Nonlinear Phenomena in Complex Systems, 2006, Vol.9, No.2, pp.163-172

NONLINEAR PHENOMENA IN COMPLEX SYSTEMS
An Interdisciplinary Journal

2006, Vol.9, No.2, pp.163-172

Beyond Attractor Neural Networks for Pattern Recognition.
H. Haken

This paper develops a novel approach to model pattern recognition by the human brain. Pattern recognition is, as usual, conceived as action of an associative memory. We start from a pulse-coupled neural network with a minimal nonlinear coupling between neurons. We treat the limiting case of dense pulses. Because of chopped signals the processes can be described as initial value problem leading to attractors in form of sets of limit cycles. Because of saturation of attention the attractors close and become quasi-attractors. Depending on the interplay between signals and attention saturation, the recognition system wanders from quasi-attractor to quasi-attractor.
Key words: pattern recognition, pulse-coupling, neural network, quasi-attractors

Full text: Acrobat PDF (180KB)