Analysis on bidirectional associative memories with multiplicative weight noise

Chi Sing Leung; Pui Fai Sum; Tien-Tsin Wong

doi:10.1007/978-3-540-69158-7_31

Analysis on bidirectional associative memories with multiplicative weight noise

Chi Sing Leung, Pui Fai Sum, Tien-Tsin Wong

Research output: Chapter in Book/Report/Conference proceeding › Conference Paper › Research › peer-review

Abstract

In neural networks, network faults can be exhibited in different forms, such as node fault and weight fault. One kind of weight faults is due to the hardware or software precision. This kind of weight faults can be modelled as multiplicative weight noise. This paper analyzes the capacity of a bidirectional associative memory (BAM) affected by multiplicative weight noise. Assuming that weights are corrupted by multiplicative noise, we study how many number of pattern pairs can be stored as fixed points. Since capacity is not meaningful without considering the error correction capability, we also present the capacity of a BAM with multiplicative noise when there are some errors in the input pattern. Simulation results have been carried out to confirm our derivations.

Original language	English
Title of host publication	Neural Information Processing - 14th International Conference, ICONIP 2007, Revised Selected Papers
Publisher	Springer
Pages	289-298
Number of pages	10
Edition	PART 1
ISBN (Print)	3540691545, 9783540691549
DOIs	https://doi.org/10.1007/978-3-540-69158-7_31
Publication status	Published - 2008
Externally published	Yes
Event	International Conference on Neural Information Processing 2007 - Kitakyushu, Japan Duration: 13 Nov 2007 → 16 Nov 2007 Conference number: 14th https://link.springer.com/book/10.1007/978-3-540-69158-7 (Proceedings)

Publication series

Name	Lecture Notes in Computer Science
Publisher	Springer
Number	PART 1
Volume	4984
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	International Conference on Neural Information Processing 2007
Abbreviated title	ICONIP 2007
Country/Territory	Japan
City	Kitakyushu
Period	13/11/07 → 16/11/07
Internet address	https://link.springer.com/book/10.1007/978-3-540-69158-7 (Proceedings)

Access to Document

10.1007/978-3-540-69158-7_31

Cite this

Leung, C. S., Sum, P. F., & Wong, T.-T. (2008). Analysis on bidirectional associative memories with multiplicative weight noise. In Neural Information Processing - 14th International Conference, ICONIP 2007, Revised Selected Papers (PART 1 ed., pp. 289-298). (Lecture Notes in Computer Science ; Vol. 4984, No. PART 1). Springer. https://doi.org/10.1007/978-3-540-69158-7_31

@inproceedings{921e70ebed5047c4895e140b697f9aee,

title = "Analysis on bidirectional associative memories with multiplicative weight noise",

abstract = "In neural networks, network faults can be exhibited in different forms, such as node fault and weight fault. One kind of weight faults is due to the hardware or software precision. This kind of weight faults can be modelled as multiplicative weight noise. This paper analyzes the capacity of a bidirectional associative memory (BAM) affected by multiplicative weight noise. Assuming that weights are corrupted by multiplicative noise, we study how many number of pattern pairs can be stored as fixed points. Since capacity is not meaningful without considering the error correction capability, we also present the capacity of a BAM with multiplicative noise when there are some errors in the input pattern. Simulation results have been carried out to confirm our derivations.",

author = "Leung, {Chi Sing} and Sum, {Pui Fai} and Tien-Tsin Wong",

year = "2008",

doi = "10.1007/978-3-540-69158-7_31",

language = "English",

isbn = "3540691545",

series = "Lecture Notes in Computer Science ",

publisher = "Springer",

number = "PART 1",

pages = "289--298",

booktitle = "Neural Information Processing - 14th International Conference, ICONIP 2007, Revised Selected Papers",

edition = "PART 1",

note = "International Conference on Neural Information Processing 2007, ICONIP 2007 ; Conference date: 13-11-2007 Through 16-11-2007",

url = "https://link.springer.com/book/10.1007/978-3-540-69158-7",

}

Leung, CS, Sum, PF & Wong, T-T 2008, Analysis on bidirectional associative memories with multiplicative weight noise. in Neural Information Processing - 14th International Conference, ICONIP 2007, Revised Selected Papers. PART 1 edn, Lecture Notes in Computer Science , no. PART 1, vol. 4984, Springer, pp. 289-298, International Conference on Neural Information Processing 2007, Kitakyushu, Japan, 13/11/07. https://doi.org/10.1007/978-3-540-69158-7_31

Analysis on bidirectional associative memories with multiplicative weight noise. / Leung, Chi Sing; Sum, Pui Fai; Wong, Tien-Tsin.
Neural Information Processing - 14th International Conference, ICONIP 2007, Revised Selected Papers. PART 1. ed. Springer, 2008. p. 289-298 (Lecture Notes in Computer Science ; Vol. 4984, No. PART 1).

Research output: Chapter in Book/Report/Conference proceeding › Conference Paper › Research › peer-review

TY - GEN

T1 - Analysis on bidirectional associative memories with multiplicative weight noise

AU - Leung, Chi Sing

AU - Sum, Pui Fai

AU - Wong, Tien-Tsin

N1 - Conference code: 14th

PY - 2008

Y1 - 2008

N2 - In neural networks, network faults can be exhibited in different forms, such as node fault and weight fault. One kind of weight faults is due to the hardware or software precision. This kind of weight faults can be modelled as multiplicative weight noise. This paper analyzes the capacity of a bidirectional associative memory (BAM) affected by multiplicative weight noise. Assuming that weights are corrupted by multiplicative noise, we study how many number of pattern pairs can be stored as fixed points. Since capacity is not meaningful without considering the error correction capability, we also present the capacity of a BAM with multiplicative noise when there are some errors in the input pattern. Simulation results have been carried out to confirm our derivations.

AB - In neural networks, network faults can be exhibited in different forms, such as node fault and weight fault. One kind of weight faults is due to the hardware or software precision. This kind of weight faults can be modelled as multiplicative weight noise. This paper analyzes the capacity of a bidirectional associative memory (BAM) affected by multiplicative weight noise. Assuming that weights are corrupted by multiplicative noise, we study how many number of pattern pairs can be stored as fixed points. Since capacity is not meaningful without considering the error correction capability, we also present the capacity of a BAM with multiplicative noise when there are some errors in the input pattern. Simulation results have been carried out to confirm our derivations.

UR - http://www.scopus.com/inward/record.url?scp=54249111959&partnerID=8YFLogxK

U2 - 10.1007/978-3-540-69158-7_31

DO - 10.1007/978-3-540-69158-7_31

M3 - Conference Paper

AN - SCOPUS:54249111959

SN - 3540691545

SN - 9783540691549

T3 - Lecture Notes in Computer Science

SP - 289

EP - 298

BT - Neural Information Processing - 14th International Conference, ICONIP 2007, Revised Selected Papers

PB - Springer

T2 - International Conference on Neural Information Processing 2007

Y2 - 13 November 2007 through 16 November 2007

ER -

Analysis on bidirectional associative memories with multiplicative weight noise

Abstract

Publication series

Conference

Access to Document

Other files and links

Cite this