Global model of neural networks, stability and learning

Statistical modeling of nuclear data using artificial neural networks (ANNs) and, more re- cently, support vector machines (SVMs), is providing novel approaches to systematics that are complementary to phenomenological and semi-microscopic theories. We present a global model of β−-decay halflives of the class of nuclei that decay 100% by β− mode in their ground states. A fully-connected multilayered feed forward network has been trained using the Levenberg- Marquardt algorithm, Bayesian regularization, and cross-validation. The halflife estimates gen- erated by the model are discussed and compared with the available experimental data, with previous results obtained with neural networks, and with estimates coming from traditional global nuclear models. Predictions of the new neural-network model are given for nuclei far from stability, with particular attention to those involved in r-process nucleosynthesis. This study demonstrates that in the framework of the β−-decay problem considered here, global models based on ANNs can at least match the predictive performance of the best conventional global models rooted in nuclear theory. Accordingly, such statistical models can provide a valuable tool for further mapping of the nuclidic chart.

Download Full-text

Node-wise Localization of Graph Neural Networks

Proceedings of the Thirtieth International Joint Conference on Artificial Intelligence ◽

10.24963/ijcai.2021/210 ◽

2021 ◽

Author(s):

Zemin Liu ◽

Yuan Fang ◽

Chenghao Liu ◽

Steven C.H. Hoi

Keyword(s):

Neural Networks ◽

State Of The Art ◽

Representation Learning ◽

Global Model ◽

Local Context ◽

Local Contexts ◽

Unique Model ◽

Graph Neural Networks ◽

Global And Local ◽

Different Parts

Graph neural networks (GNNs) emerge as a powerful family of representation learning models on graphs. To derive node representations, they utilize a global model that recursively aggregates information from the neighboring nodes. However, different nodes reside at different parts of the graph in different local contexts, making their distributions vary across the graph. Ideally, how a node receives its neighborhood information should be a function of its local context, to diverge from the global GNN model shared by all nodes. To utilize node locality without overfitting, we propose a node-wise localization of GNNs by accounting for both global and local aspects of the graph. Globally, all nodes on the graph depend on an underlying global GNN to encode the general patterns across the graph; locally, each node is localized into a unique model as a function of the global model and its local context. Finally, we conduct extensive experiments on four benchmark graphs, and consistently obtain promising performance surpassing the state-of-the-art GNNs.

Download Full-text

One and Two Proton Separation Energies from Nuclear Mass Systematics using Neural Networks

HNPS Proceedings ◽

10.12681/hnps.2980 ◽

2020 ◽

Vol 13 ◽

pp. 305

Author(s):

S. Athanassopoulos ◽

E. Mavrommatis ◽

K. A. Gernoth ◽

J. W. Clark

Keyword(s):

Neural Networks ◽

Theoretical Models ◽

Atomic Mass ◽

Global Model ◽

Nuclear Mass ◽

Rp Process ◽

The One ◽

The Masses

We deal with the systematics of one and two proton separation energies as predicted by our latest global model for the masses of nuclides developed with the use of neural networks. Among others, such systematics is useful as input to the astrophysical rp-process and to the one and two proton radioactive studies. Our results are compared with the experimental separation energies referred to in the 2003 Atomic Mass Evaluation and with those evaluated from theoretical models for the masses of nuclides, like the FRDM of Möller et al. and the HFB2 of Pearson et al. We focus in particular on the proton separation energies for nuclides that are involved in the rp-process (29<Z<40) but they have not yet been studied experimentally.

Download Full-text