Fast physics-informed nonparametric smoothing of space-time data

Alessandro Palummo, Politenico di Milano

Co-authors: Eleonora Arnone, Univesita degli Studi di Torino; Letizia Clementi, Fondazione IRCCS Istituto Neurologico Carlo Besta; Laura M. Sangalli, Politecnico di Milano

Abstract: We introduce a novel method for nonparametric smoothing of large space-time datasets observed over complex, high-dimensional domains. Our approach falls within the class of physics-informed statistical learning methods, and offers a significant improvement in computational efficiency with respect to existing techniques of the same class. The high computational efficiency makes the proposed method well-suited for analyzing massive datasets, such as those encountered in large-scale population-level studies. To demonstrate its effectiveness, we apply the method to the analysis of neuroimaging data collected via functional Magnetic Resonance Imaging (fMRI).