Comparing the latent features of universal machine-learning interatomic potentials

Abstract

The past few years have seen the development of ``universal'' machine-learning interatomic potentials (uMLIPs) capable of approximating the ground-state potential energy surface across a wide range of chemical structures and compositions with reasonable accuracy. While these models differ in the architecture and the dataset used, they share the ability to compress a staggering amount of chemical information into descriptive latent features. Herein, we systematically analyze what the different uMLIPs have learned by quantitatively assessing the relative information content of their latent features with feature reconstruction errors, and observing how the trends are affected by the choice of training set and training protocol. We find that uMLIPs encode the chemical space in significantly distinct ways, with substantial cross-model feature reconstruction errors. When variants of the same model architecture are considered, trends become dependent on the dataset, target, and training protocol of choice. We also observe that fine-tuning of a uMLIP retains a strong pre-training bias in the latent features. Finally, we discuss how atom-level features, which are directly output by MLIPs, can be compressed into global structure-level features via concatenation of progressive cumulants, each adding significantly new information about the variability across the atomic environments within a given system.

References in corpus (12)

• The Open Catalyst 2020 (OC20) Dataset and Community Challenges
• Materials Cartography: Representing and Mining Material Space Using Structural and Electronic Fingerprints
• A practical guide to machine learning interatomic potentials -- Status and future
• Learning the electronic density of states in condensed matter
• DeePMD-kit v3: A Multiple-Backend Framework for Machine Learning Potentials
• Fast Uncertainty Estimates in Deep Learning Interatomic Potentials
• Ranking the information content of distance measures
• Bayesian Force Fields from Active Learning for Simulation of Inter-Dimensional Transformation of Stanene
• Thermal conductivity of Li${}_3$PS${}_4$ solid electrolytes with ab initio accuracy
• Exploring the design space of machine-learning models for quantum chemistry with a fully differentiable framework
• Adaptive energy reference for machine-learning models of the electronic density of states
• Reconstructions and Dynamics of $\beta$-Lithium Thiophosphate Surfaces