Learning nonsparse kernels by self-organizing maps for structured data

Fabio Aiolli, Giovanni Martino, Markus Hagenbuchner, Alessandro Sperduti

Research output: Contribution to journalArticle

18 Citations (Scopus)

Abstract

The development of neural network (NN) models able to encode structured input, and the more recent definition of kernels for structures, makes it possible to directly apply machine learning approaches to generic structured data. However, the effectiveness of a kernel can depend on its sparsity with respect to a specific data set. In fact, the accuracy of a kernel method typically reduces as the kernel sparsity increases. The sparsity problem is particularly common in structured domains involving discrete variables which may take on many different values. In this paper, we explore this issue on two well-known kernels for trees, and propose to face it by recurring to self-organizing maps (SOMs) for structures. Specifically, we show that a suitable combination of the two approaches, obtained by defining a new class of kernels based on the activation map of a SOM for structures, can be effective in avoiding the sparsity problem and results in a system that can be significantly more accurate for categorization tasks on structured data. The effectiveness of the proposed approach is demonstrated experimentally on two relatively large corpora of XML formatted data and a data set of user sessions extracted from website logs.

Original languageEnglish
Article number5290054
Pages (from-to)1938-1949
Number of pages12
JournalIEEE Transactions on Neural Networks
Volume20
Issue number12
DOIs
Publication statusPublished - Dec 2009
Externally publishedYes

Fingerprint

Self organizing maps
Learning
Neural Networks (Computer)
XML
Learning systems
Websites
Chemical activation
Neural networks
Datasets
Machine Learning

Keywords

  • Kernel methods
  • Self-organizing maps (SOMs)
  • Structured data
  • Tree kernels

ASJC Scopus subject areas

  • Artificial Intelligence
  • Computer Networks and Communications
  • Computer Science Applications
  • Software
  • Medicine(all)

Cite this

Learning nonsparse kernels by self-organizing maps for structured data. / Aiolli, Fabio; Martino, Giovanni; Hagenbuchner, Markus; Sperduti, Alessandro.

In: IEEE Transactions on Neural Networks, Vol. 20, No. 12, 5290054, 12.2009, p. 1938-1949.

Research output: Contribution to journalArticle

Aiolli, Fabio ; Martino, Giovanni ; Hagenbuchner, Markus ; Sperduti, Alessandro. / Learning nonsparse kernels by self-organizing maps for structured data. In: IEEE Transactions on Neural Networks. 2009 ; Vol. 20, No. 12. pp. 1938-1949.
@article{beb7a5d0915a493c99bf50dc5e48f503,
title = "Learning nonsparse kernels by self-organizing maps for structured data",
abstract = "The development of neural network (NN) models able to encode structured input, and the more recent definition of kernels for structures, makes it possible to directly apply machine learning approaches to generic structured data. However, the effectiveness of a kernel can depend on its sparsity with respect to a specific data set. In fact, the accuracy of a kernel method typically reduces as the kernel sparsity increases. The sparsity problem is particularly common in structured domains involving discrete variables which may take on many different values. In this paper, we explore this issue on two well-known kernels for trees, and propose to face it by recurring to self-organizing maps (SOMs) for structures. Specifically, we show that a suitable combination of the two approaches, obtained by defining a new class of kernels based on the activation map of a SOM for structures, can be effective in avoiding the sparsity problem and results in a system that can be significantly more accurate for categorization tasks on structured data. The effectiveness of the proposed approach is demonstrated experimentally on two relatively large corpora of XML formatted data and a data set of user sessions extracted from website logs.",
keywords = "Kernel methods, Self-organizing maps (SOMs), Structured data, Tree kernels",
author = "Fabio Aiolli and Giovanni Martino and Markus Hagenbuchner and Alessandro Sperduti",
year = "2009",
month = "12",
doi = "10.1109/TNN.2009.2033473",
language = "English",
volume = "20",
pages = "1938--1949",
journal = "IEEE Transactions on Neural Networks and Learning Systems",
issn = "2162-237X",
publisher = "IEEE Computational Intelligence Society",
number = "12",

}

TY - JOUR

T1 - Learning nonsparse kernels by self-organizing maps for structured data

AU - Aiolli, Fabio

AU - Martino, Giovanni

AU - Hagenbuchner, Markus

AU - Sperduti, Alessandro

PY - 2009/12

Y1 - 2009/12

N2 - The development of neural network (NN) models able to encode structured input, and the more recent definition of kernels for structures, makes it possible to directly apply machine learning approaches to generic structured data. However, the effectiveness of a kernel can depend on its sparsity with respect to a specific data set. In fact, the accuracy of a kernel method typically reduces as the kernel sparsity increases. The sparsity problem is particularly common in structured domains involving discrete variables which may take on many different values. In this paper, we explore this issue on two well-known kernels for trees, and propose to face it by recurring to self-organizing maps (SOMs) for structures. Specifically, we show that a suitable combination of the two approaches, obtained by defining a new class of kernels based on the activation map of a SOM for structures, can be effective in avoiding the sparsity problem and results in a system that can be significantly more accurate for categorization tasks on structured data. The effectiveness of the proposed approach is demonstrated experimentally on two relatively large corpora of XML formatted data and a data set of user sessions extracted from website logs.

AB - The development of neural network (NN) models able to encode structured input, and the more recent definition of kernels for structures, makes it possible to directly apply machine learning approaches to generic structured data. However, the effectiveness of a kernel can depend on its sparsity with respect to a specific data set. In fact, the accuracy of a kernel method typically reduces as the kernel sparsity increases. The sparsity problem is particularly common in structured domains involving discrete variables which may take on many different values. In this paper, we explore this issue on two well-known kernels for trees, and propose to face it by recurring to self-organizing maps (SOMs) for structures. Specifically, we show that a suitable combination of the two approaches, obtained by defining a new class of kernels based on the activation map of a SOM for structures, can be effective in avoiding the sparsity problem and results in a system that can be significantly more accurate for categorization tasks on structured data. The effectiveness of the proposed approach is demonstrated experimentally on two relatively large corpora of XML formatted data and a data set of user sessions extracted from website logs.

KW - Kernel methods

KW - Self-organizing maps (SOMs)

KW - Structured data

KW - Tree kernels

UR - http://www.scopus.com/inward/record.url?scp=72149131453&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=72149131453&partnerID=8YFLogxK

U2 - 10.1109/TNN.2009.2033473

DO - 10.1109/TNN.2009.2033473

M3 - Article

C2 - 19846372

AN - SCOPUS:72149131453

VL - 20

SP - 1938

EP - 1949

JO - IEEE Transactions on Neural Networks and Learning Systems

JF - IEEE Transactions on Neural Networks and Learning Systems

SN - 2162-237X

IS - 12

M1 - 5290054

ER -