{"id":"arxiv/2611.13795","url":"https://arxiv.org/abs/2611.13795","tags":["math.NA"],"score":0,"title":"Learning the Brownian Motion in Machine Learning for Parametric Parabolic PDEs: A Backward Quadrature Approach","domain":"math.NA","source":"arxiv","authors":["Qiang Han","Shaolin Ji","Yanan Qi"],"reviews":0,"subject":"math.NA","summary":"We propose a backward quadrature learning (BQL) method for solving high dimensional parametric semilinear parabolic partial differential equations (PDEs). To the best of our knowledge, this work is the first to extend the applicability of BSDE-based stochastic numerical methods to this class of PDEs. Specifically, inspired by the approach of learning the random variables (LRV)~\\cite{BeckerJentzenMullerWurstemberger2024}, BQL incorporates learnable quadrature for Brownian increments at each time level into the backward recursion of BSDEs, enabling more efficient layerwise approximation of…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-11-10","reactions":{}}