IC-1336MobileNetV2 (PyTorch pre-trained on ImageNet) exhibits a failure mode under global unstructured L1 pruning at the Pareto-optimal point, with its high kurtosis of kurtoses (64.40) causing very-low-magnitude layers to be entirely pruned and disconnect the network

Gabryel Mason-Williams, Fredrik Dahlqvist

SourceWhat Makes a Good Prune? Maximal Unstructured Pruning for Maximal Cosine Similarity

The paper applies its proposed optimal L1 pruning method to three PyTorch pre-trained ImageNet models. While VGG11 and ResNet18 maintain performance after pruning at the Pareto-optimal point, MobileNetV2 suffers dramatic degradation. The mechanism is that MobileNetV2's parameter distribution has a very high kurtosis of layer-wise kurtoses (64.40 versus 1.42 for VGG11 and 5.81 for ResNet18), meaning it contains layers composed almost entirely of low-magnitude parameters. Global L1 pruning removes these small parameters first, eventually zeroing entire layers and disconnecting them from the network. A conservative adjustment (multiplying the optimal pruning percentage by 1/ln κ²) reduces the pruning to 21% for MobileNetV2 versus 51% for ResNet18.

Evidence
interventional
Key metric
kurtosis of kurtoses: 1.42 (VGG11), 5.81 (ResNet18), 64.40 (MobileNetV2); conservative pruning estimate: 51% (ResNet18), 21% (MobileNetV2)
Caveat
The authors note this is based on a 'limited set of experiments' and the 1/ln κ² adjustment is described as a 'sensible, conservative choice' rather than a proven formula.
Model
MobileNetV2, VGG / VGG13 VGG-11, ResNet / ResNet-152 / ResNet-101 / ResNet-50-BN ResNet-18
Concepts
Failure mode
Datasets
ImageNet-1k / ImageNet / ImageNet-1k-val / ImageNet-Val [eval]
Methods
L1 Magnitude Pruning [primary]
Related work
MobileNetV2 [context], VGG / VGG13 [context], ResNet / ResNet-152 / ResNet-101 / ResNet-50-BN [context]
Extraction
automatic-extraction