Supervised learning of local projection kernels

Mehmet Gonen, Ethem Alpaydn

Research output: Contribution to journalArticle

5 Citations (Scopus)

Abstract

We formulate a supervised, localized dimensionality reduction method using a gating model that divides up the input space into regions and selects the dimensionality reduction projection separately in each region. The gating model, the locally linear projections, and the kernel-based supervised learning algorithm which uses them in its kernels are coupled and their training is performed with an alternating optimization procedure. Our proposed local projection kernel projects a data instance into different feature spaces by using the local projection matrices, combines them with the gating model, and performs the dot product in the combined feature space. Empirical results on benchmark data sets for visualization and classification tasks validate the idea. The method is generalizable to regression estimation and novelty detection.

Original languageEnglish (US)
Pages (from-to)1694-1703
Number of pages10
JournalNeurocomputing
Volume73
Issue number10-12
DOIs
StatePublished - Jun 2010
Externally publishedYes

Fingerprint

Supervised learning
Learning
Benchmarking
Learning algorithms
Linear Models
Visualization

Keywords

  • Dimensionality reduction
  • Kernel machines
  • Local embedding
  • Subspace learning

ASJC Scopus subject areas

  • Artificial Intelligence
  • Computer Science Applications
  • Cognitive Neuroscience

Cite this

Supervised learning of local projection kernels. / Gonen, Mehmet; Alpaydn, Ethem.

In: Neurocomputing, Vol. 73, No. 10-12, 06.2010, p. 1694-1703.

Research output: Contribution to journalArticle

Gonen, Mehmet ; Alpaydn, Ethem. / Supervised learning of local projection kernels. In: Neurocomputing. 2010 ; Vol. 73, No. 10-12. pp. 1694-1703.
@article{8df44b8304a542589235a06761e27850,
title = "Supervised learning of local projection kernels",
abstract = "We formulate a supervised, localized dimensionality reduction method using a gating model that divides up the input space into regions and selects the dimensionality reduction projection separately in each region. The gating model, the locally linear projections, and the kernel-based supervised learning algorithm which uses them in its kernels are coupled and their training is performed with an alternating optimization procedure. Our proposed local projection kernel projects a data instance into different feature spaces by using the local projection matrices, combines them with the gating model, and performs the dot product in the combined feature space. Empirical results on benchmark data sets for visualization and classification tasks validate the idea. The method is generalizable to regression estimation and novelty detection.",
keywords = "Dimensionality reduction, Kernel machines, Local embedding, Subspace learning",
author = "Mehmet Gonen and Ethem Alpaydn",
year = "2010",
month = "6",
doi = "10.1016/j.neucom.2009.11.043",
language = "English (US)",
volume = "73",
pages = "1694--1703",
journal = "Neurocomputing",
issn = "0925-2312",
publisher = "Elsevier",
number = "10-12",

}

TY - JOUR

T1 - Supervised learning of local projection kernels

AU - Gonen, Mehmet

AU - Alpaydn, Ethem

PY - 2010/6

Y1 - 2010/6

N2 - We formulate a supervised, localized dimensionality reduction method using a gating model that divides up the input space into regions and selects the dimensionality reduction projection separately in each region. The gating model, the locally linear projections, and the kernel-based supervised learning algorithm which uses them in its kernels are coupled and their training is performed with an alternating optimization procedure. Our proposed local projection kernel projects a data instance into different feature spaces by using the local projection matrices, combines them with the gating model, and performs the dot product in the combined feature space. Empirical results on benchmark data sets for visualization and classification tasks validate the idea. The method is generalizable to regression estimation and novelty detection.

AB - We formulate a supervised, localized dimensionality reduction method using a gating model that divides up the input space into regions and selects the dimensionality reduction projection separately in each region. The gating model, the locally linear projections, and the kernel-based supervised learning algorithm which uses them in its kernels are coupled and their training is performed with an alternating optimization procedure. Our proposed local projection kernel projects a data instance into different feature spaces by using the local projection matrices, combines them with the gating model, and performs the dot product in the combined feature space. Empirical results on benchmark data sets for visualization and classification tasks validate the idea. The method is generalizable to regression estimation and novelty detection.

KW - Dimensionality reduction

KW - Kernel machines

KW - Local embedding

KW - Subspace learning

UR - http://www.scopus.com/inward/record.url?scp=77952554018&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=77952554018&partnerID=8YFLogxK

U2 - 10.1016/j.neucom.2009.11.043

DO - 10.1016/j.neucom.2009.11.043

M3 - Article

VL - 73

SP - 1694

EP - 1703

JO - Neurocomputing

JF - Neurocomputing

SN - 0925-2312

IS - 10-12

ER -