Abstract
We introduce two constructions in geometric deep learning for 1) transportingorientation-dependent convolutional filters over a manifold in a continuous wayand thereby defining a convolution operator that naturally incorporates therotational effect of holonomy; and 2) allowing efficient evaluation of manifoldconvolution layers by sampling manifold valued random variables that centeraround a weighted Brownian motion maximum likelihood mean. Both methods areinspired by stochastics on manifolds and geometric statistics, and provideexamples of how stochastic methods -- here horizontal frame bundle flows andnon-linear bridge sampling schemes, can be used in geometric deep learning. Weoutline the theoretical foundation of the two methods, discuss their relationto Euclidean deep networks and existing methodology in geometric deep learning,and establish important properties of the proposed constructions.