PUB - Publikationen an der Universität Bielefeld

In this contribution we utilize recent advances in feature coding strategies for a hierarchical Neocognitron-like neu- ral architecture, which can be used for invariant recogni- tion of natural visual stimuli like objects or faces. Several researchers have identied that sparseness is an important coding principle for learning receptive eld proles that resemble response properties of simple cells in visual cor- tex. However, an ongoing discussion is concerned with the question whether sparseness should be imposed on the la- tent variables ñ as implicitly done in ICA or Sparse Coding ñ or if it should rather be imposed directly on the feature matrix. Since answers to this question have so far not been unique and were rather qualitative in nature, this paper in- vestigates the two possibilities by applying a recently in- troduced algorithm for Non-negative Matrix Factorization with Sparseness Constraints (NMFSC) to feature learning in a hierarchical recognition network. For this network, we compare recognition performance on several difcult image datasets under varying sparseness settings