Uncertainty Quantification by Ensemble Learning for Computational Optical Form Measurements

TL;DR

This paper demonstrates how ensemble learning can effectively quantify uncertainty in solving large-scale, nonlinear inverse problems in optical form measurements, enhancing the reliability of deep learning predictions in real-world applications.

Contribution

It extends a deep learning approach with ensemble methods to provide uncertainty quantification for optical form measurement inverse problems.

Findings

Ensemble methods improve trustworthiness of predictions.

Uncertainty quantification remains reliable with out-of-distribution errors.

Approach is effective on high-dimensional, real-world data.

Abstract

Uncertainty quantification by ensemble learning is explored in terms of an application from computational optical form measurements. The application requires to solve a large-scale, nonlinear inverse problem. Ensemble learning is used to extend a recently developed deep learning approach for this application in order to provide an uncertainty quantification of its predicted solution to the inverse problem. By systematically inserting out-of-distribution errors as well as noisy data the reliability of the developed uncertainty quantification is explored. Results are encouraging and the proposed application exemplifies the ability of ensemble methods to make trustworthy predictions on high dimensional data in a real-world application.