Machine Learning for Atomic-Scale Materials Modeling and Simulation

This workshop will introduce researchers to cutting-edge machine learning (ML) methods that are transforming computational chemistry and materials science. Participants will learn how ML models can accelerate atomistic simulations, bridging the gap between quantum-level accuracy and realistic system sizes. The session will highlight practical workflows, including training potentials on quantum mechanical data, developing ensembles of models for a real system, and identifying the most promising candidates for practical applications. Examples will focus on heterogeneous catalysis—such as modeling catalysts and identifying active sites—but the methodologies are broadly transferable across materials science and related fields.

Target Audience: UCLA graduate students, postdoctoral researchers, and faculty in chemistry, chemical engineering, materials science, physics, and related disciplines.

Learning Outcomes:

a. Understand the “accuracy versus scale” challenge in computational modeling of complex materials.

b. Learn how neural network potentials can be trained on quantum mechanical data to achieve both high accuracy and scalability.

c. Explore how machine learning based data screening can be used to identify the most promising candidates within large ensembles of models.

d. Recognize UCLA resources (such as the Hoffman2 cluster) that can support ML-driven simulations in practice.

This workshop is hosted by IDRE Fellow, Dr. Dongxiao Chen.