Fix observation points and suppose , where the noise has zero expected value and covariance matrix . Use the unscaled objective with and . The kernel-ridge hat matrix is now . For ,
This is prediction error at the fixed observation points, with separate squared-bias and variance terms.
For eigenvalues of the kernel matrix,
The reproducing property and Bessel inequality represent with , including singular . Hence
The bound makes the opposing effects of the ridge explicit: it increases the displayed bias bound while reducing the variance bound.

Articles by others on the same topic (0)

There are currently no matching articles.