Comparison of the mean absolute error (MAE) on a test set in the formation energy (Ef in eV/atom), the band gap (Eg in eV, the superscript nz refers to datasets restricted to non-zero band gaps), the refractive index (n) for MODNet, two variants of MEGNet and SISSO as a function of the training set size (Ntrain). Underlined values correspond to the lowest error for each task. aThe MEGNet variant including transfer learning.
De Breuck, Pierre-Paul, Geoffroy Hautier, and Gian-Marco Rignanese. "Materials property prediction for limited datasets enabled by feature selection and joint learning with MODNet." npj Computational Materials 7, no. 1 (2021): 1-8.
Materials property prediction , limited datasets ,, by feature selection,
Curator: hujianju@gmail.com