List your company
ResearchFormerly WizWand

Pruning Accuracy Retention on CIFAR-100 (test)

0.774AUC (Accuracy Retention, FLOPs-based)CAP-QGW
0.387120.487560.5880.68844May 8, 2026
Updated 5mo ago

Evaluation results

MethodLinks
CAP-QGWBackbone=ResNet-18, Axis=Local axis (replaceability), Pruning Protocol=Global-threshold pruning, Sparsity levels=102026.050.774
Composite-IXBackbone=ResNet-18, Axis=Local axis (replaceability), Pruning Protocol=Global-threshold pruning, Sparsity levels=102026.050.713
MagnitudeBackbone=VGG-16, Axis=Baselines, Pruning Protocol=Global-threshold pruning, Sparsity levels=102026.050.706
Composite-PIDBackbone=ResNet-18, Axis=Target axis (task-directed PID), Pruning Protocol=Global-threshold pruning, Sparsity levels=102026.050.694
Spectral-PID-RSBackbone=ResNet-18, Axis=Target axis (task-directed PID), Pruning Protocol=Global-threshold pruning, Sparsity levels=102026.050.675
Spectral-RSBackbone=ResNet-18, Axis=Local axis (replaceability), Pruning Protocol=Global-threshold pruning, Sparsity levels=102026.050.646
Best hybrid (fixed)Backbone=ResNet-18, Axis=Hybrid (allocation × ranking), Pruning Protocol=Global-threshold pruning, Sparsity levels=102026.050.637
Best hybrid (fixed)Backbone=VGG-16, Axis=Hybrid (allocation × ranking), Pruning Protocol=Global-threshold pruning, Sparsity levels=102026.050.632
CAP-QGWBackbone=VGG-16, Axis=Local axis (replaceability), Pruning Protocol=Global-threshold pruning, Sparsity levels=102026.050.611
TaylorBackbone=VGG-16, Axis=Baselines, Pruning Protocol=Global-threshold pruning, Sparsity levels=102026.050.6
Composite-IXBackbone=VGG-16, Axis=Local axis (replaceability), Pruning Protocol=Global-threshold pruning, Sparsity levels=102026.050.592
Composite-IXBackbone=MobileNetV2, Axis=Local axis (replaceability), Pruning Protocol=Global-threshold pruning, Sparsity levels=102026.050.564
Best hybrid (fixed)Backbone=MobileNetV2, Axis=Hybrid (allocation × ranking), Pruning Protocol=Global-threshold pruning, Sparsity levels=102026.050.562
Spectral-RSBackbone=VGG-16, Axis=Local axis (replaceability), Pruning Protocol=Global-threshold pruning, Sparsity levels=102026.050.561
CAP-QGWBackbone=MobileNetV2, Axis=Local axis (replaceability), Pruning Protocol=Global-threshold pruning, Sparsity levels=102026.050.55
Spectral-RSBackbone=MobileNetV2, Axis=Local axis (replaceability), Pruning Protocol=Global-threshold pruning, Sparsity levels=102026.050.55
Spectral-PID-RSBackbone=VGG-16, Axis=Target axis (task-directed PID), Pruning Protocol=Global-threshold pruning, Sparsity levels=102026.050.543
Composite-PIDBackbone=VGG-16, Axis=Target axis (task-directed PID), Pruning Protocol=Global-threshold pruning, Sparsity levels=102026.050.504
Spectral-PID-RSBackbone=MobileNetV2, Axis=Target axis (task-directed PID), Pruning Protocol=Global-threshold pruning, Sparsity levels=102026.050.464
Composite-PIDBackbone=MobileNetV2, Axis=Target axis (task-directed PID), Pruning Protocol=Global-threshold pruning, Sparsity levels=102026.050.457
MagnitudeBackbone=ResNet-18, Axis=Baselines, Pruning Protocol=Global-threshold pruning, Sparsity levels=102026.050.454
TaylorBackbone=MobileNetV2, Axis=Baselines, Pruning Protocol=Global-threshold pruning, Sparsity levels=102026.050.442
MagnitudeBackbone=MobileNetV2, Axis=Baselines, Pruning Protocol=Global-threshold pruning, Sparsity levels=102026.050.42
TaylorBackbone=ResNet-18, Axis=Baselines, Pruning Protocol=Global-threshold pruning, Sparsity levels=102026.050.402