Reducing Symmetry Increase in Equivariant Neural Networks
A detailed and in-depth characterization of symmetry increase is provided together with a principled framework for its reduction and practical guidelines for feature design to prevent harmful symmetry increases are proposed.