Theory for groupoid equivariant neural networks: an approach for steerable CNNs on bounded domains
A theory of groupoid-equivariant neural networks in which the symmetry datum consists of a groupoid, a selected pseudogroup of local bisections, a measure, and input and output representation bundles is developed, which provides significant advantages when symmetries cannot be globally implemented by group actions.
Alberto Ibort, María Jiménez-Vázquez, J. M. Pérez-Pardo
· 0 citations