Finite size scaling in neural networks

Walter Nadler; Wolfgang Fink

doi:10.1103/PhysRevLett.78.555

Finite size scaling in neural networks

Walter Nadler, Wolfgang Fink

Research output: Contribution to journal › Article › peer-review

5 Scopus citations

Abstract

We demonstrate that the fraction of pattern sets that can be stored in single- and hidden-layer perceptrons exhibits finite size scaling. This feature allows one to estimate the critical storage capacity α_c from simulations of relatively small systems. We illustrate this approach by determining α_c, together with the finite size scaling exponent υ, for storing Gaussian patterns in committee and parity machines with binary couplings and up to K = 5 hidden units.

Original language	English (US)
Pages (from-to)	555-558
Number of pages	4
Journal	Physical review letters
Volume	78
Issue number	3
DOIs	https://doi.org/10.1103/PhysRevLett.78.555
State	Published - Jan 20 1997
Externally published	Yes

ASJC Scopus subject areas

General Physics and Astronomy

Access to Document

10.1103/PhysRevLett.78.555

Cite this

@article{1001214213de4ce6a9f74facfe8b0e53,

title = "Finite size scaling in neural networks",

abstract = "We demonstrate that the fraction of pattern sets that can be stored in single- and hidden-layer perceptrons exhibits finite size scaling. This feature allows one to estimate the critical storage capacity αc from simulations of relatively small systems. We illustrate this approach by determining αc, together with the finite size scaling exponent υ, for storing Gaussian patterns in committee and parity machines with binary couplings and up to K = 5 hidden units.",

author = "Walter Nadler and Wolfgang Fink",

year = "1997",

month = jan,

day = "20",

doi = "10.1103/PhysRevLett.78.555",

language = "English (US)",

volume = "78",

pages = "555--558",

journal = "Physical review letters",

issn = "0031-9007",

publisher = "American Physical Society",

number = "3",

}

TY - JOUR

T1 - Finite size scaling in neural networks

AU - Nadler, Walter

AU - Fink, Wolfgang

PY - 1997/1/20

Y1 - 1997/1/20

N2 - We demonstrate that the fraction of pattern sets that can be stored in single- and hidden-layer perceptrons exhibits finite size scaling. This feature allows one to estimate the critical storage capacity αc from simulations of relatively small systems. We illustrate this approach by determining αc, together with the finite size scaling exponent υ, for storing Gaussian patterns in committee and parity machines with binary couplings and up to K = 5 hidden units.

AB - We demonstrate that the fraction of pattern sets that can be stored in single- and hidden-layer perceptrons exhibits finite size scaling. This feature allows one to estimate the critical storage capacity αc from simulations of relatively small systems. We illustrate this approach by determining αc, together with the finite size scaling exponent υ, for storing Gaussian patterns in committee and parity machines with binary couplings and up to K = 5 hidden units.

UR - http://www.scopus.com/inward/record.url?scp=4243318064&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=4243318064&partnerID=8YFLogxK

U2 - 10.1103/PhysRevLett.78.555

DO - 10.1103/PhysRevLett.78.555

M3 - Article

AN - SCOPUS:4243318064

SN - 0031-9007

VL - 78

SP - 555

EP - 558

JO - Physical review letters

JF - Physical review letters

IS - 3

ER -

Finite size scaling in neural networks

Abstract

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this