Speaker
Description
In this work, a modern computational method was developed with the goal to predict the total kinetic energy of neutron-induced fission fragments across a wide range of nuclei. This data-driven approach was trained on a large-scale, simulated dataset generated by the semi-empirical GEF code, which provides a comprehensive baseline of fission observables. The methodology was employed with a Bayesian neural network (BNN) architecture, augmented with a mixture-density output layer, in order to showcase accordingly the probabilistic, multimodal nature of fission, while also allowing the network to separate and quantify the aleatoric uncertainty inherent in the stochastic fission process from the epistemic uncertainty associated with the model's parameters and the training data. By enabling principled uncertainty quantifications, the model aims in increasing the conclusion capabilities in nuclear data products, clarifying the credibility of predictions derived from model-generated data and providing a flexible tool for future theoretical, experimental and applied investigations into fission phenomena.