AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

SInGE: Sparsity via Integrated Gradients Estimation of Neuron Relevance

Edouard YVINEC, Arnaud Dapogny, Matthieu Cord, Kevin Bailly

NeurIPS 2022 · Architectures and efficiency › Pruning and efficient tuning · approach: Structured network pruning · 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 L2 norm instead of L1 for magnitude criterion (ablation)magnitude pruning · CIFAR10 ResNet-56, 75% params removedtop-1 accuracy: 42.01 -> 42.35No clear effect
use gradient criterion instead of magnitude (ablation)pruning criterion · CIFAR10 ResNet-56, 75% params removedtop-1 accuracy: 42.35 -> 77.68Improved
combine magnitude x gradient instead of gradient only (ablation)pruning criterion · CIFAR10 ResNet-56, 85% params removedtop-1 accuracy: 28.31 -> 90.28Improved
integrate gradient along path to zero (IG) instead of magnitude x gradient (ablation)pruning criterion · CIFAR10 ResNet-56, 85% params removedtop-1 accuracy: 90.28 -> 92.80Improved
integrate gradient along path to zero (IG) instead of magnitude x gradient (ablation)pruning criterion · CIFAR10 ResNet-56, 90% params removedtop-1 accuracy: 10.00 -> 85.38Improved
keep magnitude term in integrated criterion (IG vs SG) (ablation)pruning criterion · CIFAR10 ResNet-56, 90% params removedtop-1 accuracy: 77.20 -> 85.38Improved
entwine pruning and fine-tuning instead of post-pruning fine-tuning (ablation)SInGE · CIFAR10 ResNet-56, 90% params removed, 1000 stepstop-1 accuracy: 77.2 -> 85.38Improved
entwine pruning and fine-tuning instead of post-pruning (ablation)SInGE · CIFAR10 ResNet-56, 85% params removed, 5000 stepstop-1 accuracy: 93.13 -> 93.31Improved
increase fine-tuning steps from 1000 to 5000 (entwined) (ablation)SInGE · CIFAR10 ResNet-56, 90% params removedtop-1 accuracy: 85.38 -> 90.02Improved
increase fine-tuning steps from 1000 to 5000 (entwined) (ablation)SInGE · CIFAR10 ResNet-56, 85% params removedtop-1 accuracy: 93.18 -> 93.31Improved
SInGE vs HAP at lossless accuracySInGE · CIFAR10 ResNet-56 (93.5 acc)% parameters removed: 76.2 -> 84.3Improved
SInGE vs MDP at 40% goalSInGE · ImageNet MobileNetV2top-1 accuracy: 69.58 -> 70.47Improved

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