Bayesian neural networks for the evaluation of reaction cross-section of interest in nucleosynthesis studies

Nuclear reaction cross-sections are of primary interest in the nucleosynthesis studies. The nuclear reactions involved in such studies are difficult to be measured experimentally at all the projectile energies of interest. Therefore, the theoretical model predictions are generally optimized using the experimental data available. However, difficulty arises in the estimation of the uncertainties appropriately with such predictions. In this study we have used a Bayesian neural network to produce cross-section predictions along with their uncertainty estimates. In order to leverage on the data from the available theoretical models, we have used a low-fidelity Bayesian neural network. As a test case we have trained a multi-fidelity learning model by the data from the Talys prediction and available experimental data. We have used experimental data for ¹⁴N(n,p)¹⁴C reaction from EXFOR data library and theoretical predictions from TENDL data library for our low-fidelity estimate which will be discussed at length in the conference