Binary Factorization by Neural Autoassociator

D. Husek (Czech Republic), A.A. Frolov, I. Muraviev (Russia), H. Rezankova, V. Snasel (Czech Republic), and P. Polyakov (Russia)


Boolean factorization, recurrent neural network, SingleStep approximation, neurodynamics, computersimulation, feature extraction, Hebbian learning


The unsupervised learning of feature extraction in high dimensional patterns is a central problem for neural network approach. Feature extraction is the procedure which maps original patterns into the features (or factors) space of reduced dimension. In this paper we demonstrate that Hebbian learning in Hopfield-like neural network is a natural procedure for unsupervised learning of feature extraction. Due to this learning, factors become the attractors of network dynamics, hence they can be revealed by the random search. The neurodynamics is analyzed by Single-Step approximation, which is known [1] to be rather accurate for sparsely encoded Hopfield network. Thus, the analysis is restricted by the case of sparsely encoded factors. The accuracy of Single-Step approximation is confirmed by computer simulations.

Important Links:

Go Back