Physically constrained autoencoder-assisted Bayesian optimization for refinement of high-dimensional defect-sensitive single crystalline structure

•AI-accelerated structure refinement: This work combines physically constrained generative machine learning with Bayesian optimization to improve defect-sensitive crystal analysis.

•Sharper detection of subtle disorder: We demonstrate how to resolve small atomic defects and irregularities that conventional refinement can miss, linking structure more directly to materials function.

•New capability for AI-enabled materials discovery: Results advance MRSEC program goals by turning complex experimental data into more precise, defect-aware structural models.

•Workforce development through team science: Trained students across career stages, from undergraduate and REU researchers to graduate students, in collaborative, AI-driven materials research.