(a) Piezoresponse amplitude image of ferroelectric thin film. Microscope measures spectra and presents them to researcher, who makes quantitative assessments about each spectrum. Data used to shape a target spectrum and subsequently (b) the AI algorithm finds regions of the sample that closely align with target spectrum.
Scientific Achievement
A human-AI collaborative workflow, Bayesian optimized active recommender system (BOARS), was developed and implemented to combine the best of human and AI agents toward material characterization and discovery.
Significance and Impact
BOARS enables accelerated learning of structure-property relationships aligned with the researcher’s goals and allows for new types of experiments previously out of reach.
Research Details
- Workflow finds structure-property relationships in a ferroelectric thin film with scanning probe techniques.
- In exploration stage, human improves the search direction with on-the-fly intervention and feedback; in later stage, AI increases the speed of learning towards the human-aligned goal
Arpan Biswas, Yongtao Liu, Nicole Creange, Yu-Chen Liu, Stephen Jesse, Jan-Chi Yang, Sergei Kalinin, Maxim Ziatdinov, and Rama Vasudevan, npj Computational Materials 10, 29 (2024).
Work conducted at the Center for Nanophase Materials Sciences
