Paper on MLatom

The mathematical and implementation details of the techniques available in MLatom: A Package for Atomistic Simulations with Machine Learning are reported.

MLatom has been applied since its inception in 2013 for developing approaches for assisting quantum chemical (QC) research with machine learning (ML). The program has used ML to predict molecule-specific parameters of semiempirical QC methods, to improve predictions made by low-level QC methods (Δ-learning), to generate very accurate molecular potential surfaces with significantly reduced computational cost (via structure-based sampling and self-correction), and to run nonadiabatic excited-state dynamics.

My recent publication[1] provides the theoretical background and reports all the relevant mathematical and implementation details of the methods available in the program package (its recently released version 1.0, to be precise).

MLatom has been written from the beginning to provide a user-friendly, out-of-the-box tool for performing ML atomistic simulations. It can be used without extensive knowledge of machine learning or scripting. The program is a stand-alone package, which does not require any other ML libraries. It has been also optimized for efficient, shared-memory parallelization.

For more information about the program including the list of its capabilities and online manual, please visit its website MLatom.com.

1. Pavlo O. Dral, MLatom: A Program Package for Quantum Chemical Research Assisted by Machine Learning. J. Comput. Chem. 2019, Early View. DOI: 10.1002/jcc.26004.