Accurate Prediction of the Optical Absorption Energies by Neural Network Ensemble Approach

The neural network ensemble approach (NNE) is proposed for improving the generalization ability of neural networks and to reduce the calculation errors of density functional theory (DFT). The simple averaging approach (NNEA) and weighted averaging approach (NNEW) for combining the predictions of component neural networks we adopted respectively. As a demonstration, this combined DFT and NNE correction approach has been applied to accurately predict the optical absorption energies of organic molecules. The NNEA and NNEW approach improved DFT calculation results and reduced the rms deviations from 0.41 to 0.20 and 0.18 eV for the testing set of organic molecules, respectively. In general, the NNE correction approach leads to better results and shows the good generalization ability.