Nonlinear Phenomena in Complex Systems, 2002, Vol.5, No.3, pp.214-227

NONLINEAR PHENOMENA IN COMPLEX SYSTEMS
An Interdisciplinary Journal

2002, Vol.5, No.3, pp.214-227

Geometrical Properties of Phase Spaces of Hopfield Analog Neural Networks after Learning pp.214-227
M. Jaszuk, W. A. Kaminski, and A. D. Linkevich

Phase space of the Hopfield analog neural network is investigated in the case of inter-neuronal (synaptic) couplings adjusted with the aid of the projection learning algorithm which exploits outer products of vectors. The shape and size of basins of attraction around the memorized patterns are found to be drastically dependent on the network and the learning rule parameters.
Key words: neural network, associative memory, attractor.

Full text: Acrobat PDF (395KB) PostScript (4402KB) PostScript.gz (461KB)