Recovering a Feed-Forward Net From Its Output

Charles Fefferman; Scott Markel

Recovering a Feed-Forward Net From Its Output

Charles Fefferman
, Scott Markel

Mathematics

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

16 Scopus citations

Abstract

We study feed-forward nets with arbitrarily many layers, using the standard sigmoid, tanh x. Aside from technicalities, our theorems are: 1. Complete knowledge of the output of a neural net for arbitrary inputs uniquely specifies the architecture, weights and thresholds; and 2. There are only finitely many critical points on the error surface for a generic training problem.

Original language	English (US)
Title of host publication	Advances in Neural Information Processing Systems 6, NIPS 1993
Editors	J. Cowan, G. Tesauro, J. Alspector
Publisher	Neural information processing systems foundation
Pages	335-342
Number of pages	8
ISBN (Electronic)	1558603220, 9781558603226
State	Published - 1993
Event	6th Advances in Neural Information Processing Systems, NIPS 1993 - Denver, United States Duration: Nov 29 1993 → Dec 2 1993

Publication series

Name	Advances in Neural Information Processing Systems
Volume	6
ISSN (Print)	1049-5258

Conference

Conference	6th Advances in Neural Information Processing Systems, NIPS 1993
Country/Territory	United States
City	Denver
Period	11/29/93 → 12/2/93

All Science Journal Classification (ASJC) codes

Signal Processing
Information Systems
Computer Networks and Communications

Cite this

@inproceedings{757e80f3ef2144d0811dbc771325caec,

title = "Recovering a Feed-Forward Net From Its Output",

abstract = "We study feed-forward nets with arbitrarily many layers, using the standard sigmoid, tanh x. Aside from technicalities, our theorems are: 1. Complete knowledge of the output of a neural net for arbitrary inputs uniquely specifies the architecture, weights and thresholds; and 2. There are only finitely many critical points on the error surface for a generic training problem.",

author = "Charles Fefferman and Scott Markel",

note = "Publisher Copyright: {\textcopyright} 1993 Neural information processing systems foundation. All rights reserved.; 6th Advances in Neural Information Processing Systems, NIPS 1993 ; Conference date: 29-11-1993 Through 02-12-1993",

year = "1993",

language = "English (US)",

series = "Advances in Neural Information Processing Systems",

publisher = "Neural information processing systems foundation",

pages = "335--342",

editor = "J. Cowan and G. Tesauro and J. Alspector",

booktitle = "Advances in Neural Information Processing Systems 6, NIPS 1993",

}

Fefferman, C & Markel, S 1993, Recovering a Feed-Forward Net From Its Output. in J Cowan, G Tesauro & J Alspector (eds), Advances in Neural Information Processing Systems 6, NIPS 1993. Advances in Neural Information Processing Systems, vol. 6, Neural information processing systems foundation, pp. 335-342, 6th Advances in Neural Information Processing Systems, NIPS 1993, Denver, United States, 11/29/93.

Recovering a Feed-Forward Net From Its Output. / Fefferman, Charles; Markel, Scott.
Advances in Neural Information Processing Systems 6, NIPS 1993. ed. / J. Cowan; G. Tesauro; J. Alspector. Neural information processing systems foundation, 1993. p. 335-342 (Advances in Neural Information Processing Systems; Vol. 6).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

TY - GEN

T1 - Recovering a Feed-Forward Net From Its Output

AU - Fefferman, Charles

AU - Markel, Scott

PY - 1993

Y1 - 1993

N2 - We study feed-forward nets with arbitrarily many layers, using the standard sigmoid, tanh x. Aside from technicalities, our theorems are: 1. Complete knowledge of the output of a neural net for arbitrary inputs uniquely specifies the architecture, weights and thresholds; and 2. There are only finitely many critical points on the error surface for a generic training problem.

AB - We study feed-forward nets with arbitrarily many layers, using the standard sigmoid, tanh x. Aside from technicalities, our theorems are: 1. Complete knowledge of the output of a neural net for arbitrary inputs uniquely specifies the architecture, weights and thresholds; and 2. There are only finitely many critical points on the error surface for a generic training problem.

UR - https://www.scopus.com/pages/publications/105021090472

UR - https://www.scopus.com/pages/publications/105021090472#tab=citedBy

M3 - Conference contribution

AN - SCOPUS:105021090472

T3 - Advances in Neural Information Processing Systems

SP - 335

EP - 342

BT - Advances in Neural Information Processing Systems 6, NIPS 1993

A2 - Cowan, J.

A2 - Tesauro, G.

A2 - Alspector, J.

PB - Neural information processing systems foundation

T2 - 6th Advances in Neural Information Processing Systems, NIPS 1993

Y2 - 29 November 1993 through 2 December 1993

ER -

Recovering a Feed-Forward Net From Its Output

Abstract

Publication series

Conference

All Science Journal Classification (ASJC) codes

Other files and links

Fingerprint

Cite this