IC-673Trained depthwise convolutional kernels in DS-CNN architectures converge to identifiable DoG-like patterns, with over 95% of ConvNeXtV2 and over 90% of ConvNeXt filters classifiable into a small set of clusters

Zahra Babaiee, Peyman Kiasari, Daniela Rus, Radu Grosu

SourceUnveiling the Unseen: Identifiable Clusters in Trained Depthwise Convolutional Kernels

The authors applied unsupervised autoencoder-based clustering to millions of depthwise convolutional kernels extracted from pre-trained DS-CNN models (ConvNeXt, ConvNeXtV2, HoNet, MoGANet, ConvMixer, EfficientNet, MobileNetV3, MixNet, MNASNet, ReplkNet-XL). The kernels converge into approximately ten identifiable clusters resembling difference-of-Gaussian (DoG) functions, their first and second-order derivatives, and cross-shaped patterns. The patterns persist across all layers, with the composition shifting: DoG patterns dominate early layers while cross-shaped patterns become predominant in the final convolutional layer. The degree of convergence varies by architecture, being highest in ConvNeXtV2 (up to 98.16%) and lowest in ConvMixer (56.64%).

Evidence
observational
Key metric
ConvNeXtV2 tiny 22k: 98.16% filters clustered; ConvNeXtV2 tiny 1k: 97.33%; ConvNeXt tiny: 95.21%; HoNet tiny: 80.71%; MoGANet small: 78.87%; ConvMixer 768-32: 56.64%; EfficientNet-B4: 73.78%; MobileNetV3 large: 87.09%
Caveat
The ConvMixer model shows the weakest convergence (56.64%) with noisier kernels. The 3x3 kernels (MobileNetV3) could not be well-fit by the autoencoder and required a separate k-means method. The authors note the cause of cross-shaped filters remains uncertain.
Model
ConvNeXtV2, ConvNeXt, MoGANet, ConvMixer, EfficientNet, MobileNetV3, MNASNet, ReplkNet-XL
Datasets
ImageNet-1k / ImageNet / ImageNet-1k-val / ImageNet-Val [source], ImageNet-21k / ImageNet-22k [source]
Methods
Principal component analysis [supporting], K-means clustering [supporting]
Extraction
automatic-extraction