Ji Qi gave a tutorial talk on “Machine Learning and High-Throughput Discovery and Design of Next Generation Electrode and Superionic Materials and Their Interfaces for SSBs” at the MRS Fall 2023! This tutorial provides an overview of how our group is using ML techniques to gain insights and discovery alkali superionic conductors, as well as the many open-source software packages that we have developed for these purposes. A recording of this talk is available on our group’s YouTube channel (and embedded above).
Prof Ong gave an invited seminar talk at the National University of Singapore on Jul 5 2023. In this talk, Prof Ong discusses the different ways in which machine learning (ML) can be used to improve or accelerate the various steps of in silico materials design. The general goal is to preserve the universality and accuracy of ab initio approaches as far as possible while achieving orders of magnitude speed-ups and improved scaling. Prof Ong shared his view that graph deep learning models trained on large diverse materials datasets, such as the M3GNet universal potential, are the “foundation” models for materials science. He further argues that the most robust approach is to replace the smallest, most expensive step in the materials design workflow with ML and preserve as much as the physics of thermodynamics, kinetics, etc. in the computation of materials properties.
Graphs are a natural way to represent atoms and bonds. In this lecture titled “Mathematical Graphs as a Representation for Materials”, Prof Shyue Ping Ong introduces the basics of graph deep learning and its application in materials science. MatErials Graph Networks (MEGNet) models have immense flexibility and expressiveness that can be adapted to datasets of diverse quality and quantity. We also demonstrate how the application of simple principles like energy minimization or interatomic development with materials graph models with 3-body interactions (M3GNet) can be used in the discovery of new materials **without** ab initio calculations, paving the way for massive-scale computational materials design. Prof Ong also introduces the matterverse.ai initiative, an open initiative to use ML to greatly expand the explorable matterverse. This lecture also includes two hands-on tutorials using Google Colab to demonstrate key concepts and the application of MEGNet and M3GNet models for property predictions and crystal structure relaxation. This Lecture is part of the Lawrence Livermore National Laboratory (LLNL) Computational Chemistry & Materials Science (CCMS) Summer Institute held from June 6 to August 12, 2022. The program offers graduate students the opportunity to work directly with leading LLNL researchers on the development and application of cutting-edge methods […]
Prof Ong gave a talk on Design of Alkali Superionic Conductors with ML at the MRS Spring 2022 meeting. In this talk, he highlights how new developments in using ML to construct interatomic potentials can greatly extend the accessible time and length scales in simulations, which can in turn lead to improved accuracy and new insights into the behavior of alkali superionic conductors. He also talks about the development of a universal graph potential for the periodic table utilizing data on structural relaxations performed by the Materials Project over the past 10 years.
Prof Ong gave a talk on “Addressing Errors in Ab Initio Molecular Dynamics Predictions through Machine Learning” at the Electronic Structure 21 Workshop held on Jul 14 2021. In this talk, we discuss the errors that AIMD simulations make in predicting the properties of lithium superionic conductors, which can be traced to the small simulation cell sizes, high temperatures and short simulation times. We then demonstrate how machine learning can be used to construct highly accurate interatomic potentials (ML-IAPs) to enable long time scale, large system simulations of complex materials. In particular, a standardized approach to construct ML-IAPs for lithium superionic conductors such as Li7P3S11, Li3YCl6 and Li0.33La0.56TiO3 (LLTO) is shown. All three superionic conductor solid electrolytes exhibit a transition between Arrhenius regimes are relatively low temperatures due to a change in diffusion dimensionality.
Prof Ong gave a talk on bridging computational and experimental predictions in materials machine learning models at the CNLS Virtual Workshop Machine Learning in Chemical and Materials Sciences held on May 12 2021. This talk discusses the sources of ML prediction errors, namely model errors and data errors, and demonstrate how these errors can be mitigated using appropriate techniques. For example, multi-fidelity/multi-task models can help small data models learn from larger, less accurate data models, while choosing an appropriate DFT functional for computing energies and forces for ML interatomic potentials can significantly improve the agreement with experimental measurements. You can jump to the relevant chapter of interest below!
Dr Chi Chen gave a talk at the Global XAS Journal Club on the Materials Virtual Lab’s efforts at constructing large X-ray absorption spectra databases using high-throughput computation and the development of machine learning models that can supercharge the interpretation of such spectra.
Prof Ong gave a talk on “Discovering New Materials in a Fraction of the Time with Graph Networks” at the NVIDIA GTC 2021 conference. This talk discusses our recent work on using GPU-trained multi-fidelity graph networks, together with Bayseian optimization techniques, to discover novel materials.
Chi Chen gave a talk at nanoHub’s Hands-on Data Science and Machine Learning Training Series on how to develop MatErials Graph Network (MEGNet) models for predicting various materials properties from crystal structure. He also demonstrates how the MEGNet framework can be adapted to work with multi-fidelity data sources to improve predictions on high-value small datasets (e.g., experimental data). Extensive examples are shown using Jupyter notebooks. The video is available on the Materials Virtual Lab Youtube Channel. The megnet package used extensively in these tutorials can be found on Github.
Yunxing gave a talk at NanoHUB’s Hands-on Data Science and Machine Learning Training Series today on how to conveniently develop machine learning interatomic potentials (ML-IAPs) using the Materials Machine Learning (maml) library. ML-IAPs describe the potential energy surface using local environment descriptors and has been demonstrated to be able to achieve near-DFT accuracy with linear scaling with respect to the number of atoms. The recording of this talk is now available on the Materials Virtual Lab’s Youtube channel. To find out more about the maml package, check out our Github repository. You can also read Yunxing’s excellent paper benchmarking the performance and cost of various ML-IAPs to learn more.