Evolving memory cell structures for sequence learning

Justin Bayer; Daan Wierstra; Julian Togelius; Jürgen Schmidhuber

doi:10.1007/978-3-642-04277-5_76

Evolving memory cell structures for sequence learning

Justin Bayer, Daan Wierstra, Julian Togelius, Jürgen Schmidhuber

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

Abstract

Long Short-Term Memory (LSTM) is one of the best recent supervised sequence learning methods. Using gradient descent, it trains memory cells represented as differentiable computational graph structures. Interestingly, LSTM's cell structure seems somewhat arbitrary. In this paper we optimize its computational structure using a multi-objective evolutionary algorithm. The fitness function reflects the structure's usefulness for learning various formal languages. The evolved cells help to understand crucial features that aid sequence learning.

Original language	English (US)
Title of host publication	Artificial Neural Networks - ICANN 2009 - 19th International Conference, Proceedings
Pages	755-764
Number of pages	10
Edition	PART 2
DOIs	https://doi.org/10.1007/978-3-642-04277-5_76
State	Published - 2009
Event	19th International Conference on Artificial Neural Networks, ICANN 2009 - Limassol, Cyprus Duration: Sep 14 2009 → Sep 17 2009

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Number	PART 2
Volume	5769 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Other

Other	19th International Conference on Artificial Neural Networks, ICANN 2009
Country/Territory	Cyprus
City	Limassol
Period	9/14/09 → 9/17/09

ASJC Scopus subject areas

Theoretical Computer Science
General Computer Science

Access to Document

10.1007/978-3-642-04277-5_76

Cite this

Bayer, J., Wierstra, D., Togelius, J., & Schmidhuber, J. (2009). Evolving memory cell structures for sequence learning. In Artificial Neural Networks - ICANN 2009 - 19th International Conference, Proceedings (PART 2 ed., pp. 755-764). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 5769 LNCS, No. PART 2). https://doi.org/10.1007/978-3-642-04277-5_76

Evolving memory cell structures for sequence learning. / Bayer, Justin; Wierstra, Daan; Togelius, Julian et al.
Artificial Neural Networks - ICANN 2009 - 19th International Conference, Proceedings. PART 2. ed. 2009. p. 755-764 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 5769 LNCS, No. PART 2).

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

Bayer, J, Wierstra, D, Togelius, J & Schmidhuber, J 2009, Evolving memory cell structures for sequence learning. in Artificial Neural Networks - ICANN 2009 - 19th International Conference, Proceedings. PART 2 edn, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), no. PART 2, vol. 5769 LNCS, pp. 755-764, 19th International Conference on Artificial Neural Networks, ICANN 2009, Limassol, Cyprus, 9/14/09. https://doi.org/10.1007/978-3-642-04277-5_76

Bayer J, Wierstra D, Togelius J, Schmidhuber J. Evolving memory cell structures for sequence learning. In Artificial Neural Networks - ICANN 2009 - 19th International Conference, Proceedings. PART 2 ed. 2009. p. 755-764. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); PART 2). doi: 10.1007/978-3-642-04277-5_76

Bayer, Justin ; Wierstra, Daan ; Togelius, Julian et al. / Evolving memory cell structures for sequence learning. Artificial Neural Networks - ICANN 2009 - 19th International Conference, Proceedings. PART 2. ed. 2009. pp. 755-764 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); PART 2).

@inproceedings{a1685b2fc95f48fa9d6d0226793ee8ae,

title = "Evolving memory cell structures for sequence learning",

abstract = "Long Short-Term Memory (LSTM) is one of the best recent supervised sequence learning methods. Using gradient descent, it trains memory cells represented as differentiable computational graph structures. Interestingly, LSTM's cell structure seems somewhat arbitrary. In this paper we optimize its computational structure using a multi-objective evolutionary algorithm. The fitness function reflects the structure's usefulness for learning various formal languages. The evolved cells help to understand crucial features that aid sequence learning.",

author = "Justin Bayer and Daan Wierstra and Julian Togelius and J{\"u}rgen Schmidhuber",

year = "2009",

doi = "10.1007/978-3-642-04277-5_76",

language = "English (US)",

isbn = "3642042767",

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

number = "PART 2",

pages = "755--764",

booktitle = "Artificial Neural Networks - ICANN 2009 - 19th International Conference, Proceedings",

edition = "PART 2",

note = "19th International Conference on Artificial Neural Networks, ICANN 2009 ; Conference date: 14-09-2009 Through 17-09-2009",

}

TY - GEN

T1 - Evolving memory cell structures for sequence learning

AU - Bayer, Justin

AU - Wierstra, Daan

AU - Togelius, Julian

AU - Schmidhuber, Jürgen

PY - 2009

Y1 - 2009

N2 - Long Short-Term Memory (LSTM) is one of the best recent supervised sequence learning methods. Using gradient descent, it trains memory cells represented as differentiable computational graph structures. Interestingly, LSTM's cell structure seems somewhat arbitrary. In this paper we optimize its computational structure using a multi-objective evolutionary algorithm. The fitness function reflects the structure's usefulness for learning various formal languages. The evolved cells help to understand crucial features that aid sequence learning.

AB - Long Short-Term Memory (LSTM) is one of the best recent supervised sequence learning methods. Using gradient descent, it trains memory cells represented as differentiable computational graph structures. Interestingly, LSTM's cell structure seems somewhat arbitrary. In this paper we optimize its computational structure using a multi-objective evolutionary algorithm. The fitness function reflects the structure's usefulness for learning various formal languages. The evolved cells help to understand crucial features that aid sequence learning.

UR - http://www.scopus.com/inward/record.url?scp=70450190492&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=70450190492&partnerID=8YFLogxK

U2 - 10.1007/978-3-642-04277-5_76

DO - 10.1007/978-3-642-04277-5_76

M3 - Conference contribution

AN - SCOPUS:70450190492

SN - 3642042767

SN - 9783642042768

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

SP - 755

EP - 764

BT - Artificial Neural Networks - ICANN 2009 - 19th International Conference, Proceedings

T2 - 19th International Conference on Artificial Neural Networks, ICANN 2009

Y2 - 14 September 2009 through 17 September 2009

ER -

Evolving memory cell structures for sequence learning

Abstract

Publication series

Other

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this