Unsupervised Learning of Representations Invariant and Equivariant to Group Transformations
This work introduces a novel framework for unsupervised learning of representations that are separated into a group-invariant component and a group-equivariant component, valid for any group G.