Feedforward nets for interpolation and classification

Eduardo D. Sontag

doi:10.1016/0022-0000(92)90039-L

Feedforward nets for interpolation and classification

Eduardo D. Sontag

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

104 Scopus citations

Abstract

This paper deals with single-hidden-layer feedforward nets, studying various aspects of classification power and interpolation capability. In particular, a worst-case analysis shows that direct input to output connections in threshold nets double the recognition but not the interpolation power, while using sigmoids rather than thresholds allows doubling both. For other measures of classification, including the Vapnik-Chervonenkis dimension, the effect of direct connections or sigmoidal activations is studied in the special case of two-dimensional inputs.

Original language	English (US)
Pages (from-to)	20-48
Number of pages	29
Journal	Journal of Computer and System Sciences
Volume	45
Issue number	1
DOIs	https://doi.org/10.1016/0022-0000(92)90039-L
State	Published - Aug 1992

All Science Journal Classification (ASJC) codes

Theoretical Computer Science
Computer Networks and Communications
Computational Theory and Mathematics
Applied Mathematics

Access to Document

10.1016/0022-0000(92)90039-L

Cite this

@article{cb1f96f8b4ba40eea5f4ea49f400e10b,

title = "Feedforward nets for interpolation and classification",

abstract = "This paper deals with single-hidden-layer feedforward nets, studying various aspects of classification power and interpolation capability. In particular, a worst-case analysis shows that direct input to output connections in threshold nets double the recognition but not the interpolation power, while using sigmoids rather than thresholds allows doubling both. For other measures of classification, including the Vapnik-Chervonenkis dimension, the effect of direct connections or sigmoidal activations is studied in the special case of two-dimensional inputs.",

author = "Sontag, {Eduardo D.}",

note = "Funding Information: * Supported in part by US Air Force Grant AFOSR-880235 and by Siemens Corporate Research.",

year = "1992",

month = aug,

doi = "10.1016/0022-0000(92)90039-L",

language = "English (US)",

volume = "45",

pages = "20--48",

journal = "Journal of Computer and System Sciences",

issn = "0022-0000",

publisher = "Academic Press Inc.",

number = "1",

}

TY - JOUR

T1 - Feedforward nets for interpolation and classification

AU - Sontag, Eduardo D.

N1 - Funding Information: * Supported in part by US Air Force Grant AFOSR-880235 and by Siemens Corporate Research.

PY - 1992/8

Y1 - 1992/8

N2 - This paper deals with single-hidden-layer feedforward nets, studying various aspects of classification power and interpolation capability. In particular, a worst-case analysis shows that direct input to output connections in threshold nets double the recognition but not the interpolation power, while using sigmoids rather than thresholds allows doubling both. For other measures of classification, including the Vapnik-Chervonenkis dimension, the effect of direct connections or sigmoidal activations is studied in the special case of two-dimensional inputs.

AB - This paper deals with single-hidden-layer feedforward nets, studying various aspects of classification power and interpolation capability. In particular, a worst-case analysis shows that direct input to output connections in threshold nets double the recognition but not the interpolation power, while using sigmoids rather than thresholds allows doubling both. For other measures of classification, including the Vapnik-Chervonenkis dimension, the effect of direct connections or sigmoidal activations is studied in the special case of two-dimensional inputs.

UR - http://www.scopus.com/inward/record.url?scp=0026904597&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=0026904597&partnerID=8YFLogxK

U2 - 10.1016/0022-0000(92)90039-L

DO - 10.1016/0022-0000(92)90039-L

M3 - Article

AN - SCOPUS:0026904597

SN - 0022-0000

VL - 45

SP - 20

EP - 48

JO - Journal of Computer and System Sciences

JF - Journal of Computer and System Sciences

IS - 1

ER -

Feedforward nets for interpolation and classification

Abstract

All Science Journal Classification (ASJC) codes

Access to Document

Other files and links

Fingerprint

Cite this