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

Summary: Data-driven materials discovery promises to accelerate the design of soft, polymeric and bio-inspired materials, but real materials data are often limited, heterogeneous and incomplete. In this talk, I will first introduce the basic idea of learning structure–property relationships in materials and explain why physical knowledge, uncertainty and interpretability are essential. I will then present my research on machine-learning methods for polymer glass transition, molecular-simulation data analysis, and experimental soft-matter data, including collaborations with academic and industrial partners. Finally, I will outline future plans at AMI: materials informatics for limited-data regimes, reconstruction of hidden structural ensembles from measurable properties, and uncertainty-aware workflows toward sustainable and functional soft materials.