Relevance Preprocessing for Generalized Learning Vector Quantization Methods

van Wezel, Jelle (2018) Relevance Preprocessing for Generalized Learning Vector Quantization Methods. Master's Internship Report, Computing Science.

	Text Relevance Preprocessing for Generalized Learning Vector Quantization Methods - van Wezel, J 2018.tex Download (2kB)
	Text toestemming.pdf Restricted to Registered users only Download (95kB)

Abstract

Datasets often contain statistical regularities. These regularities can be used to improve the distance metric in Learning Vector Quantization (LVQ) with a relevance matrix. Methods like Global Metric Learning (GML) and Supervised Variational Relevance Learning (SUVREL) try to find this matrix. We tried to find if various LVQ methods would converge faster with a predetermined relevance matrix on multiple datasets. We found that the various LVQ methods did not converge noticeably faster with a predetermined matrix but achieved a better classification result on all but one of the tested datasets. Furthermore, a variant of SUVREL, called SUVREL Localized, is introduced and tested. It showed comparable performance to SUVREL and GML.

Item Type:	Thesis (Master's Internship Report)
Supervisor name:	Biehl, M. and Bunte, K.
Degree programme:	Computing Science
Thesis type:	Master's Internship Report
Language:	English
Date Deposited:	20 Aug 2018
Last Modified:	20 Aug 2018 12:51
URI:	https://fse.studenttheses.ub.rug.nl/id/eprint/18331

Actions (login required)

View Item