Self-organizing maps for structured domains: Theory, models, and learning of kernels

Fabio Aiolli, Giovanni Martino, Markus Hagenbuchner, Alessandro Sperduti

Research output: Chapter in Book/Report/Conference proceedingChapter

2 Citations (Scopus)

Abstract

Self-Organizing Maps (SOMs) are a form of Machine Learning methods which are popularly applied as a tool to either cluster vectorial information, or to produce a topology preserving projection of high dimensional data vectors onto a low dimensional (often 2-dimensional) display space [20]. A SOM is generally trained unsupervised. The computational complexity of the underlying algorithms grows linearly with the size and number of inputs, which renders the SOM suitable for data mining tasks. The standard SOM algorithm is defined on input domains involving fixed sized data vectors. It is however recognized that many problem domains are naturally represented by structured data which are more complex than fixed sized vectors. Just to give some examples, in speech recognition, data is available in the form of variable length temporal vectors, while in Chemistry data is most naturally represented through molecular graphs.Moreover, numerous data mining tasks provide structural information which may be important to consider during the processing. For example, document mining in the world wide web involves both inter-document structure due to the formatting or hypertext structure, and intra-document structure due to hyperlink or reference dependencies. Note that any model capable of dealing with graphs can be used also in applications involving vectors, sequences, and trees, since these are special cases of graphs.

Original languageEnglish
Title of host publicationStudies in Computational Intelligence
Pages9-42
Number of pages34
Volume247
DOIs
Publication statusPublished - 2009
Externally publishedYes

Publication series

NameStudies in Computational Intelligence
Volume247
ISSN (Print)1860949X

Fingerprint

Self organizing maps
Data mining
Speech recognition
World Wide Web
Learning systems
Computational complexity
Display devices
Topology
Processing

ASJC Scopus subject areas

  • Artificial Intelligence

Cite this

Aiolli, F., Martino, G., Hagenbuchner, M., & Sperduti, A. (2009). Self-organizing maps for structured domains: Theory, models, and learning of kernels. In Studies in Computational Intelligence (Vol. 247, pp. 9-42). (Studies in Computational Intelligence; Vol. 247). https://doi.org/10.1007/978-3-642-04003-0_2

Self-organizing maps for structured domains : Theory, models, and learning of kernels. / Aiolli, Fabio; Martino, Giovanni; Hagenbuchner, Markus; Sperduti, Alessandro.

Studies in Computational Intelligence. Vol. 247 2009. p. 9-42 (Studies in Computational Intelligence; Vol. 247).

Research output: Chapter in Book/Report/Conference proceedingChapter

Aiolli, F, Martino, G, Hagenbuchner, M & Sperduti, A 2009, Self-organizing maps for structured domains: Theory, models, and learning of kernels. in Studies in Computational Intelligence. vol. 247, Studies in Computational Intelligence, vol. 247, pp. 9-42. https://doi.org/10.1007/978-3-642-04003-0_2
Aiolli F, Martino G, Hagenbuchner M, Sperduti A. Self-organizing maps for structured domains: Theory, models, and learning of kernels. In Studies in Computational Intelligence. Vol. 247. 2009. p. 9-42. (Studies in Computational Intelligence). https://doi.org/10.1007/978-3-642-04003-0_2
Aiolli, Fabio ; Martino, Giovanni ; Hagenbuchner, Markus ; Sperduti, Alessandro. / Self-organizing maps for structured domains : Theory, models, and learning of kernels. Studies in Computational Intelligence. Vol. 247 2009. pp. 9-42 (Studies in Computational Intelligence).
@inbook{b64723c40de644d687af2d4b13145192,
title = "Self-organizing maps for structured domains: Theory, models, and learning of kernels",
abstract = "Self-Organizing Maps (SOMs) are a form of Machine Learning methods which are popularly applied as a tool to either cluster vectorial information, or to produce a topology preserving projection of high dimensional data vectors onto a low dimensional (often 2-dimensional) display space [20]. A SOM is generally trained unsupervised. The computational complexity of the underlying algorithms grows linearly with the size and number of inputs, which renders the SOM suitable for data mining tasks. The standard SOM algorithm is defined on input domains involving fixed sized data vectors. It is however recognized that many problem domains are naturally represented by structured data which are more complex than fixed sized vectors. Just to give some examples, in speech recognition, data is available in the form of variable length temporal vectors, while in Chemistry data is most naturally represented through molecular graphs.Moreover, numerous data mining tasks provide structural information which may be important to consider during the processing. For example, document mining in the world wide web involves both inter-document structure due to the formatting or hypertext structure, and intra-document structure due to hyperlink or reference dependencies. Note that any model capable of dealing with graphs can be used also in applications involving vectors, sequences, and trees, since these are special cases of graphs.",
author = "Fabio Aiolli and Giovanni Martino and Markus Hagenbuchner and Alessandro Sperduti",
year = "2009",
doi = "10.1007/978-3-642-04003-0_2",
language = "English",
isbn = "9783642040023",
volume = "247",
series = "Studies in Computational Intelligence",
pages = "9--42",
booktitle = "Studies in Computational Intelligence",

