A characterization of the existence of energies for neural networks
In this paper we give under an appropriate theoretical frame-work a characterization about neural networks which admit an energy. We prove that a neural network admits an energy if and only if the weight matrix verifies two conditions: the diagonal elements are non-negative and the associated incidence graph does not admit non-quasi-symmetric circuits.
Unable to display preview. Download preview PDF.
- 1.Fogelman-Soulié, F., Goles, E., Weisbuch, G.: Transient length in sequential iteration of threshold functions. Disc. Appl. Math. 6, (1983) 95–98Google Scholar
- 2.Goles, E., Fogelman-Soulié, F., Pellegrin, D.: Decreasing energy functions as a tool for studying threshold netwoks. Disc. Appl. Math. 12, (1985) 261–277Google Scholar
- 3.Goles, E., Martinez, S.: Neural and automata networks. Math. and Applications, 58, Kluwer Acad. Pub., (1990)Google Scholar
- 4.Hopfield, J.J.: Neural networks and physical systems with emergent collective computational abilities. Proc. Natl. Acad. Sci., USA, 79, (1982) 2554–2558Google Scholar
- 5.Kobuchi, Y.: State evaluation functions and Lyapunov functions for neural networks. Neural Networks, 4, (1991) 505–510Google Scholar
- 6.X. Xu W.T. Tsai, Construction associative memories using neural networks. Neural Networks, 3, (1990) 301–309Google Scholar