Investigating how ReLU-networks encode symmetries
This provides insights into how symmetries are encoded in ReLU-networks, offering a new perspective on existing conjectures for researchers in equivariant deep learning.
The paper investigates whether equivariance in neural networks implies layerwise equivariance, finding theoretical cases where it does and does not, and supports this with experiments on VGG-nets and ResNets, showing it is easier to merge a network with a transformed version than merging different networks.
Many data symmetries can be described in terms of group equivariance and the most common way of encoding group equivariances in neural networks is by building linear layers that are group equivariant. In this work we investigate whether equivariance of a network implies that all layers are equivariant. On the theoretical side we find cases where equivariance implies layerwise equivariance, but also demonstrate that this is not the case generally. Nevertheless, we conjecture that CNNs that are trained to be equivariant will exhibit layerwise equivariance and explain how this conjecture is a weaker version of the recent permutation conjecture by Entezari et al. [2022]. We perform quantitative experiments with VGG-nets on CIFAR10 and qualitative experiments with ResNets on ImageNet to illustrate and support our theoretical findings. These experiments are not only of interest for understanding how group equivariance is encoded in ReLU-networks, but they also give a new perspective on Entezari et al.'s permutation conjecture as we find that it is typically easier to merge a network with a group-transformed version of itself than merging two different networks.