Options
2020
Conference Paper
Title
Novelty Discovery with Kernel Minimum Enclosing Balls
Abstract
We introduce the idea of utilizing ensembles of Kernel Minimum Enclosing Balls to detect novel datapoints. To this end, we propose a novelty scoring methodology that is based on combining outcomes of the corresponding characteristic functions of a set of fitted balls. We empirically evaluate our model by presenting experiments on synthetic as well as real world datasets.