Modeling masses with an artificial neural network
Fall DNP
Contributed presentation on 10/2022
We present a new model of masses based on an artificial neural network. Using a randomized fraction (~20%) of the Atomic Mass Evaluation we predict ~80% of measured masses to with an accuracy of approximately 300 keV. We employ a Mixture Density Network to produce probabilistic output. Thus our methodology also provides confidence intervals for each prediction. Addition of a physical constraint, here the Garvey-Kelson relations, greatly improves the predictive capabilities of this modeling.
LA-UR-22-31213