Prediction of the band gap in 2D materials using active learning

Abstract

Accurate band gap prediction of two-dimensional materials holds significant scientific and technological value for the development of electronic and optoelectronic devices. In contrast to the high computational cost associated with traditional first-principles methods, machine learning offers a promising and cost-effective alternative for band gap prediction. In this work, we demonstrate that the combination of artificial neural networks and an active learning algorithm leads to a highly data-efficient method for predicting band gaps of 2D materials while maintaining accuracy, with L1-regularization analyzing feature selection. This approach achieves a computational cost reduction by shrinking the original dataset by 80% compared to traditional training approaches