The insight gained from nuclear reaction data is key to our description of the atomic nucleus and astrophysical processes, as well as having applications in nuclear medicine and security.
One challenge faced for reaction measurements of this type is the competition between ideal particle accelerator parameters – which focus on producing the largest reaction yields – versus ideal particle detection settings – which center around the most selectivity. To add to this quandary, nuclear reactions typically produce tens to hundreds of nuclei at once with only a select few being of interest.
Therefore, a fully-connected neural network (NN) was developed, trained under supervision with only a sparse amount of correctly labelled data, and finally implemented to improve upon the sensitivity of isotope classification.
The NN approach, when applied to experimental data collected at the ATLAS user facility, produced enhanced results over traditional techniques by nearly an order of magnitude.
The discovery of numerous new electromagnetic transitions and quantum excited levels in the isotope of interest, 38S, was facilitated by the new approach.
Both the far-reaching impact of the new reaction data extracted, as well as the NN approach to isotope classification, are under continued exploration.