Clipping algorithms for solving the nearest point problem over reduced convex hulls

The nearest point problem (NPP), i.e., finding the closest points between two disjoint convex hulls, has two classical solutions, the Gilbert-Schlesinger-Kozinec (GSK) and Mitchell-Dem'yanov-Malozemov (MDM) algorithms. When the convex hulls do intersect, NPP has to be stated in terms of reduced convex hulls (RCHs), made up of convex pattern combinations whose coefficients are bound by a @m<1 value and that are disjoint for suitable @m. The GSK and MDM methods have recently been extended to solve NPP for RCHs using the particular structure of the extreme points of a RCH. While effective, their reliance on extreme points may make them computationally costly, particularly when applied in a kernel setting. In this work we propose an alternative clipped extension of classical MDM that results in a simpler algorithm with the same classification accuracy than that of the extensions already mentioned, but also with a much faster numerical convergence.

[1]  José R. Dorronsoro,et al.  Cycle-breaking acceleration of SVM training , 2009, Neurocomputing.

[2]  José R. Dorronsoro,et al.  On the Equivalence of the SMO and MDM Algorithms for SVM Training , 2008, ECML/PKDD.

[3]  S. Sathiya Keerthi,et al.  Computation of certain measures of proximity between convex polytopes: a complexity viewpoint , 1992, Proceedings 1992 IEEE International Conference on Robotics and Automation.

[4]  Jue Wang,et al.  A generalized S-K algorithm for learning v-SVM classifiers , 2004, Pattern Recognit. Lett..

[5]  Jue Wang,et al.  A general soft method for learning SVM classifiers with L1-norm penalty , 2008, Pattern Recognit..

[6]  S. Sathiya Keerthi,et al.  A fast iterative nearest point algorithm for support vector machine classifier design , 2000, IEEE Trans. Neural Networks Learn. Syst..

[7]  Andreu Català,et al.  Dual unification of bi-class support vector machine formulations , 2006, Pattern Recognit..

[8]  E. Gilbert.  An Iterative Procedure for Computing the Minimum of a Quadratic Form on a Convex Set , 1966 .

[9]  Andreu Català,et al.  Rapid and brief communication: Unified dual for bi-class SVM approaches , 2005 .

[10]  V. N. Malozemov,et al.  Finding the Point of a Polyhedron Closest to the Origin , 1974 .

[11]  Sergios Theodoridis,et al.  A Geometric Nearest Point Algorithm for the Efficient Solution of the SVM Classification Task , 2007, IEEE Transactions on Neural Networks.

[12]  Václav Hlavác,et al.  Simple Solvers for Large Quadratic Programming Tasks , 2005, DAGM-Symposium.

[13]  Jinbo Bi,et al.  A geometric approach to support vector regression , 2003, Neurocomputing.

[14]  Bernhard Schölkopf,et al.  A tutorial on support vector regression , 2004, Stat. Comput..

[15]  Kristin P. Bennett,et al.  Duality and Geometry in SVM Classifiers , 2000, ICML.

[16]  Václav Hlavác,et al.  An iterative algorithm learning the maximal margin classifier , 2003, Pattern Recognit..

[17]  Sergios Theodoridis,et al.  A geometric approach to Support Vector Machine (SVM) classification , 2006, IEEE Transactions on Neural Networks.

[18]  José R. Dorronsoro,et al.  A Common Framework for the Convergence of the GSK, MDM and SMO Algorithms , 2010, ICANN.