Nuclear mass prediction using a neural network approach
Abstract
The nuclear mass contains abundant information about the nuclear structure, which is one of the most important properties of atomic nuclei. In this work, feedforward neural network and Bayesian neural network methods are used to study nuclear mass. By introducing two physical quantities related to the nuclear pairing effect and the shell effect at the input layer of the network, and considering the mass prediction of existing nuclear models at the output layer of the network, that is, learning the mass difference between the predicted experimental data by existing nuclear models, the predictive ability of the proposed model is significantly improved. Compared with the optimization algorithms commonly used in feedforward neural networks, Bayesian methods have great advantages in optimizing neural network parameters and significantly improve the description of the pairing and shell effects in mass predictions.
