Speaker
Description
Accurate knowledge of fission product yields (FPY) is fundamental for nuclear reactor physics, waste management, and the understanding of nuclear structure. However, experimental data for neutron-induced fission are often sparse, especially across wide incident energy ranges. This work explores the application of Machine Learning techniques, specifically Gaussian Process Regression (GPR), to evaluate independent fission charge yields and investigate their energy dependence.
The methodology focuses on two distinct approaches. First, the GPR model is trained on evaluated data from the JENDL-5 library to establish a robust baseline for predicting charge distributions in regions where experimental measurements are missing. Second, comprehensive simulations are performed using the GEF (General-purpose Fission-model) code to generate synthetic datasets for a wide range of isotopes, from Thorium to Fermium. These simulations provide a theoretical framework for comparative analysis and allow the GPR model to learn complex physical patterns, such as odd-even staggering and the filling of the symmetry valley at higher energies.
Different covariance functions (kernels), including Radial Basis Function (RBF), Matern, and Rational Quadratic, are benchmarked using Automatic Relevance Determination (ARD) to identify the most influential physical features, such as the atomic number ($Z$) and the incident neutron energy ($E_n$). Special attention is given to the impact of noise modeling through the integration of White Noise kernels to account for aleatoric uncertainties in the data.
The results demonstrate that the GPR framework effectively reproduces the charge yield distributions and generalizes well to unseen systems, such as $^{235}$U. Furthermore, the sensitivity analysis reveals the importance of balancing local structural details with global energy trends. This study aims to provide an automated, uncertainty-aware computational tool for nuclear data evaluation, contributing to the development of more reliable fission models for both fundamental research and industrial applications.