Correlative and causal ML augmented by theoretical and phenomenological models of expected system behaviors is used for efficient discovery, screening, and optimization of metal halide perovskites.
Scientific Achievement
Established a comprehensive framework for correlative and causal machine learning (ML)-based discovery and optimization of halide perovskites.
Significance and Impact
The proposed framework bridges the gap between instrumentation and control, leading to efficient automation of synthesis and characterization of metal halide perovskites (MHPs).
Research Details
- Bayesian optimization based on deep kernel learning augmented with phenomenological models of expected system behaviors leads to more efficient physics-based exploration of high-dimensional parameter spaces.
- Causal ML enables discovery of causal links between synthesis variables, material parameters, and physical functionalities.
Theory co-navigation: combined data-model uncertainty is minimized for more rapid optimization of MHP properties.
Mashid Ahmadi, Maxim Ziatdinov, Yuanyuan Zhou, Eric A. Lass, and Sergei V. Kalinin, "Machine Learning for High-Throughput Experimental Exploration of Metal Halide Perovskites," Joule (2021). DOI: 10.1016/j.joule.2021.10.001