In this paper lower and upper bounds for the number of support vectors are derived for support vector machines (SVMs) based on the epsilon-insensitive loss function. It turns out that these bounds are asymptotically tight under mild assumptions on the data generating distribution. Finally, we briefly discuss a trade-off in epsilono between sparsity and accuracy if the SVM is used to estimate the conditional median.
I. Steinwart and A. Christmann, Sparsity of SVMs that use the Epsilon-Insensitive Loss. In Neural Information Processing Systems 21, pp. 1569-1576, 2009. Los Alamos National Laboratory Technical Report LA-UR-08-3631. [ Abstract | PDF (202 KB) ]
