Timezone: »

Learning with Transformation Invariant Kernels
Christian Walder · Olivier Chapelle

Mon Dec 03 08:10 PM -- 08:25 PM (PST) @ None

This paper considers kernels invariant to translation, rotation and dilation. We show that no non-trivial positive definite (p.d.) kernels exist which are radial and dilation invariant, only conditionally positive definite (c.p.d.) ones. Accordingly, we discuss the c.p.d. case and provide some novel analysis, including an elementary derivation of a c.p.d. representer theorem. On the practical side, we give a support vector machine (s.v.m.) algorithm for arbitrary c.p.d. kernels. For the thinplate kernel this leads to a classifier with only one parameter (the amount of regularisation), which we demonstrate to be as effective as an s.v.m. with the Gaussian kernel, even though the Gaussian involves a second parameter (the length scale).

Author Information

Christian Walder (Max Planck Institute)
Olivier Chapelle (Google)

Related Events (a corresponding poster, oral, or spotlight)

More from the Same Authors