Bauckhage, ChristianChristianBauckhageSifa, RafetRafetSifa2022-03-132022-03-132015https://publica.fraunhofer.de/handle/publica/392395We explore the idea of clustering according to extremal rather than to central data points. To this end, we introduce the notion of the maxoid of a data set and present an algorithm for k-maxoids clustering which can be understood as a variant of classical k-means clustering. Exemplary results demonstrate that extremal cluster prototypes are more distinctive and hence more interpretable than central ones.enconference paper