Physics-Embedded Bayesian Neural Network (PE-BNN) to predict Energy Dependence of Fission Product Yields with Fine Structures

2025/04/24 by Chen, Jingde, Mukobara, Yuta, Fujio, Kazuki +3
#Data Analysis #FOS: Physical sciences #Nuclear Experiment (nucl-ex) #Nuclear Theory (nucl-th) #Statistics and Probability (physics.data-an)

paper · doi:10.48550/arxiv.2504.17275

Abstract

We present a physics-embedded Bayesian neural network (PE-BNN) framework that integrates fission product yields (FPYs) with prior nuclear physics knowledge to predict energy-dependent FPY data with fine structure. By incorporating an energy-independent phenomenological shell factor as a single input feature, the PE-BNN captures both fine structures and global energy trends. The combination of this physics-informed input with hyperparameter optimization via the Watanabe-Akaike Information Criterion (WAIC) significantly enhances predictive performance. Our results demonstrate that the PE-BNN framework is well-suited for target observables with systematic features that can be embedded as model inputs, achieving close agreement with known shell effects and prompt neutron multiplicities.

Citations

• Final Search for Short-Baseline Neutrino Oscillations with the PROSPECT-I Detector at HFIR
• Bayesian approach to energy dependence of fission product yields of 235 U by data augmentation
• Japanese evaluated nuclear data library version 5: JENDL-5
• Composable Effects for Flexible and Accelerated Probabilistic Programming in NumPyro
• Bayesian Evaluation of Incomplete Fission Yields
• Nuclear mass predictions based on Bayesian neural network approach with pairing and shell effects
• A 4-dimensional Langevin approach to low-energy nuclear fission of ²³⁶U
• The No-U-Turn Sampler: Adaptively Setting Path Lengths in Hamiltonian Monte Carlo
• MCMC using Hamiltonian dynamics
• Asymptotic Equivalence of Bayes Cross Validation and Widely Applicable Information Criterion in Singular Learning Theory

Related

• Robust extrapolation using physics-related activation functions in neural networks for nuclear masses
• Improved predictions of phenomenological nuclear charge radius formulae with Bayesian optimization approach
• Nuclear Mass Predictions Using a Neural Network with Additive Gaussian Process Regression-Optimized Activation Functions
• Nuclear mass systematics by complementing the Finite Range Droplet Model with neural networks
• Fast Uncertainty Quantification of Spent Nuclear Fuel with Neural Networks