Accurate first principles data and their role in machine learning material models

In the era of AI-driven materials design, models and data are the two most frequently discussed elements. For models, there are two distinct strategies: prioritizing general-purpose models and prioritizing task-specific models. For data, there are also two strategies: prioritizing data scale and prioritizing data accuracy. In this talk, I will focus on exploring strategies to advance the calculation of molecules and materials toward the highest accuracy, including how to use deep learning methods to further improve the accuracy of first-principles calculations, how to optimize material machine learning force field models using small amounts of high-accuracy first principles data, and how to train complex and amorphous materials' generative models using limited but reliable experimental data. I will show accurate first principles deep learning quantum Monte Carlo results on various molecule and solid systems, as well as machine learning models and simulations of water and carbon materials.