Got this from the paper "Graph Kernels for Chemical Classification", by Ralaivola et. al, published in Neural Networks 2005.
It looks like, to prove a kernel, you need to show that the function is symmetric (ie, k(u,v) = k(v,u) ), that it's continuous, and that it is "a positive definite kernel" (ie, that the square n x n matrix K = (k(ui, uj))1 <= i, j <= n is positive semi-definite...ie, all its eigenvalues are non-negative).
This would make it a Mercer Kernel.
Friday, March 13, 2009
Wednesday, March 4, 2009
Subscribe to:
Comments (Atom)