{"id":"arxiv/2603.06380","url":"https://arxiv.org/abs/2603.06380","tags":["math.NA","cs.AI","cs.LG"],"score":0,"title":"Kinetic-based regularization: Learning spatial derivatives and PDE applications","domain":"math.NA","source":"arxiv","authors":["Abhisek Ganguly","Santosh Ansumali","Sauro Succi"],"reviews":0,"subject":"math.NA","summary":"Accurate estimation of spatial derivatives from discrete and noisy data is central to scientific machine learning and numerical solutions of PDEs. We extend kinetic-based regularization (KBR), a localized multidimensional kernel regression method with a single trainable parameter, to learn spatial derivatives with provable second-order accuracy in 1D. Two derivative-learning schemes are proposed: an explicit scheme based on the closed-form prediction expressions, and an implicit scheme that solves a perturbed linear system at the points of interest. The fully localized formulation enables…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-03-06","reactions":{}}