Neyman Seminar — Shyue Ong

Machine learning (ML) models have demonstrated human, or even superhuman, performance in many tasks. From AlphaFold to self-driving cars, ML models have delivered solutions to once-intractable scientific problems and made possible technologies that once seemed the realm of science fiction. In this talk, I will discuss how we can harness ML to achieve a similar transformative impact in materials science by transcending the trade-off between accuracy and performance in traditional in silico methods. I will highlight our recent advances in sophisticated graph deep learning models for materials property predictions. Using such models, we have explored vast, diverse chemical spaces and discovered novel materials for rechargeable lithium-ion batteries (LIBs) and aerospace applications. I will also discuss the application of ML to construct accurate interatomic potentials. Using ML potentials, we have simulated complex materials at unprecedented accuracies across large time and length scales. These simulations have provided new insights into the diffusion and reactivity in solid electrolytes for LIBs and chemical short-range order in high entropy alloys. Finally, I will share my perspectives on the path forward for ML in materials science.