Javascript must be enabled for the correct page display

GPLVQ - Prototype-based Probabilistic Classification

Scheepens, M.C.P.M. (2017) GPLVQ - Prototype-based Probabilistic Classification. Master's Thesis / Essay, Computing Science.

thesis.pdf - Published Version

Download (684kB) | Preview
[img] Text
toestemming.pdf - Other
Restricted to Backend only

Download (81kB)


Learning Vector Quantization (LVQ) is a straight-forward machine learning algorithm that builds a classification model by placing prototypes in the feature space. The decision boundaries between different classes are a discontinuity, i.e. class membership can change promptly by slight movements around the decision boundary. In this thesis, a new derivative of LVQ is formulated which introduces probabilistic elements on the prototypes. This is achieved by using mechanisms of the Gaussian Process (GP) machine learning algorithm and combining those with LVQ, which yields the Gaussian Process Learning Vector Quantization (GPLVQ) algorithm. GPLVQ was tested on different data sets, where it showed a performance similar to GP and LVQ on simple datasets and yielded a fuzzy decision boundary by assigning probabilities for the class membership. However, the new algorithm inherited some problems from its ancestors. GPLVQ has the same weaknesses for high-dimensional data as GP. Furthermore, GPLVQ is also sensitive to underfitting when using too few prototypes like LVQ.

Item Type: Thesis (Master's Thesis / Essay)
Degree programme: Computing Science
Thesis type: Master's Thesis / Essay
Language: English
Date Deposited: 15 Feb 2018 08:28
Last Modified: 15 Feb 2018 08:28

Actions (login required)

View Item View Item