An empirical study on budget-aware online kernel algorithms for streams of graphs

Giovanni Martino, Nicolò Navarin, Alessandro Sperduti

Research output: Contribution to journalArticle

1 Citation (Scopus)

Abstract

Kernel methods are considered as an effective technique for on-line learning. Many approaches have been developed for compactly representing the dual solution of a kernel method when the problem imposes memory constraints. However, in the literature no work is specifically tailored to streams of graphs. Motivated by the fact that the size of the feature space representation of many state-of-the-art graph kernels is relatively small and thus it is explicitly computable, we study whether executing kernel algorithms in the feature space can be more effective than the classical dual approach. We study three different algorithms and various strategies for managing the budget. Efficiency and efficacy of the proposed approaches are experimentally assessed on relatively large graph streams exhibiting concept drift. It turns out that, when strict memory budget constraints have to be enforced, working in feature space, given the current state-of-the-art on graph kernels, is more than a viable alternative to dual approaches, both in terms of speed and classification performance.

Original languageEnglish
Pages (from-to)163-182
Number of pages20
JournalNeurocomputing
Volume216
DOIs
Publication statusPublished - 5 Dec 2016

Fingerprint

Budgets
Data storage equipment
Learning
Efficiency

Keywords

  • Graph kernels
  • Graph streams
  • Online learning
  • Online passive aggressive

ASJC Scopus subject areas

  • Computer Science Applications
  • Cognitive Neuroscience
  • Artificial Intelligence

Cite this

An empirical study on budget-aware online kernel algorithms for streams of graphs. / Martino, Giovanni; Navarin, Nicolò; Sperduti, Alessandro.

In: Neurocomputing, Vol. 216, 05.12.2016, p. 163-182.

Research output: Contribution to journalArticle

Martino, Giovanni ; Navarin, Nicolò ; Sperduti, Alessandro. / An empirical study on budget-aware online kernel algorithms for streams of graphs. In: Neurocomputing. 2016 ; Vol. 216. pp. 163-182.
@article{24f9211b85ab4bbab648ce69014923cf,
title = "An empirical study on budget-aware online kernel algorithms for streams of graphs",
abstract = "Kernel methods are considered as an effective technique for on-line learning. Many approaches have been developed for compactly representing the dual solution of a kernel method when the problem imposes memory constraints. However, in the literature no work is specifically tailored to streams of graphs. Motivated by the fact that the size of the feature space representation of many state-of-the-art graph kernels is relatively small and thus it is explicitly computable, we study whether executing kernel algorithms in the feature space can be more effective than the classical dual approach. We study three different algorithms and various strategies for managing the budget. Efficiency and efficacy of the proposed approaches are experimentally assessed on relatively large graph streams exhibiting concept drift. It turns out that, when strict memory budget constraints have to be enforced, working in feature space, given the current state-of-the-art on graph kernels, is more than a viable alternative to dual approaches, both in terms of speed and classification performance.",
keywords = "Graph kernels, Graph streams, Online learning, Online passive aggressive",
author = "Giovanni Martino and Nicol{\`o} Navarin and Alessandro Sperduti",
year = "2016",
month = "12",
day = "5",
doi = "10.1016/j.neucom.2016.07.029",
language = "English",
volume = "216",
pages = "163--182",
journal = "Neurocomputing",
issn = "0925-2312",
publisher = "Elsevier",

}

TY - JOUR

T1 - An empirical study on budget-aware online kernel algorithms for streams of graphs

AU - Martino, Giovanni

AU - Navarin, Nicolò

AU - Sperduti, Alessandro

PY - 2016/12/5

Y1 - 2016/12/5

N2 - Kernel methods are considered as an effective technique for on-line learning. Many approaches have been developed for compactly representing the dual solution of a kernel method when the problem imposes memory constraints. However, in the literature no work is specifically tailored to streams of graphs. Motivated by the fact that the size of the feature space representation of many state-of-the-art graph kernels is relatively small and thus it is explicitly computable, we study whether executing kernel algorithms in the feature space can be more effective than the classical dual approach. We study three different algorithms and various strategies for managing the budget. Efficiency and efficacy of the proposed approaches are experimentally assessed on relatively large graph streams exhibiting concept drift. It turns out that, when strict memory budget constraints have to be enforced, working in feature space, given the current state-of-the-art on graph kernels, is more than a viable alternative to dual approaches, both in terms of speed and classification performance.

AB - Kernel methods are considered as an effective technique for on-line learning. Many approaches have been developed for compactly representing the dual solution of a kernel method when the problem imposes memory constraints. However, in the literature no work is specifically tailored to streams of graphs. Motivated by the fact that the size of the feature space representation of many state-of-the-art graph kernels is relatively small and thus it is explicitly computable, we study whether executing kernel algorithms in the feature space can be more effective than the classical dual approach. We study three different algorithms and various strategies for managing the budget. Efficiency and efficacy of the proposed approaches are experimentally assessed on relatively large graph streams exhibiting concept drift. It turns out that, when strict memory budget constraints have to be enforced, working in feature space, given the current state-of-the-art on graph kernels, is more than a viable alternative to dual approaches, both in terms of speed and classification performance.

KW - Graph kernels

KW - Graph streams

KW - Online learning

KW - Online passive aggressive

UR - http://www.scopus.com/inward/record.url?scp=84994387433&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84994387433&partnerID=8YFLogxK

U2 - 10.1016/j.neucom.2016.07.029

DO - 10.1016/j.neucom.2016.07.029

M3 - Article

VL - 216

SP - 163

EP - 182

JO - Neurocomputing

JF - Neurocomputing

SN - 0925-2312

ER -