Neural network models of nuclear systematics

• Published in: Physics letters. B Vol. 300; no. 1; pp. 1 - 7
• Main Authors: Gernoth, K.A., Clark, J.W., Prater, J.S., Bohr, H.
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 04.02.1993
• ISSN: 0370-2693; 1873-2445
• DOI: 10.1016/0370-2693(93)90738-4

We describe a novel phenomenological approach to many-body systems based on multilayer feedforward neural networks. When subjected to appropriate training schemes, such networks are capable of learning the systematics of atomic masses and nuclear spins and parities with high accuracy while forming an efficient internal representation of the training data. When tested on nuclei outside the training set, these neural-network models demonstrate a predictive power competitive with that of traditional theoretical approaches, provided the test nuclei are not too different from those of the training set. The relative performance on training and test sets may be used as a measure of the homology of nuclei with respect to given observables.