Geometric Back-Propagation in Morphological Neural Networks.

Groenendijk, Rick; Dorst, Leo; Gevers, Theo · IEEE Trans Pattern Anal Mach Intell · 2023

basic_science · Level V

Where this comes from

Abstract

This paper provides a definition of back-propagation through geometric correspondences for morphological neural networks. In addition, dilation layers are shown to learn probe geometry by erosion of layer inputs and outputs. A proof-of-principle is provided, in which predictions and convergence of morphological networks significantly outperform convolutional networks.