College of Science & Engineering
Twin Cities
The University of Minnesota School of Physics and Astronomy participates in an experiment called NOvA, which is designed to study the properties of sub-nuclear particles called neutrinos. NOvA uses a neutrino beam produced at Fermi National Accelerator Laboratory near Chicago, a Near Detector located at Fermilab, and a 14,000 ton Far Detector located 811km from Fermilab in the University’s Ash River Laboratory near International Falls, Minnesota.
The NOvA Far Detector produces a 330,000 pixel data stream for 550 µs, each 1.3 s, synchronized to the neutrino beam particle production. The key data analysis challenge is the identification and classification of each energy deposit that is clustered in space and time. Because neutrinos only weakly interact with ordinary matter, interesting events are extremely rare, while the number of “background” events is large. Even for interesting events, there is a significant probability of an incorrect classification, which directly affect the reliability of the observations made by the experiment.
The NOvA Experiment has pioneered the use of convolutional neural-nets to classify neutrino experiments, yielding an improvement in statistical power equivalent to 30% more data. This initial improvement was obtained without any significant effort in the optimization of the network configuration or inputs. These researchers are exploring the systematic optimization of these parameters, as well as use of this technique in regression analyses to estimate continuous parameters such as the energy of the incoming neutrinos. These techniques are likely to play a significant role in future neutrino experiments, as well as other particle physics experiments. The optimization of CAFFE neutral-nets by Minnesota researchers working on NOvA is considerably facilitated by access to MSI vector processors.