Quantum-mechanical tunneling in associative neural networks

  • M. Dugic
  • D. Rakovic


We investigate the quantum-mechanical tunneling between the “patterns" of the, so-called, associative neural networks. Being the relatively stable minima of the “configuration-energy" space of the networks, the “patterns" represent the macroscopically distinguishable states of the neural nets. Therefore, the tunneling represents a macroscopic quantum effect, but with some special characteristics. Particularly, we investigate the tunneling between the minima of approximately equal depth, thus requiring no energy exchange. If there are at least a few such minima, the tunneling represents a sort of the “random walk" process, which implies the quantum fluctuations in the system, and therefore “malfunctioning" in the information processing of the nets. Due to the finite number of the minima, the “random walk" reduces to a dynamics modeled by the, so-called, Pauli master equation. With some plausible assumptions, the set(s) of the Pauli master equations can be analytically solved. This way comes the main result of this paper: the quantum fluctuations due to the quantum-mechanical tunneling can be “minimized" if the “pattern"-formation is such that there are mutually “distant" groups of the “patterns", thus providing the “zone" structure of the “pattern" formation. This qualitative result can be considered as a basis of the efficient deterministic functioning of the associative neural nets.

PACS. 02.50.Ey Stochastic processes - 05.40.-a Fluctuation phenomena, random processes, noise, and Brownian motion 


Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.

Copyright information

© EDP Sciences, Springer-Verlag, Società Italiana di Fisica 2000

Authors and Affiliations

  • M. Dugic
    • 1
  • D. Rakovic
    • 2
  1. 1.Faculty of Science, Dept. Phys., P.O. Box 60, 34 000 Kragujevac, Yugoslavia
  2. 2.Faculty of Electrical Engineering, P.O. Box 35-54, 11120 Belgrade, Yugoslavia

Personalised recommendations