Advertisement

Lyapunov Functionals Associated to Neural Network

  • Eric Goles
  • Servet Martínez
Chapter
Part of the Mathematics and Its Applications book series (MAIA, volume 58)

Abstract

In Chapter 2 we analized the steady state behavior of Neural and Majority Networks by means of algebraic invariants which enabled the characterization of periods. Unfortunately, the transient behavior does not yield as readily to a study based on such class of invariants. To overcome the difficulty, we introduce here Lyapunov functionals driving the network dynamics. Using this kind of functionals, explicit bounds to the transient length will be found. Moreover, Lyapunov functionals provide a physical interpretation of Neural Networks. These operators were first introduced by Hopfield [Ho2] to analize the fixed point behaviour of random sequential iterations of associative networks. As for their application in the study of synchronous updating rules and of memory updating, they were defined and developed in [G2,GFP]. Besides, they were applied to a reversible automaton in [Po].

Keywords

Neural Network Symmetric Matrix Sequential Iteration Integer Matrix Real Symmetric Matrix 
These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

Preview

Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.

References

  1. [FG2]
    Fogelman-Soulie, F., E. Goles, G. Weisbuch, Transient length in Sequential Iteration of Threshold Functions, Disc. App. Maths., 6, 1983, 95–98.CrossRefGoogle Scholar
  2. [G2]
    Goles, E., Positive Automata Networks, in Disordered Systems and Biological Organization, edited by E. Bienenstock, F. Fogelman-Soulie, G. Weisbuch, NATO ASI, Series, F20, Springer Verlag, 1986, 101–112.Google Scholar
  3. [G8]
    Goles, E., Lyapunov Functions Associated to Automata Networks, in Automata Networks in Computer Science, F. Fogelman, Y. Robert, M. Tchuente eds., Manchester University Press, 1987, 58–81.Google Scholar
  4. [G10]
    Goles, E., Antisymmetrical Neural Networks, Disc. App. Math., 13, 1986, 97–100.MathSciNetzbMATHCrossRefGoogle Scholar
  5. [GFP]
    Goles, E., F. Fogelman-Soulie, D. Pellegrin, The Energy as a Tool for the Study of Threshold Networks, Disc. App. Math., 12, 1985, 261–277.zbMATHCrossRefGoogle Scholar
  6. [GM6]
    Goles, E., S. Martínez, Exponential Transient Classes of Symmetric Neural Networks for Synchronous and Sequential Updating, Complex Systems, 3, 1989, 589–597.MathSciNetzbMATHGoogle Scholar
  7. [G02]
    Goles, E., J., Olivos, The Convergence of Symmetric Threshold Automata, Inf. and Control, 51(2), 1981, 98–104.MathSciNetzbMATHCrossRefGoogle Scholar
  8. [HaL]
    Haken, A., M. Luby, Steepest Descent can take Exponential Time for Symmetric Connection Networks, Complex Systems, 2, 1988, 191–196.zbMATHGoogle Scholar
  9. [Ho2]
    Hopfield, J.J., Neural Networks and Physical Systems with Emergent Collective Computational Abilities, Proc. Natl. Acad. Sci. USA., 79, 1982, 2554–2558.MathSciNetCrossRefGoogle Scholar
  10. [Po]
    Pomeau, Y., Invariant in Cellular Automata, J. Phys. A17, 1984, L415–L418.MathSciNetCrossRefGoogle Scholar
  11. [T2]
    Tchuente, M., Sequential Iteration of Parallel Iteration, Theor. Comp. Sci., 48, 1986, 135–144.MathSciNetzbMATHCrossRefGoogle Scholar

Copyright information

© Kluwer Academic Publishers 1990

Authors and Affiliations

  • Eric Goles
    • 1
  • Servet Martínez
    • 1
  1. 1.Departamento de Ingeniería Matemática, Facultad de Ciencias Físicas y MatemáticasUniversidad de ChileSantiagoChile

Personalised recommendations