A simple denoising approach to exploit multi-fidelity data for machine learning materials properties

TL;DR

This paper introduces a denoising method to better utilize multi-fidelity data, such as experimental and computational results, for predicting materials properties like band gaps, enhancing data quality and model accuracy.

Contribution

It presents a novel denoising approach that improves the integration of multi-fidelity data in machine learning models for materials property prediction.

Findings

Denoising enhances prediction accuracy of band gaps.

Multiple denoising iterations improve data quality.

The method outperforms existing multi-fidelity data exploitation techniques.

Abstract

Machine-learning models have recently encountered enormous success for predicting the properties of materials. These are often trained based on data that present various levels of accuracy, with typically much less high- than low-fidelity data. In order to extract as much information as possible from all available data, we here introduce an approach which aims to improve the quality of the data through denoising. We investigate the possibilities that it offers in the case of the prediction of the band gap relying on both limited experimental data and density-functional theory relying different exchange-correlation functionals (with an increasing amount of data as the accuracy of the functional decreases). We explore different ways to combine the data into training sequences and analyze the effect of the chosen denoiser. Finally, we analyze the effect of applying the denoising procedure several times until convergence. Our approach provides an improvement over existing methods to exploit multi-fidelity data.