No label defined (Q4343455)
From MaRDI portal
!
This is the item page for this Wikibase entity, intended for internal use and editing purposes. Please use the normal view instead:
scientific article; zbMATH DE number 1031011
| Language | Label | Description | Also known as |
|---|---|---|---|
| default for all languages | No label defined |
||
| English | No label defined |
scientific article; zbMATH DE number 1031011 |
Statements
18 December 1997
0 references
Potts-Hopfield model
0 references
neural networks and optimization
0 references
storage capacity problem
0 references
discrete optimization problem
0 references
0.8166807293891907
0 references
0.7944494485855103
0 references
0.78934645652771
0 references
0.7888270020484924
0 references
We study the storage capacity of the Potts-Hopfield model on associative memories and on the relations between neural networks and optimization. Concerning the storage capacity problem, we study more general neural networks where each node can take a finite number of states (Potts-Hopfield model). In Section 3 we supply the result obtained by the author, \textit{P. Ferrari} and \textit{P. Picco} [J. Stat. Phys. 66, 55 (1992)] which asserts that the storage capacity is of order \(M=\alpha N\) when the size of the network \(N\to\infty\), this is a generalization of the Newman theorem for the Hopfield model. In Section 4 we show the scaling \(M\approx N/(6+\varepsilon)\log N\) for preserving the original patterns as fixed points, a result of \textit{G. Weisbuch} and \textit{F. Fogelman-Soulié} [J. Phys. Lett. 46, L623--L630 (1985)]. The relations between neural networks and optimization is set in a general framework in Section 5, we consider a class of local functions studied by \textit{E. Goles} and the author [``Neural and automata networks'' (Kluwer, 1990)] which generalize the sign or the threshold functions. We supply a duality theorem between norm optimization problems [\textit{E. Goles} and the author, in: Statistical physics, automata networks and dynamical systems. Math. Appl. 75, 65-87 (1992)], which allows to show the equivalence between a discrete optimization problem and a continuous one, a result of \textit{A. J. E. M. Janssen} [Inf. Process. Lett. 37, No. 3, 155-157 (1991; Zbl 0713.68009)]. The algorithm which solves this class of optimization problems corresponds to the synchronous evolution of neural networks. For the entire collection see [Zbl 0864.00066].
0 references