神经网络方法对原子核质量的预言

Nuclear mass prediction using a neural network approach

原子核质量包含丰富的核结构信息,是原子核的基本物理性质之一.本文采用前馈神经网络和贝叶斯神经网络方法研究原子核质量,通过在网络输入层引入与原子核对效应和壳效应相关的两个物理量,在网络输出层考虑现有核模型的质量预言,即学习原子核质量实验值与核模型预言值之差,大幅提高了核模型的质量预言精确度.相比前馈神经网络中常用的优化算法,贝叶斯方法在优化神经网络参数方面具有较大优势,显著改进了神经网络对原子核质量预言中对效应和壳效应的描述.

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.