Abstract

Even without complete information about the spatial coordinates of every cell in the body, we can still learn about its anatomy. This illustrates the broader idea of compressing complex information into key descriptors, a concept that we have proposed to extend to autonomous studies of functional nanomaterials. In recent years, interest in autonomous experimentation has expanded across a wide range of fields, from chemical synthesis to applied materials to drug design, and across diverse implementations, from self-driving las to closed-loop discovery platforms. In functional nanomaterials, however, a bottleneck on the path toward autonomy is the lack of knowledge of active units responsible for materials properties and performance. The relevant descriptors of the active units – structural, electronic, and geometric – are hidden within features of experimental data, such as spectra and scattering patterns. We have recently demonstrated that supervised machine learning methods can be used to extract the descriptors from data through inversion methods. I will present our recent results on descriptor-based optimization in operando experiments, where structural and functional data are measured simultaneously. Although the main emphasis will be on X-ray absorption spectroscopy and catalysis, I will also discuss how these approaches can be extended to other techniques and functionalities.