Adaptive margin support vector machines

Jason Weston, Ralf Herbrich
1999-01-01
Abstract:The study of classi cation learning has shown that algorithms which learn a real {valued function for classi cation can control their generalization error by making use of a quantity known as the margin (see Section??). Based on these results, learning machines which directly control the margin (eg SVMs, LP {SVMs) have been proven to be successful in classi cation learning (Mason and Bartlett, 1998; Vapnik, 1998; Smola, 1998). Moreover, it turned out to be favourable to formulate the decision functions in terms of a symmetric, positive semide nite, and square integrable function 39 k (;) referred to as a kernel (see Section??). The class of decision functions| also known as kernel classi ers (Smola, 1998; Jaakkola and Haussler, 1999)| is then given by1 decision function f (x)= m X i= 1 iyik (xi; x) 0:(1.1)
What problem does this paper attempt to address?