Nuclear masses learned from a probabilistic neural network
A. E. Lovell, Arvind Mohan, T. M. Sprouse, Matthew R. Mumpower
Abstract
Modeling of nuclear masses is important for many areas of nuclear science including nuclear astrophysics, reaction modeling, and nuclear data evaluations, but accuracy is challenging. This paper shows how judicious use of physics knowledge---so-called feature-space engineering---in machine learning, coupled with sophisticated models of theoretical uncertainties, can lead to better predictions.
Topics & Concepts
Probabilistic logicArtificial neural networkComputer scienceArtificial intelligenceNuclear physics research studiesParticle physics theoretical and experimental studiesQuantum Chromodynamics and Particle Interactions