}

TY - CHAP

T1 - Self-organizing maps for structured domains

T2 - Theory, models, and learning of kernels

AU - Aiolli, Fabio

AU - Martino, Giovanni

AU - Hagenbuchner, Markus

AU - Sperduti, Alessandro

PY - 2009

Y1 - 2009

N2 - Self-Organizing Maps (SOMs) are a form of Machine Learning methods which are popularly applied as a tool to either cluster vectorial information, or to produce a topology preserving projection of high dimensional data vectors onto a low dimensional (often 2-dimensional) display space [20]. A SOM is generally trained unsupervised. The computational complexity of the underlying algorithms grows linearly with the size and number of inputs, which renders the SOM suitable for data mining tasks. The standard SOM algorithm is defined on input domains involving fixed sized data vectors. It is however recognized that many problem domains are naturally represented by structured data which are more complex than fixed sized vectors. Just to give some examples, in speech recognition, data is available in the form of variable length temporal vectors, while in Chemistry data is most naturally represented through molecular graphs.Moreover, numerous data mining tasks provide structural information which may be important to consider during the processing. For example, document mining in the world wide web involves both inter-document structure due to the formatting or hypertext structure, and intra-document structure due to hyperlink or reference dependencies. Note that any model capable of dealing with graphs can be used also in applications involving vectors, sequences, and trees, since these are special cases of graphs.

AB - Self-Organizing Maps (SOMs) are a form of Machine Learning methods which are popularly applied as a tool to either cluster vectorial information, or to produce a topology preserving projection of high dimensional data vectors onto a low dimensional (often 2-dimensional) display space [20]. A SOM is generally trained unsupervised. The computational complexity of the underlying algorithms grows linearly with the size and number of inputs, which renders the SOM suitable for data mining tasks. The standard SOM algorithm is defined on input domains involving fixed sized data vectors. It is however recognized that many problem domains are naturally represented by structured data which are more complex than fixed sized vectors. Just to give some examples, in speech recognition, data is available in the form of variable length temporal vectors, while in Chemistry data is most naturally represented through molecular graphs.Moreover, numerous data mining tasks provide structural information which may be important to consider during the processing. For example, document mining in the world wide web involves both inter-document structure due to the formatting or hypertext structure, and intra-document structure due to hyperlink or reference dependencies. Note that any model capable of dealing with graphs can be used also in applications involving vectors, sequences, and trees, since these are special cases of graphs.

UR - http://www.scopus.com/inward/record.url?scp=70350223425&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=70350223425&partnerID=8YFLogxK

U2 - 10.1007/978-3-642-04003-0_2

DO - 10.1007/978-3-642-04003-0_2

M3 - Chapter

AN - SCOPUS:70350223425

SN - 9783642040023

VL - 247

T3 - Studies in Computational Intelligence

SP - 9

EP - 42

BT - Studies in Computational Intelligence

ER -