AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

GDP: Stabilized Neural Network Pruning via Gates With Differentiable Polarization

Yi Guo, Huan Yuan, Jianchao Tan, Zhangyang Wang, Sen Yang, Ji Liu

ICCV 2021 · Architectures and efficiency › Pruning and efficient tuning · approach: Structured network pruning · 41 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
introduce separate learnable gate parameters alpha instead of using conv kernel norms as gate input (ablation)GDP · ImageNet MobileNet-V2top-1 accuracyImproved
use gates g(alpha) in forward pass instead of naive smooth-L0 regularization on weights (ablation)GDP · ImageNet MobileNet-V2top-1 accuracyImproved
prune from a well-trained model instead of from scratch (ablation)GDP · ImageNet MobileNet-V2top-1 accuracyImproved
use proximal l1 relaxation instead of proximal l0 (ablation)GDP optimizer · pruningstabilityImproved
vary decay rate of epsilon from 0.86 to 0.96 (ablation)GDP · ImageNet MobileNet-V2top-1 accuracy / compressionNo clear effect
vary initial epsilon value (ablation)GDP · ImageNet MobileNet-V2top-1 accuracy / compressionNo clear effect
prune DeepLabv3+ ResNet-50 to 38.4% FLOPsGDP · Pascal VOC 2012 valmIoU: 0.774 -> 0.779Improved
prune DeepLabv3+ ResNet-50 to 5.2% FLOPsGDP · Pascal VOC 2012 valmIoU: 0.774 -> 0.666Hurt
prune VGG-16 with GDP vs unpruned baselineGDP · CIFAR-10 VGG-16accuracy: 93.89 -> 93.99Improved
prune ResNet-56 to 34.36% FLOPsGDP · CIFAR-10 ResNet-56accuracy: 93.9 -> 93.55Hurt
prune ResNet-56 to 46.65% FLOPsGDP · CIFAR-10 ResNet-56accuracy: 93.9 -> 93.97Improved
prune MobileNet-V2 to 53.78% FLOPsGDP · CIFAR-10 MobileNet-V2accuracy: 94.89 -> 95.15Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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