Evaluate Deep Learning Discrepancies Between High Fidelity and Low Fidelity Simulations

Utilization of deep learning (DL) models has shown promising results in extraction of structural information from complex, three-dimensional atomistic data obtained from Atom Probe Tomography (APT). The primary requirement for DL models is a substantial amount of training data. To overcome this limitation, we propose the use of simulated training data. However, an open question arises regarding the level of accuracy (fidelity) that the synthetic training data should possess.<br/>The objective of this research is to investigate the relationship between the fidelity of synthetic training data and the performance of the DL model when applied to real APT data. By varying the fidelity of the synthetic data, we assess the impact on the accuracy and reliability of the DL model's structural predictions. Through our findings, we aim to provide insights into the optimal fidelity requirements for synthetic training data to achieve accurate and robust results when analyzing real APT data. This research contributes to the advancement of DL-based approaches for extracting valuable structural information from atomistic data, enhancing our understanding of complex materials, and enabling more precise analysis in various scientific fields.