Abstract

Representing data in hyperbolic space can effectively capture latenthierarchical relationships. With the goal of enabling accurate classificationof points in hyperbolic space while respecting their hyperbolic geometry, weintroduce hyperbolic SVM, a hyperbolic formulation of support vector machineclassifiers, and elucidate through new theoretical work its connection to theEuclidean counterpart. We demonstrate the performance improvement of hyperbolicSVM for multi-class prediction tasks on real-world complex networks as well assimulated datasets. Our work allows analytic pipelines that take the inherenthyperbolic geometry of the data into account in an end-to-end fashion withoutresorting to ill-fitting tools developed for Euclidean space.

 

Quick Read (beta)