P. Ruj\'an and M. Marchand (2000)
Computing the Bayes Kernel Classifier
In: Advances in Large Margin Classifiers, ed. by A.J. Smola and P.L. Bartlett and B. Schölkopf and D. Schuurmans, pp. 329-348, Cambridge, MA, MIT Press.
In the so-called version space view of classification, the SVM solution of a separable learning problem corresponds to the center of the largest inscribable sphere in a polytope determined by the training examples. Statistically, however, it would be preferable to find a solution corresponding to the center of mass. Ruj\'an and Marchand propose a Billiard algorithm which, under the assumption of ergodicity, converges towards the latter.