Stability of Feedback-Type Neural Networks Having Nonsymmetric Interconnections
Hitoshi Mada · Japanese Journal of Applied Physics · 1993
A Hopfield network has symmetric interconnections that carry many patterns of memory. Dynamics of the Hopfield model are explained with the energy function constructed by a quadratic form of input vectors and interconnections. However, it is difficult to implement complete symmetric interconnection with practical hardwares. Neural networks having nonsymmetric interconnections cannot be discussed using the energy function. This paper describes a method to treat stability of a Hopfield model having nonsymmetric interconnections. The discussed mathematical forms are analogous to nonequilibrium thermodynamics. A stable condition is theoretically obtained for nonsymmetrically interconnected neural networks.