Error-correcting neural networks for two-dimensional curvature computation in the level-set method (Q2674272)

An error-neural-modeling-based strategy for approximating two-dimensional curvature in the level-set method is presented. As the authors mention this work is a revised version of their preliminary research presented in [\textit{L. Larios-Cárdenas} and \textit{F. Gibou}, SIAM J. Sci. Comput. 43, No. 3, A1754--A1779 (2021; Zbl 07364333)] and later extended in [\textit{L. Larios-Cárdenas} and \textit{F. Gibou}, J. Comput. Phys. 463, Article ID 111291, 20 p. (2022; Zbl 07536781)]. The paper is organized as follows. Section 1 is an introduction. In Section 2, the level set method is briefly reviewed. After that, a thorough description of the authors' strategy is given in Section 3, including inference, data-set generation, and preprocessing algorithms, alongside relevant technical details. Section 4 provides a series of experiments evaluating the curvature hybrid solver's efficacy and efficiency. Finally, in Section 5 a summary and a few pointers for future work are given. The authors state that now they are working on the three-dimensional extension of their error-correcting approach.