Authoritative citation KNN learning in multiple-instance problems

Joseph Bernadt, Leen Kiat Soh

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

In this paper, we propose an authoritative citation K-nearest neighbor (ACKNN) algorithm for learning and classification in multiple-instance problems. We devise an authority measure for each instance or each bag of instances. This authority measure records how well an instance or a bag of instances has contributed to a correct classification, thus documenting how well an instance or a bag has been cited as a nearest neighbor. Based on our experiments with the Musk1 and Musk2 datasets, by learning the authority measures, the ACKNN algorithm outperforms most other algorithms in Musk1 classification accuracy, but only performs reasonably well in Musk2 classification accuracy.

Original languageEnglish (US)
Title of host publicationProceedings of the 2004 International Conference on Machine Learning and Applications, ICMLA '04
EditorsM. Kantardzic, O. Nasraoui, M. Milanova
Pages410-417
Number of pages8
StatePublished - Dec 1 2004
Event2004 International Conference on Machine Learning and Applications, ICMLA '04 - Louisville, KY, United States
Duration: Dec 16 2004Dec 18 2004

Publication series

NameProceedings of the 2004 International Conference on Machine Learning and Applications, ICMLA '04

Conference

Conference2004 International Conference on Machine Learning and Applications, ICMLA '04
CountryUnited States
CityLouisville, KY
Period12/16/0412/18/04

    Fingerprint

ASJC Scopus subject areas

  • Engineering(all)

Cite this

Bernadt, J., & Soh, L. K. (2004). Authoritative citation KNN learning in multiple-instance problems. In M. Kantardzic, O. Nasraoui, & M. Milanova (Eds.), Proceedings of the 2004 International Conference on Machine Learning and Applications, ICMLA '04 (pp. 410-417). (Proceedings of the 2004 International Conference on Machine Learning and Applications, ICMLA '04).