Information theoretic feature selection in multi-label data through composite likelihood

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

147 Downloads (Pure)

Abstract

In this paper we present a framework to unify information theoretic feature selection criteria for multi-label data. Our framework combines two different ideas; expressing multi-label decomposition methods as composite likelihoods and then showing how feature selection criteria can be derived by maximizing these likelihood expressions. Many existing criteria, until now proposed as heuristics, can be reproduced from a single basis under the proposed framework. Furthermore we can derive new problem-specific criteria by making different independence assumptions over the feature and label spaces. One such derived criterion is shown experimentally to outperform other approaches proposed in the literature on real-world datasets.
Original languageEnglish
Title of host publicationhost publication
Publication statusPublished - Aug 2014
EventStructural, Syntactic, and Statistical Pattern Recognition - SSPR 2014 - Joensuu, Finland
Duration: 20 Aug 201422 Aug 2014

Conference

ConferenceStructural, Syntactic, and Statistical Pattern Recognition - SSPR 2014
CityJoensuu, Finland
Period20/08/1422/08/14

Keywords

  • Multi-label;Feature selection; Composite likelhood; Information theory

Fingerprint

Dive into the research topics of 'Information theoretic feature selection in multi-label data through composite likelihood'. Together they form a unique fingerprint.

Cite this