The covariance matrix of Green's functions and its application to machine learning
This is an incremental contribution that applies Green's function theory to machine learning regression, potentially benefiting researchers in mathematical machine learning.
The authors tackled regression by proposing a new algorithm based on Green's function theory, which uses a covariance matrix from normalized Green's functions to provide predictive distributions with confidence intervals.
In this paper, a regression algorithm based on Green's function theory is proposed and implemented. We first survey Green's function for the Dirichlet boundary value problem of 2nd order linear ordinary differential equation, which is a reproducing kernel of a suitable Hilbert space. We next consider a covariance matrix composed of the normalized Green's function, which is regarded as aprobability density function. By supporting Bayesian approach, the covariance matrix gives predictive distribution, which has the predictive mean $μ$ and the confidence interval [$μ$-2s, $μ$+2s], where s stands for a standard deviation.