A characterization of simple recurrent neural networks with two hidden units as a language recognizer

Azusa Iwata; Yoshihisa Shinozawa; Akito Sakurai

doi:10.1007/978-3-540-69158-7_46

A characterization of simple recurrent neural networks with two hidden units as a language recognizer

Azusa Iwata, Yoshihisa Shinozawa, Akito Sakurai

Department of Industrial and Systems Engineering

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

Abstract

We give a necessary condition that a simple recurrent neural network with two sigmoidal hidden units to implement a recognizer of the formal language {a ⁿ b ⁿ | n > 0 } which is generated by a set of generating rules {S→aSb, S→ab } and show that by setting parameters so as to conform to the condition we get a recognizer of the language. The condition implies instability of learning process reported in previous studies. The condition also implies, contrary to its success in implementing the recognizer, difficulty of getting a recognizer of more complicated languages.

Original language	English
Title of host publication	Neural Information Processing - 14th International Conference, ICONIP 2007, Revised Selected Papers
Pages	436-445
Number of pages	10
Edition	PART 1
DOIs	https://doi.org/10.1007/978-3-540-69158-7_46
Publication status	Published - 2008
Event	14th International Conference on Neural Information Processing, ICONIP 2007 - Kitakyushu, Japan Duration: 2007 Nov 13 → 2007 Nov 16

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Number	PART 1
Volume	4984 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Other

Other	14th International Conference on Neural Information Processing, ICONIP 2007
Country/Territory	Japan
City	Kitakyushu
Period	07/11/13 → 07/11/16

ASJC Scopus subject areas

Theoretical Computer Science
Computer Science(all)

Access to Document

10.1007/978-3-540-69158-7_46

Cite this

Iwata, A., Shinozawa, Y., & Sakurai, A. (2008). A characterization of simple recurrent neural networks with two hidden units as a language recognizer. In Neural Information Processing - 14th International Conference, ICONIP 2007, Revised Selected Papers (PART 1 ed., pp. 436-445). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 4984 LNCS, No. PART 1). https://doi.org/10.1007/978-3-540-69158-7_46

A characterization of simple recurrent neural networks with two hidden units as a language recognizer. / Iwata, Azusa; Shinozawa, Yoshihisa; Sakurai, Akito.
Neural Information Processing - 14th International Conference, ICONIP 2007, Revised Selected Papers. PART 1. ed. 2008. p. 436-445 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 4984 LNCS, No. PART 1).

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

Iwata, A, Shinozawa, Y & Sakurai, A 2008, A characterization of simple recurrent neural networks with two hidden units as a language recognizer. in Neural Information Processing - 14th International Conference, ICONIP 2007, Revised Selected Papers. PART 1 edn, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), no. PART 1, vol. 4984 LNCS, pp. 436-445, 14th International Conference on Neural Information Processing, ICONIP 2007, Kitakyushu, Japan, 07/11/13. https://doi.org/10.1007/978-3-540-69158-7_46

Iwata A, Shinozawa Y, Sakurai A. A characterization of simple recurrent neural networks with two hidden units as a language recognizer. In Neural Information Processing - 14th International Conference, ICONIP 2007, Revised Selected Papers. PART 1 ed. 2008. p. 436-445. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); PART 1). doi: 10.1007/978-3-540-69158-7_46

Iwata, Azusa ; Shinozawa, Yoshihisa ; Sakurai, Akito. / A characterization of simple recurrent neural networks with two hidden units as a language recognizer. Neural Information Processing - 14th International Conference, ICONIP 2007, Revised Selected Papers. PART 1. ed. 2008. pp. 436-445 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); PART 1).

@inproceedings{6b886a342c8c49e28d95444db58195cf,

title = "A characterization of simple recurrent neural networks with two hidden units as a language recognizer",

abstract = "We give a necessary condition that a simple recurrent neural network with two sigmoidal hidden units to implement a recognizer of the formal language {a n b n | n > 0 } which is generated by a set of generating rules {S→aSb, S→ab } and show that by setting parameters so as to conform to the condition we get a recognizer of the language. The condition implies instability of learning process reported in previous studies. The condition also implies, contrary to its success in implementing the recognizer, difficulty of getting a recognizer of more complicated languages.",

author = "Azusa Iwata and Yoshihisa Shinozawa and Akito Sakurai",

year = "2008",

doi = "10.1007/978-3-540-69158-7_46",

language = "English",

isbn = "3540691545",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

number = "PART 1",

pages = "436--445",

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

edition = "PART 1",

note = "14th International Conference on Neural Information Processing, ICONIP 2007 ; Conference date: 13-11-2007 Through 16-11-2007",

}

TY - GEN

T1 - A characterization of simple recurrent neural networks with two hidden units as a language recognizer

AU - Iwata, Azusa

AU - Shinozawa, Yoshihisa

AU - Sakurai, Akito

PY - 2008

Y1 - 2008

N2 - We give a necessary condition that a simple recurrent neural network with two sigmoidal hidden units to implement a recognizer of the formal language {a n b n | n > 0 } which is generated by a set of generating rules {S→aSb, S→ab } and show that by setting parameters so as to conform to the condition we get a recognizer of the language. The condition implies instability of learning process reported in previous studies. The condition also implies, contrary to its success in implementing the recognizer, difficulty of getting a recognizer of more complicated languages.

AB - We give a necessary condition that a simple recurrent neural network with two sigmoidal hidden units to implement a recognizer of the formal language {a n b n | n > 0 } which is generated by a set of generating rules {S→aSb, S→ab } and show that by setting parameters so as to conform to the condition we get a recognizer of the language. The condition implies instability of learning process reported in previous studies. The condition also implies, contrary to its success in implementing the recognizer, difficulty of getting a recognizer of more complicated languages.

UR - http://www.scopus.com/inward/record.url?scp=54249088081&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=54249088081&partnerID=8YFLogxK

U2 - 10.1007/978-3-540-69158-7_46

DO - 10.1007/978-3-540-69158-7_46

M3 - Conference contribution

AN - SCOPUS:54249088081

SN - 3540691545

SN - 9783540691549

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 436

EP - 445

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

T2 - 14th International Conference on Neural Information Processing, ICONIP 2007

Y2 - 13 November 2007 through 16 November 2007

ER -

A characterization of simple recurrent neural networks with two hidden units as a language recognizer

Abstract

Publication series

Other

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this