Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Locality Density-Based Fuzzy Multiple Empirical Kernel Learning

Locality Density-Based Fuzzy Multiple Empirical Kernel Learning Multiple Empirical Kernel Learning (MEKL) explicitly maps the samples to empirical feature spaces, in which the feature vectors of the mapped samples are explicitly presented. Thus with the explicit representation of the samples, almost all algorithms can be kernelized directly, which is much easier in processing and analyzing the structure of the empirical feature spaces. However, in conventional MEKL, samples are treated to belong to one exact class, and contribute the same importance to the decision surface. However, in many real-world applications, input samples may not be fully assigned to one class. MEKL suffers the instinct drawbacks in representing these samples. To overcome this problem, we assign a fuzzy membership to each mapped sample in each feature space and reformulate MEKL to the Fuzzy MEKL (FMEKL) in which different samples in different feature spaces can make different contributions to the learning of decision surface. Moreover, we propose a novel fuzzy membership evaluation approach named locality density-based fuzzy membership evaluation, which assign larger fuzzy membership to the samples with higher local density. Thus, FMEKL by adopting the locality density-based fuzzy membership evaluation is named as Locality Density-based Fuzzy Multiple Empirical Kernel Learning (LD-FMEKL). Experimental results on both balanced and imbalanced real-world datasets validate that LD-FMEKL outperforms the compared algorithms. The contributions of this work are: (i) reformulating traditional MEKL to fuzzy multiple empirical kernel learning; (ii) introducing an alternative locality density-based fuzzy membership evaluation approach; (iii) proposing the locality density-based FMEKL. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Neural Processing Letters Springer Journals

Locality Density-Based Fuzzy Multiple Empirical Kernel Learning

Neural Processing Letters , Volume 49 (3) – Jul 17, 2018

Loading next page...
 
/lp/springer-journals/locality-density-based-fuzzy-multiple-empirical-kernel-learning-Hf5u3DSUrO

References (41)

Publisher
Springer Journals
Copyright
Copyright © 2018 by Springer Science+Business Media, LLC, part of Springer Nature
Subject
Computer Science; Artificial Intelligence; Complex Systems; Computational Intelligence
ISSN
1370-4621
eISSN
1573-773X
DOI
10.1007/s11063-018-9881-x
Publisher site
See Article on Publisher Site

Abstract

Multiple Empirical Kernel Learning (MEKL) explicitly maps the samples to empirical feature spaces, in which the feature vectors of the mapped samples are explicitly presented. Thus with the explicit representation of the samples, almost all algorithms can be kernelized directly, which is much easier in processing and analyzing the structure of the empirical feature spaces. However, in conventional MEKL, samples are treated to belong to one exact class, and contribute the same importance to the decision surface. However, in many real-world applications, input samples may not be fully assigned to one class. MEKL suffers the instinct drawbacks in representing these samples. To overcome this problem, we assign a fuzzy membership to each mapped sample in each feature space and reformulate MEKL to the Fuzzy MEKL (FMEKL) in which different samples in different feature spaces can make different contributions to the learning of decision surface. Moreover, we propose a novel fuzzy membership evaluation approach named locality density-based fuzzy membership evaluation, which assign larger fuzzy membership to the samples with higher local density. Thus, FMEKL by adopting the locality density-based fuzzy membership evaluation is named as Locality Density-based Fuzzy Multiple Empirical Kernel Learning (LD-FMEKL). Experimental results on both balanced and imbalanced real-world datasets validate that LD-FMEKL outperforms the compared algorithms. The contributions of this work are: (i) reformulating traditional MEKL to fuzzy multiple empirical kernel learning; (ii) introducing an alternative locality density-based fuzzy membership evaluation approach; (iii) proposing the locality density-based FMEKL.

Journal

Neural Processing LettersSpringer Journals

Published: Jul 17, 2018

There are no references for this article.