The article discusses the use of second-order solvers in scientific machine learning (SciML) for regression tasks. It introduces a software framework based on PETSc to bridge deep-learning software with conventional solvers. Empirical evidence shows the efficacy of trust region methods based on Gauss-Newton approximation in improving generalization errors. The content is structured into sections discussing the introduction, background, related work, contributions, and numerical results of various test cases.
Til et andet sprog
fra kildeindhold
arxiv.org
Vigtigste indsigter udtrukket fra
by Stefano Zamp... kl. arxiv.org 03-20-2024
https://arxiv.org/pdf/2403.12188.pdfDybere Forespørgsler