Table 1 Prediction performance (<MAE> in eV/atom and R2) of the ML models based on CE, Voronoi, and ElemNet features, respectively, using various datasets: spinel oxides (source domain Modelspinel), perovskite oxides (target domain Modelperovskite), direct prediction of perovskite oxides using the source domain model (ModelSP), transfer learning model (ModelTL).

From: Center-environment deep transfer machine learning across crystal structures: from spinel oxides to perovskite oxides

Model

Modelspinel

Modelperocskite

ModelSP

ModelTL

Metric

<MAE>

<R2>

<MAE>

<R2>

<MAE>

<R2>

<MAE>

<R2>

CE feature

0.072

0.98

0.132

0.95

0.514

0.53

0.106

0.96

Voronoi feature69

0.116

0.92

0.176

0.92

0.495

0.52

0.166

0.91

ElemNet68

0.372

0.65

N/A

 

N/A

 

N/A

 
  1. The results were averaged over 10 runs in the 10-fold cross-validation during the training process.