RNN with a recurrent output layer for learning of naturalness

Ján Dolinský; Hideyuki Takagi

doi:10.1007/978-3-540-69158-7_27

RNN with a recurrent output layer for learning of naturalness

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

Abstract

The behavior of recurrent neural networks with a recurrent output layer (ROL) is described mathematically and it is shown that using ROL is not only advantageous, but is in fact crucial to obtaining satisfactory performance for the proposed naturalness learning. Conventional belief holds that employing ROL often substantially decreases the performance of a network or renders the network unstable, and ROL is consequently rarely used. The objective of this paper is to demonstrate that there are cases where it is necessary to use ROL. The concrete example shown models naturalness in handwritten letters.

Original language	English
Title of host publication	Neural Information Processing - 14th International Conference, ICONIP 2007, Revised Selected Papers
Pages	248-257
Number of pages	10
Edition	PART 1
DOIs	https://doi.org/10.1007/978-3-540-69158-7_27
Publication status	Published - 2008
Event	14th International Conference on Neural Information Processing, ICONIP 2007 - Kitakyushu, Japan Duration: Nov 13 2007 → Nov 16 2007

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	11/13/07 → 11/16/07

All Science Journal Classification (ASJC) codes

Theoretical Computer Science
Computer Science(all)

Access to Document

10.1007/978-3-540-69158-7_27

Cite this

Dolinský, J., & Takagi, H. (2008). RNN with a recurrent output layer for learning of naturalness. In Neural Information Processing - 14th International Conference, ICONIP 2007, Revised Selected Papers (PART 1 ed., pp. 248-257). (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_27

RNN with a recurrent output layer for learning of naturalness. / Dolinský, Ján; Takagi, Hideyuki.
Neural Information Processing - 14th International Conference, ICONIP 2007, Revised Selected Papers. PART 1. ed. 2008. p. 248-257 (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

Dolinský, J & Takagi, H 2008, RNN with a recurrent output layer for learning of naturalness. 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. 248-257, 14th International Conference on Neural Information Processing, ICONIP 2007, Kitakyushu, Japan, 11/13/07. https://doi.org/10.1007/978-3-540-69158-7_27

Dolinský J, Takagi H. RNN with a recurrent output layer for learning of naturalness. In Neural Information Processing - 14th International Conference, ICONIP 2007, Revised Selected Papers. PART 1 ed. 2008. p. 248-257. (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_27

Dolinský, Ján ; Takagi, Hideyuki. / RNN with a recurrent output layer for learning of naturalness. Neural Information Processing - 14th International Conference, ICONIP 2007, Revised Selected Papers. PART 1. ed. 2008. pp. 248-257 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); PART 1).

@inproceedings{aa7744c3bad5465db28ce892c698b10d,

title = "RNN with a recurrent output layer for learning of naturalness",

abstract = "The behavior of recurrent neural networks with a recurrent output layer (ROL) is described mathematically and it is shown that using ROL is not only advantageous, but is in fact crucial to obtaining satisfactory performance for the proposed naturalness learning. Conventional belief holds that employing ROL often substantially decreases the performance of a network or renders the network unstable, and ROL is consequently rarely used. The objective of this paper is to demonstrate that there are cases where it is necessary to use ROL. The concrete example shown models naturalness in handwritten letters.",

author = "J{\'a}n Dolinsk{\'y} and Hideyuki Takagi",

year = "2008",

doi = "10.1007/978-3-540-69158-7_27",

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 = "248--257",

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 - RNN with a recurrent output layer for learning of naturalness

AU - Dolinský, Ján

AU - Takagi, Hideyuki

PY - 2008

Y1 - 2008

N2 - The behavior of recurrent neural networks with a recurrent output layer (ROL) is described mathematically and it is shown that using ROL is not only advantageous, but is in fact crucial to obtaining satisfactory performance for the proposed naturalness learning. Conventional belief holds that employing ROL often substantially decreases the performance of a network or renders the network unstable, and ROL is consequently rarely used. The objective of this paper is to demonstrate that there are cases where it is necessary to use ROL. The concrete example shown models naturalness in handwritten letters.

AB - The behavior of recurrent neural networks with a recurrent output layer (ROL) is described mathematically and it is shown that using ROL is not only advantageous, but is in fact crucial to obtaining satisfactory performance for the proposed naturalness learning. Conventional belief holds that employing ROL often substantially decreases the performance of a network or renders the network unstable, and ROL is consequently rarely used. The objective of this paper is to demonstrate that there are cases where it is necessary to use ROL. The concrete example shown models naturalness in handwritten letters.

UR - http://www.scopus.com/inward/record.url?scp=54249090825&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=54249090825&partnerID=8YFLogxK

U2 - 10.1007/978-3-540-69158-7_27

DO - 10.1007/978-3-540-69158-7_27

M3 - Conference contribution

AN - SCOPUS:54249090825

SN - 3540691545

SN - 9783540691549

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

SP - 248

EP - 257

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 -

RNN with a recurrent output layer for learning of naturalness

Abstract

Publication series

Other

All Science Journal Classification (ASJC) codes

Access to Document

Other files and links

Fingerprint

Cite this