Robustness of Local Predictions in Atomistic Machine Learning Models

Abstract

Machine learning (ML) models for molecules and materials commonly rely on a decomposition of the global target quantity into local, atom-centered contributions. This approach is convenient from a computational perspective, enabling large-scale ML-driven simulations with a linear-scaling cost, and also allow for the identification and post-hoc interpretation of contributions from individual chemical environments and motifs to complicated macroscopic properties. However, even though there exist practical justifications for these decompositions, only the global quantity is rigorously defined, and thus it is unclear to what extent the atomistic terms predicted by the model can be trusted. Here, we introduce a quantitative metric, which we call the local prediction rigidity (LPR), that allows one to assess how robust the locally decomposed predictions of ML models are. We investigate the dependence of LPR on the aspects of model training, particularly the composition of training dataset, for a range of different problems from simple toy models to real chemical systems. We present strategies to systematically enhance the LPR, which can be used to improve the robustness, interpretability, and transferability of atomistic ML models.

References in corpus (5)

• ANI-1: An extensible neural network potential with DFT accuracy at force field computational cost
• Machine-learning based interatomic potential for amorphous carbon
• Nearsightedness of Electronic Matter
• Plane-wave based electronic structure calculations for correlated materials using dynamical mean-field theory and projected local orbitals
• Finite-temperature materials modeling from the quantum nuclei to the hot electrons regime

Cited by in corpus (10)

• Artificial Intelligence for Science in Quantum, Atomistic, and Continuum Systems
• Analyzing Atomic Interactions in Molecules as Learned by Neural Networks
• Thermal conductivity of Li${}_3$PS${}_4$ solid electrolytes with ab initio accuracy
• Prediction rigidities for data-driven chemistry
• SPA${}^{\mathrm{H}}$M(a,b): encoding the density information from guess Hamiltonian in quantum machine learning representations
• Uncertainty Quantification in Atomistic Simulations of Silicon using Interatomic Potentials
• Adaptive energy reference for machine-learning models of the electronic density of states
• Interpolating many-body wave functions for accelerated molecular dynamics on the near-exact electronic surface
• Shifting sands of hardware and software in exascale quantum mechanical simulations
• Assessing zero-shot generalisation behaviour in graph-neural-network interatomic potentials