Multitask methods for predicting molecular properties from heterogeneous data
Open Access
- 3 July 2024
- journal article
- research article
- Published by AIP Publishing in The Journal of Chemical Physics
- Vol. 161 (1)
Abstract
Data generation remains a bottleneck in training surrogate models to predict molecular properties. We demonstrate that multitask Gaussian process regression overcomes this limitation by leveraging both expensive and cheap data sources. In particular, we consider training sets constructed from coupled-cluster (CC) and density functional theory (DFT) data. We report that multitask surrogates can predict at CC-level accuracy with a reduction in data generation cost by over an order of magnitude. Of note, our approach allows the training set to include DFT data generated by a heterogeneous mix of exchange–correlation functionals without imposing any artificial hierarchy on functional accuracy. More generally, the multitask framework can accommodate a wider range of training set structures—including the full disparity between the different levels of fidelity—than existing kernel approaches based on Δ-learning although we show that the accuracy of the two approaches can be similar. Consequently, multitask regression can be a tool for reducing data generation costs even further by opportunistically exploiting existing data sources.Keywords
All Related Versions
Funding Information
- National Nuclear Security Administration (DE-NA0003965)
- National Science Foundation Graduate Research Fellowship Program (1745302)
- National Center of Competence in Research Materials' Revolution: Computational Design and Discovery of Novel Materials (205602)
This publication has 54 references indexed in Scilit:
- Numerical stability and efficiency of response property calculations in density functional theoryLetters in Mathematical Physics, 2023
- Learning from multiple quantum chemical methods: Δ-learning, transfer learning, co-kriging, and beyond2022
- Locally induced Gaussian processes for large-scale simulation experimentsStatistics and Computing, 2021
- A multi-fidelity information-fusion approach to machine learn and predict polymer bandgapComputational Materials Science, 2020
- Multi-fidelity machine learning models for accurate bandgap predictions of solidsComputational Materials Science, 2017
- Gaussian approximation potentials: A brief tutorial introductionInternational Journal of Quantum Chemistry, 2015
- Focused multi-task learning in a Gaussian process frameworkMachine Learning, 2012
- AFLOW: An automatic framework for high-throughput materials discoveryComputational Materials Science, 2012
- A high-throughput infrastructure for density functional theory calculationsComputational Materials Science, 2011
- Knowledge-based artificial neural networksArtificial Intelligence, 1994