Mercer's theorem
Mercer's theorem is one of the most popular results of the work of
James Mercer. It is used for the
kernel trick, an important method in
machine learning.
Mercer's theorem states that any positive definite kernel K(x, y) can be expressed as a dot product in a high-dimensional space.
More specifically, if a kernel is positive semi-definite, i.e.,
then there exists a function whose image is in an
inner product space of possibly high dimension, such that