On the universality of neural encodings in CNNs

TL;DR

This paper investigates whether neural encodings in CNNs trained on natural images are universal across datasets, proposing a new method to compare learned weights and revealing fundamental universal features.

Contribution

It introduces a novel procedure to compare CNN weights directly, demonstrating the universality of neural encodings across different natural image datasets.

Findings

Eigenvectors in CNN layers are universal across datasets

Supports the existence of a universal neural encoding for natural images

Provides insights into transfer learning success

Abstract

We explore the universality of neural encodings in convolutional neural networks trained on image classification tasks. We develop a procedure to directly compare the learned weights rather than their representations. It is based on a factorization of spatial and channel dimensions and measures the similarity of aligned weight covariances. We show that, for a range of layers of VGG-type networks, the learned eigenvectors appear to be universal across different natural image datasets. Our results suggest the existence of a universal neural encoding for natural images. They explain, at a more fundamental level, the success of transfer learning. Our work shows that, instead of aiming at maximizing the performance of neural networks, one can alternatively attempt to maximize the universality of the learned encoding, in order to build a principled foundation model.