"Homogeneous vector bundles and G-equivariant convolutional neural networks"Aronsson, JimmyG-equivariant convolutional neural networks (GCNNs) can be motivated and constructed from homogeneous vector bundles over homogeneous G-spaces M = G/K. However, convolutional layers are not the only possible G-equivariant layers. We give a bandlimit-criterion that characterizes when G-equivariant layers are expressible as convolutions, in the general setting of unimodular Lie groups G of type I and compact subgroups K. |
| http://univie.ac.at/projektservice-mathematik/e/talks/Aronsson_2021-06_Online_ICCHA_2021_Jimmy_Aronsson.pdf |
« back