STORAGE CAPACITY OF A MULTILAYER NEURAL NETWORK WITH BINARY WEIGHTS

被引：32

作者：

BARKAI, E

KANTER, I

机构：

[1] Department of Physics, Bar-Ilan University, Ramat Gan

来源：

EUROPHYSICS LETTERS | 1991年 / 14卷 / 02期

关键词：

D O I：

10.1209/0295-5075/14/2/003

中图分类号：

O4 [物理学];

学科分类号：

0702 ;

摘要：

Statistical mechanics is applied to estimate the maximal capacity per weight (alpha-c) of a two-layer feed-forward network with binary weights, functioning as a parity machine of the hidden units. For K greater-than-or-equal-to 2 hidden units, the maximal theoretical capacity is achieved, alpha-c = 1, and the average overlap between different solutions is zero. These results agree with the simulations. At finite temperature one-step replica symmetry breaking solution is found, which appears to be exact.

引用

页码：107 / 112

页数：6