Variance Stable Gaussian Smoothing Estimator for Arbitrary High-order Derivatives
Shuixin Fang, Xuerui Wang, Hang Zeng, Tao Zhou
Abstract
Computing high-order partial derivatives in high dimensions by automatic differentiation can be costly in both computation and memory. Stein's identity offers an alternative through Gaussian smoothing, but the variance of the resulting estimators can diverge as the step size tends to zero, even with antithetic sampling. We propose a variance stabilization technique that extends antithetic sampling from a reflection pair to n equally spaced points on a circle in the complex plane. Under local analyticity, we prove that, when n is at least the derivative order, the bias is of order O(σn) and the variance remains bounded as the smoothing step size σ tends to zero. Numerical experiments support these predictions and demonstrate the effectiveness of the estimator in physics-informed neural networks (PINNs) for polyharmonic and Cahn--Hilliard equations. The technique mitigates the bias-variance trade-off in smoothing-based derivative estimation and is useful for PINNs solving high-order partial differential equations.
Create a lesson
Related papers
Stochastic counterpart of Nesterov's method for sampling and provable non-acceleration of samplers with heavy-ball structure
A. Sharma
A time-domain adversarial example demonstrating the failure of H1 elements in a-ϕ electromagnetic simulation
Timothy Carlson, Sanjay Govindjee
Signed QoI Gaps for Goal-Oriented Error Estimation: Reliability and Sharpness from the Same First-Order System Residuals
Shun Zhang
Perturbation Algorithm and Iterative Realization of a Four-Layer Semi-Discrete Scheme for an Evolutionary Problem with a Variable Operator
David Gulua, Jemal Rogava, Zurab Vashakidze
Regularity transfer for the Ritz projection
Lars Diening, Julian Rolfes, Abner J. Salgado
High-Order Area-Preserving Runge--Kutta Methods for Surface Diffusion Flow of Curves
Lei Cui, Gengen Zhang