The content discusses the Kernel Multigrid (KMG) algorithm to improve Back-fitting convergence using sparse Gaussian Process Regression (GPR). It introduces Additive Gaussian Processes, Bayesian Back-fitting, and Kernel Packets. The article outlines the challenges of training additive GPs due to computational complexity and proposes KMG as a solution. It explains the theoretical basis, numerical experiments, and lower bounds for convergence rates.
Til et andet sprog
fra kildeindhold
arxiv.org
Vigtigste indsigter udtrukket fra
by Lu Zou,Liang... kl. arxiv.org 03-21-2024
https://arxiv.org/pdf/2403.13300.pdfDybere Forespørgsler