Abstract

In the recent decade, materials science has seen a huge growth in available data from combinatorial experiments as well as high-throughput first principles calculations. With this data explosion, we now stand at the cusp where machine learning techniques can make meaningful predictions of many properties of materials almost instantaneously. In this talk, I will discuss the potentially transformative impact that this “instant” materials property prediction can have on materials research, from providing new chemistry insights that will greatly improve our ability to “guess” new materials with superior properties to accessing large length / time scales at near DFT accuracy. I will highlight some of the most promising machine learning approaches thus far, focusing, in particular, on techniques to address fundamental data size and diversity limitations in materials science. Finally, I will outline some of the key obstacles that still remain to ML-enabled materials science.