNo clear effect
increase MEST lambda toward infinity (ablation)MEST · vision modelstest accuracyHurt
use SET instead of magnitude pruningDST · CIFAR10 first-update masksmask overlapNo clear effect
DST instead of static sparse training at low densitysparse training · vision/tabular modelstest accuracyImproved
increase update period from small to 800 (random growth) (ablation)DST · MLP CIFAR10 density 0.2/0.1validation accuracyImproved
increase update period to 1600 (gradient growth) (ablation)DST · MLP CIFAR10validation accuracyImproved
use very small update period (25) with gradient-based pruning and gradient growth (ablation)DST · CIFAR10validation accuracyHurt
choose gradient growth over random growthDST · convolutional modelstest accuracyNo clear effect

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