Robust spherical separation. (3rd June 2017)

• Record Type: Journal Article
• Authors: Astorino, A.; Bomze, I.; Fuduli, A.; Gaudioso, M.
• Is Part Of: Optimization. Volume 66:Number 6(2017)
• Page Start: 925
• Page End: 938
• Publication Date: 2017-06-03
• Subjects: Spherical separation -- robust classification -- non-smooth optimization; Mathematical optimization -- Periodicals; 519.7
• DOI: 10.1080/02331934.2017.1306743
• Languages: English
• ISSNs: 0233-1934

Abstract

Abstract: We propose a robust spherical separation technique aimed at separating two finite sets of points and . Robustness concerns the possibility to admit uncertainties and perturbations in the data-set, which may occur when the data are corrupted by noise or are influenced by measurement errors. In particular, starting from the standard spherical separation under the assumption of spherical uncertainty, we propose a model characterized by a non-convex non-differentiable objective function, which we minimize by means of a bundle-type algorithm. Quite promising numerical results are provided on small and large data-sets drawn from well-established test beds in literature.