Neural Networks for Ultrasensitive Nuclear Measurements

Jesse D. Ward, Craig E. Aalseth, Emily K. Mace

Feb 12, 2018 (modified: Feb 12, 2018) ICLR 2018 Workshop Submission readers: everyone
  • Abstract: This report details the application of neural networks to a data challenge in the classification of pulses generated by an ultra-low background proportional counter (ULBPC) developed at Pacific Northwest National Laboratory (PNNL). In addition to true radioactive decay events there can be spurious pulses detected due to baseline noise, microdischarge, and pileup events. In order to distinguish these events, we leveraged the ability of neural networks to make fine distinctions between inputs. We find that a fully-connected neural network is able to properly classify events in datasets with significant microdischarge and noise contributions.
  • TL;DR: Neural networks can reliably distinguish true detection events from spurious events.
  • Keywords: nuclear measurements, low background, proportional counters, neural networks