What Affects Learned Equivariance in Deep Image Recognition Models?

Robert-Jan Bruintjes*, Tomasz Motyka*, Jan van Gemert
2nd Workshop on Learning with Limited Labelled Data for Image and Video Understanding @ CVPR 2023 (CVPRW 2023)
Equivariance w.r.t. geometric transformations in neural networks improves data efficiency, parameter efficiency and robustness to out-of-domain perspective shifts. When equivariance is not designed into a neural network, the network can still learn equivariant functions from the data. We quantify this learned equivariance, by proposing an improved measure for equivariance. We find evidence for a correlation between learned translation equivariance and validation accuracy on ImageNet. We therefore investigate what can increase the learned equivariance in neural networks, and find that data augmentation, reduced model capacity and inductive bias in the form of convolutions induce higher learned equivariance in neural networks.


@misc{bruintjes2023affects, title={What Affects Learned Equivariance in Deep Image Recognition Models?}, author={Robert-Jan Bruintjes and Tomasz Motyka and Jan van Gemert}, year={2023}, eprint={2304.02628}, archivePrefix={arXiv}, primaryClass={cs.CV} }