Talk as part of a colloquium in Data-Driven Materials Discovery

Summary: Most materials have fixed compositions and are optimized through lengthy synthesis–test cycles. This empirical model is poorly suited to adaptive materials, whose functions emerge from molecular architecture, interfaces, and changing environments. By combining machine learning, multiscale simulation, and experiments, we can move from trial and error to predictive design. I will illustrate this shift through adaptive imaging and theragnostic; self-healing formulations and functional coatings; and digital discovery, where machine learning identifies structure–property relationships and proposes candidates. Intelligent experimental design, digital twins, and quantum–classical methods connect scales and resolve interfaces. This framework also shapes my teaching, training students to integrate theory, use data from experiment, computation, and data to design materials that sense, reorganize, and adapt