Geometric Back-propagation in Morphological Neural Networks
Rick Groenendijk, Leo Dorst, Theo Gevers
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.
Topics & Concepts
Dilation (metric space)Mathematical morphologyConvolutional neural networkArtificial intelligenceArtificial neural networkComputer scienceBackpropagationConvergence (economics)Pattern recognition (psychology)Cellular neural networkAlgorithmComputer visionMathematicsGeometryImage processingImage (mathematics)EconomicsEconomic growthNeural Networks and ApplicationsTopological and Geometric Data AnalysisAdvanced Numerical Analysis Techniques