Science‐Driven Atomistic Machine Learning

Abstract

Machine learning (ML) algorithms are currently emerging as powerful tools in all areas of science. Conventionally, ML is understood as a fundamentally data‐driven endeavour. Unfortunately, large well‐curated databases are sparse in chemistry. In this contribution, I therefore review science‐driven ML approaches which do not rely on “big data”, focusing on the atomistic modelling of materials and molecules. In this context, the term science‐driven refers to approaches that begin with a scientific question and then ask what training data and model design choices are appropriate. As key features of science‐driven ML, the automated and purpose‐driven collection of data and the use of chemical and physical priors to achieve high data‐efficiency are discussed. Furthermore, the importance of appropriate model evaluation and error estimation is emphasized.

Advanced electrocatalytic redox processes for environmental remediation of halogenated organic water pollutants

“…442 Data-driven artificial intelligence models are capable of being integrated into active and iterative learning schemes that incorporate experimental results to improve the extrapolative, i.e., predictive, capabilities of models. 456,457 Machine learning can also be combined with multiscale simulations and quantum mechanics to predict the performance of surface sites of catalysts. 458 Advanced in situ or operando spectroscopies, together with computational first principles and machine learning This journal is © The Royal Society of Chemistry 2023 approaches enable much-needed quantitative understanding of catalytically active sites, reactions centres, and reaction mechanisms.…”

“…, predictive, capabilities of models. 456,457 Machine learning can also be combined with multiscale simulations and quantum mechanics to predict the performance of surface sites of catalysts. 458…”

Physics-inspired machine learning of localized intensive properties

“…Due to their great potential for accelerating materials discovery and design, there has been signicant interest in machine learning (ML) models that enable the fast and accurate prediction of molecular and materials properties. [1][2][3][4][5] Consequently, a wide range of neural network (NN) and Kernel ML methods have been developed and applied to systems ranging from isolated molecules to complex amorphous solids. [6][7][8][9][10][11][12][13][14] In this context, many state-of-the-art approaches exploit the approximately local nature of chemical interactions.…”

q-pac: A Python Package for Machine Learned Charge Equilibration Models

“…Atomistic machine learning (ML) methods and interatomic potentials in particular have had an enormous impact in the fields of molecular and materials simulation. [1][2][3][4][5] One of the key innovations that made this possible was the idea of decomposing the total energy of a system into atomic contributions, which could be learned as a function of each atom's chemical environment within a certain cutoff radius using Neural Networks (NNs) 6 or Gaussian Process Regression (GPR) 7 . This locality assumption has enabled the construction of highly accurate, computationally efficient and size-extensive potentials, that approach first-principles accuracy at a fraction of the cost.…”