Question
Asked 15th Apr, 2020
Can I directly obtain a few eigenvectos of a covariance matrix when the covariance function is known?
I am suffering memory problem. My problem has 1000000 unknowns (1000000 by 1000000 in terms of covariance matrix) and they are correlated by a certain relationship given by a covariance function. I want to solve the problem using PCA but, at first, I have to compose covariance matrix which takes enormous memory. If my covariance function has analytical form of k(x,x')=exp(-||x-x'||^2), I believe there should be analytical way of getting a few eigenvectors with higher eigenvalues (similar like Krylov-Schur algorithm) without composing covariance matrix at first. Thank you in advance! - Eungyu
Similar questions and discussions
Related Publications
Starting from the definition of generalized Riemannian space (GR(N)) [1], in which a non-symmetric basic tensor g(ij) is introduced, in the present paper a generalized Kahlerian space GK(2)N of the second kind is defined, as a GR(N) with almost complex structure F(i)(h), that is covariantly constant with respect to the second kind of covariant deri...
The formulation of covariant brackets on the space of solutions to a variational problem is analyzed in the framework of contact geometry. It is argued that the Poisson algebra on the space of functionals on fields should be read as a Poisson subalgebra within an algebra of functions equipped with a Jacobi bracket on a suitable contact manifold.
Photonuclear reaction research is of great interest to obtain information about the structure of nuclei. The investigation of structural effects requires certain insights into the reaction mechanisms, that have to be identified on the basis of the fundamental principles of covariance and gauge invariance. The major achievement of the chosen model i...