Multi-Axis and Multi-Vector Gradient Estimations: Using Multi-Sampled Complex Unit Vectors to Estimate Gradients of Real Functions

2023/11/30 by Ergün Akleman, Akleman, Ergun, Alan D. Freed +1
Engineering · #Advanced Numerical Analysis Techniques #FOS: Mathematics #Numerical Analysis (math.NA)

paper · pdf · doi:10.48550/arxiv.2312.00228

openalex publication_date 2023/11/30 · openalex created_date 2023/12/05 · openalex updated_date 2026/07/28

Abstract

In this preliminary study, we provide two methods for estimating the gradients of functions of real value. Both methods are built on derivative estimations that are calculated using the standard method or the Squire-Trapp method for any given direction. Gradients are computed as the average of derivatives in uniformly sampled directions. The first method uses a uniformly distributed set of axes that consists of orthogonal unit vectors that span the space. The second method only uses a uniformly distributed set of unit vectors. Both methods essentially minimize the error through an average of estimations to cancel error terms. Both methods are essentially a conceptual generalization of the method used to estimate normal fractal surfaces.

Related

• A Theoretical and Empirical Comparison of Gradient Approximations in\n Derivative-Free Optimization
• Generalizing Stochastic Smoothing for Differentiation and Gradient Estimation
• Nearest Neighbour Based Estimates of Gradients: Sharp Nonasymptotic Bounds and Applications
• Derivative Estimation from Coarse, Irregular, Noisy Samples: An MLE-Spline Approach
• Gradient Estimation with Simultaneous Perturbation and Compressive Sensing