On the universality of neural encodings in CNNs
This work addresses the problem of understanding the fundamental properties of neural networks for researchers in machine learning, though it is incremental as it builds on existing transfer learning concepts.
The authors investigated whether convolutional neural networks develop universal neural encodings across different natural image datasets, finding that learned eigenvectors in VGG-type networks appear universal, suggesting a foundational basis for transfer learning.
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.