1 paper
Daiyuan Li, Shreya Arya, Robert Ghrist
Most equivariant neural networks rely on a single global symmetry, limiting their use in domains where symmetries are instead local. We introduce Torsor CNNs, a framework for learn…