Nuclear mass predictions based on convolutional neural network

A convolutional neural network (CNN) is employed to investigate nuclear mass. By introducing the masses of neighboring nuclei and the paring effects at the input layer of the network, local features of the target nucleus are extracted to predict its mass. Then, through learning the differences between the predicted nuclear masses by the WS4 model and the experimental nuclear masses, a new global-local model (CNN-WS4) is developed, which incorporates both the global nuclear mass model and local features. This model achieves an accuracy of 0.070 MeV for the nuclei with $Z\geqslant8$ and $N\geqslant8$ in AME2016, significantly enhancing the accuracy of nuclear mass prediction. When extrapolating for newly emerged nuclei in AME2020, the CNN-WS4 also exhibits appreciable stability, thereby demonstrating its robustness.