Reward driven workflows for unsupervised explainable analysis of phases and ferroic variants from atomically resolved imaging data
Abstract
Rapid progress in aberration corrected electron microscopy necessitates development of robust methods for the identification of phases, ferroic variants, and other pertinent aspects of materials structure from imaging data. While unsupervised methods for clustering and classification are widely used for these tasks, their performance can be sensitive to hyperparameter selection in the analysis workflow. In this study, we explore the effects of descriptors and hyperparameters on the capability of unsupervised ML methods to distill local structural information, exemplified by discovery of polarization and lattice distortion in Sm doped BiFeO3 (BFO) thin films. We demonstrate that a reward-driven approach can be used to optimize these key hyperparameters across the full workflow, where rewards were designed to reflect domain wall continuity and straightness, ensuring that the analysis aligns with the material's physical behavior. This approach allows us to discover local descriptors that are best aligned with the specific physical behavior, providing insight into the fundamental physics of materials. We further extend the reward driven workflows to disentangle structural factors of variation via optimized variational autoencoder (VAE). Finally, the importance of well-defined rewards was explored as a quantifiable measure of success of the workflow.
AI-Generated Overview
-
Research Focus: The study investigates a reward-driven workflow for the unsupervised identification of phases and ferroic variants in materials through atomically resolved imaging data, particularly focusing on Sm-doped BiFeO₃ thin films.
-
Methodology: The researchers developed a reward-driven approach to optimize hyperparameters and descriptors within unsupervised machine learning methods for clustering and classification of imaging data, utilizing atomic column coordinates and Gaussian Mixture Models (GMM).
-
Results: The study successfully demonstrated that the reward-driven workflow can achieve robust segmentation of images, with findings aligning well with physical behaviors and the underlying material's characteristics. Variables, such as window sizes and the number of GMM components, were shown to significantly impact segmentation results.
-
Key Contribution(s): This research presents the first application of a reward-driven optimization concept in the context of phase and ferroic variant identification from atomically resolved STEM images, emphasizing the importance of carefully defined reward functions that reflect the physical behavior of materials.
-
Significance: The findings enhance the understanding of unsupervised analysis workflows in material science, providing a robust and explainable method for real-time data analytics that can significantly reduce the time and human bias associated with traditional data analysis processes.
-
Broader Applications: The proposed reward-driven workflow can be generalized and applied to a wide range of physics discovery tasks, potentially facilitating advancements in materials characterization and the development of autonomous microscopy operations.