Physics-Informed Automatic Differentiation for Single-Shot Nanoscale 3D Imaging in In Situ Transmission Electron Microscopy

Deepan Balakrishnan, Joel Yeo, Yixiao Yang, Tristan Bepler, N. Duane Loh

Keywords: Automatic Differentiation, Physics-Informed Neural Networks, Single-Shot 3D Reconstruction, Transmission Electron Microscopy (TEM), Computational Microscopy

TL;DR: We introduce a physics-informed automatic differentiation framework for single-shot 3D reconstruction in TEM and demonstrate that pop-out 3D metrology provides a meaningful initialization that prevents vanishing gradients and improves convergence.

Confirmation Of Submission Requirements: I submit an abstract. It uses the template provided on the submission page and is no longer than 2 pages.

Submission Number: 241