Improving neural network predictions of material properties with limited data using transfer learning

TL;DR

This paper introduces transfer learning algorithms to improve neural network predictions of material properties from limited data, enabling efficient and generalizable models in materials science.

Contribution

The paper develops novel transfer learning methods tailored for predicting material properties, reducing data requirements and enhancing model transferability in materials science.

Findings

Effective transfer learning algorithms for material property prediction.

Successful application to Gibbs free energy of transition metal oxides.

Reduced need for large training datasets.

Abstract

We develop new transfer learning algorithms to accelerate prediction of material properties from ab initio simulations based on density functional theory (DFT). Transfer learning has been successfully utilized for data-efficient modeling in applications other than materials science, and it allows transferable representations learned from large datasets to be repurposed for learning new tasks even with small datasets. In the context of materials science, this opens the possibility to develop generalizable neural network models that can be repurposed on other materials, without the need of generating a large (computationally expensive) training set of materials properties. The proposed transfer learning algorithms are demonstrated on predicting the Gibbs free energy of light transition metal oxides.