Hierarchical Autoassociators

Authors: Horst Bischof, Andreas Weingessel and Kurt Hornik (Vienna University of Technology)

Abstract: In this paper, we present a combination of autoassociative networks with hierarchical methods. The so called Principal Component Pyramid (PCA-pyramid) is a hierarchically cascaded, locally connected autoassociative neural network. This network was derived by combining the structure of image pyramids (a commonly used data structure in image processing and computer vision) with neural network learning algorithms. We will present some theoretical results for these structures as well as applications to image compression and feature extraction.