M. Opper and O. Winther (2000)
Gaussian Processes and SVM: Mean Field and Leave-One-Out
In: Advances in Large Margin Classifiers, ed. by A.J. Smola and P.L. Bartlett and B. Schölkopf and D. Schuurmans, pp. 311-326, Cambridge, MA, MIT Press.
This paper gives an overview of the connections between Gaussian Processes and SV machines and the implications for cost functions and corresponding probabilistic settings. The authors use the 'naive mean field' approximation from Statistical Mechanics to provide estimates on the leave-one-out error in kernel methods which are fast to compute and in very good agreement with the true leave-one-out error. Experimental results corroborate this finding.