Improving molecular machine learning through adaptive subsampling with active learning

Abstract

Active machine learning can be used to sample training data in an autonomous manner to improve machine learning performance. This approach is competitive with state-of-the-art data sampling approaches, especially on erroneous data.