Comparing the Latent Features of Universal Machine‐Learning Interatomic Potentials
Open Access
- 22 March 2026
- journal article
- Published by Wiley in Advanced Intelligent Systems
- Vol. 8 (5)
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 (FREs), 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 FREs. 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 pretraining 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.Keywords
All Related Versions
Funding Information
- Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung (200020_214879)
- Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung (CRSII5_202296)
- National Center of Competence in Research Materials’ Revolution: Computational Design and Discovery of Novel Materials
- European Research Council (101001890‐FIAMMA)
This publication has 53 references indexed in Scilit:
- Exploring the Design Space of Machine Learning Models for Quantum Chemistry with a Fully Differentiable FrameworkJournal of Chemical Theory and Computation, 2025
- DeePMD-kit v3: A Multiple-Backend Framework for Machine Learning PotentialsJournal of Chemical Theory and Computation, 2025
- A practical guide to machine learning interatomic potentials – Status and futureCurrent Opinion in Solid State and Materials Science, 2025
- Data Generation for Machine Learning Interatomic Potentials and BeyondChemical Reviews, 2024
- Improving machine-learning models in materials science through large datasetsMaterials Today Physics, 2024
- Mechanism of Charge Transport in Lithium ThiophosphateChemistry of Materials, 2024
- Artificial Intelligence-Aided Mapping of the Structure–Composition–Conductivity Relationships of Glass–Ceramic Lithium Thiophosphate ElectrolytesChemistry of Materials, 2022
- Accurate and Numerically Efficient r2SCAN Meta-Generalized Gradient ApproximationThe Journal of Physical Chemistry Letters, 2020
- Mapping Materials and MoleculesAccounts of Chemical Research, 2020
- Principal covariates regressionChemometrics and Intelligent Laboratory Systems, 1992