Providing physics guidance in Bayesian neural networks from the input layer: The case of giant dipole resonance predictions

Xiaohang Wang, Long Zhu, and Jun Su

Phys. Rev. C 104, 034317 – Published 20 September 2021

Abstract

Background: A Bayesian neural network (BNN) approach has been applied to evaluate and predict the nuclear data. The BNN is a numerical algorithm. When one incorporates this algorithm in nuclear physics analyses, how to maintain the scientific rigor is a key problem and presents new challenges.

Purpose: In this paper, a case study on giant dipole resonance (GDR) energy is presented to illustrate the effectiveness and maneuverability of the method to provide physics guidance in the BNN from the input layer.

Methods: Pearson's correlation coefficients are applied to assess the statistical dependence between nuclear properties in the ground state and the GDR energies. Then the optimal ground-state properties are employed as variables of the input layer in the BNN to evaluate and predict the GDR energies.

Results: Those selected ground-state properties actively contribute to reduce the predicted errors and avoid the overfitting.

Conclusions: This paper gives a demonstration to find effects of the GDR energy by using the BNN without the physics motivated model, which may be helpful to discover physics effects from the complex nuclear data.