Table 2 Test MAE of different models for the prediction task of “Model Architecture Design”.

From: Improving deep learning model performance under parametric constraints for materials informatics applications

Dataset

Size

AutoML

ElemNet

IRNet

BNet

BRNet

OQMD

345,134

0.149

0.049

0.042

0.042

0.041

AFLOWLIB

234,299

0.115

0.058

0.051

0.048

0.047

MP

89,181

0.167

0.121

0.117

0.112

0.106

JARVIS

19,994

0.129

0.083

0.094

0.071

0.070

  1. Here, we use formation energy (eV/atom) as the materials property and composition-based elemental fraction as the model input.
  2. The lowest MAE values in each row are highlighted in bold.
Back to article page