Researchers have developed a new method to improve nonstationary Gaussian processes (GPs) by modeling spatial deformations as a function of covariates. This approach addresses the limitations of static methods that cannot predict GP behavior under changing covariate conditions. The proposed technique connects deformation spaces and covariate vectors by representing deformations as generated by velocity fields, with a method to truncate high-order interactions for practical estimation. An efficient algorithm is provided for out-of-sample prediction, demonstrated on simulations and case studies in manufacturing and geostatistics. AI
Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →
IMPACT Enhances predictive modeling for spatial data influenced by covariates, potentially improving applications in fields like manufacturing and geostatistics.
RANK_REASON This is a research paper published on arXiv detailing a new methodology for nonstationary Gaussian processes.