New Deep Learning Data Analysis Method for PROSPECT using GAPE: Genetic Algorithm Powered Evolution
M. Adriamirado and O. Benevides Rodrigues and A. Irani and B. R. Littlejohn and F. Machado, N. S. Bowden, T. Classen, S. Ghosh, M. P. Mendenhall, C. Roca, X. Zhang, A. J. Conant, G. Deichert, M. Fuller, M. J. Dolinski, C. E. Lane, A. Lozano Sanchez, R. Neilson, P. E. Mueller, B. Heffron, D. Venegas-Vargas, S. Hans, R. Rosero, M. Yeh, H. P. Mumm, J. Wilhelmi, J. Maricic, A. M. Meyer, R. Milincic · 2026
Read New Deep Learning Data Analysis Method for PROSPECT using GAPE: Genetic Algorithm Powered Evolution by M. Adriamirado and O. Benevides Rodrigues and A. Irani and B. R. Littlejohn and F. Machado, N. S. Bowden, T. Classen, S. Ghosh, M. P. Mendenhall, C. Roca, X. Zhang, A. J. Conant, G. Deichert, M. Fuller, M. J. Dolinski, C. E. Lane, A. Lozano Sanchez, R. Neilson, P. E. Mueller, B. Heffron, D. Venegas-Vargas, S. Hans, R. Rosero, M. Yeh, H. P. Mumm, J. Wilhelmi, J. Maricic, A. M. Meyer, R. Milincic (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-08814.txt.
Introduction
M. Adriamirado and 42 colleagues propose GAPE, a genetic method for building deep learning tools for the PROSPECT reactor neutrino test at Oak Ridge. GAPE creates energy and place gauges and sorters for inverse beta decay signals from common backgrounds, and adapts to other particle physics machine learning problems. Tested against standard PROSPECT pathways on the same information, its sorter offers nearly 2.8 times better signal to background ratio, with period set training cutting time linked biases.
Catalog record
| CREATOR | M. Adriamirado and O. Benevides Rodrigues and A. Irani and B. R. Littlejohn and F. Machado, N. S. Bowden, T. Classen, S. Ghosh, M. P. Mendenhall, C. Roca, X. Zhang, A. J. Conant, G. Deichert, M. Fuller, M. J. Dolinski, C. E. Lane, A. Lozano Sanchez, R. Neilson, P. E. Mueller, B. Heffron, D. Venegas-Vargas, S. Hans, R. Rosero, M. Yeh, H. P. Mumm, J. Wilhelmi, J. Maricic, A. M. Meyer, R. Milincic |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 62307 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.08814
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
New Deep Learning Data Analysis Method for PROSPECT using GAPE: Genetic Algorithm Powered Evolution M. Adriamirado and O. Benevides Rodrigues and A. Irani and B. R. Littlejohn and F. Machado, N. S. Bowden, T. Classen, S. Ghosh, M. P. Mendenhall, C. Roca, X. Zhang, A. J. Conant, G. Deichert, M. Fuller, M. J. Dolinski, C. E. Lane, A. Lozano Sanchez, R. Neilson, P. E. Mueller, B. Heffron, D. Venegas-Vargas, S. Hans, R. Rosero, M. Yeh, H. P. Mumm, J. Wilhelmi, J. Maricic, A. M. Meyer, R. Milincic Abstract We propose a genetic algorithm powered evolution (GAPE) method to create deep learning solutions for energy and position estimation for reactor antineutrino interactions in the Precision Reactor Oscillation and Spectrum Experiment (PROSPECT) at the highly enriched High Flux Isotope Reactor (HFIR) at Oak Ridge National Laboratory. We also apply GAPE to create classification models to distinguish signatures of inverse beta decay (IBD) interactions of reactor antineutrinos from common background types. The GAPE method can also be adopted for optimization of other types of problems that utilize machine learning (ML) models for particle physics applications. When applied in the PROSPECT context, we find that the models selected by GAPE can, in some cases, outperform the traditional models previously used for PROSPECT data analysis. In particular, when benchmarked against conventional PROSPECT neutrino identification pathways using the same underlying information, the classifier offers the promise of improving the signal-to-background ratio by nearly 2.8 times. Performance biases uncovered during initial IBD classifier validation were primarily caused by differences in time-dependent response between background and signal training datasets. Biases were effectively mitigated through a data-period-specific training regimen, offering a pathway towards realizing an unbiased IBD signal classifier for future reactor neutrino datasets. Prospect Collaboration † † heading: 23 2026 1- TBA; Revised TBA TBA 21-0000 † † shortheadings: GAPE / Adriamirado et al. † † firstpage: 1 Correspondence: [email protected] keywords Machine Learning, Deep Neural Networks, Scintillation, Genetic Algorithm, Neutrinos, Reactor Experiment 1 Introduction Machine learning techniques are now widely used in the particle physics community for a variety of purposes, from particle tracking and reconstruction 39 to full waveform analysis in Time of Flight (TOF) experiments 34 . Recent neutrino experiments that have utilized ML methods include DUNE 4 , MicroBoone 5 , IceCube 2 , and PROSPECT 12 . A 2022 survey of ML methods in particle physics experiments was published by Nature 42 . Our paper introduces a genetically enhanced deep learning approach for reconstructing the physical attributes of MeV-energy neutrinos produced inside the High Flux Isotope Reactor (HFIR) at Oak Ridge National Laboratory (ORNL) and detected inside the PROSPECT (Precision Reactor Oscillation and SPECTrum) particle physics detector. In particular, we will determine how this deep learning approach, referred to as GAPE, compares to PROSPECT’s traditional methods in separating true neutrino interactions from common backgrounds and in reconstructing the true energy and location of the interacting neutrinos. Background rejection is an important aspect of PROSPECT performance, since the detector seeks to measure weakly-interacting neutrinos in the high-radiation environment of an on-surface nuclear reactor facility. Neutrino energy reconstruction is also important for PROSPECT measurements, since a primary deliverable of the experiment is to perform measurements of the aggregate energy spectrum of HFIR’s neutrinos. There has been sizeable interest recently in the study of reactor antineutrino energy spectra, as there have been observed disagreements 9 ; 11 ; 10 ; 3 ; 8 with predictions based on beta-spectrum conversion 26 ; 32 and calculations from nuclear databases 23 . These discrepancies report excesses of detected interactions of neutrinos in the region between 5-7 MeV. PROSPECT was designed to make precision measurements of the antineutrino spectrum and probe eV scale massive sterile neutrinos by searching for neutrino oscillations over meter-long baselines from a highly-enriched uranium reactor 14 . The 4-ton antineutrino detector target, shown in Figure 1 , consists of a 14 by 11 grid of long, rectangular segments, filled with a {}^{6}\mathrm{Li} -doped liquid scintillator to detect antineutrinos through inverse beta decay (IBD) while efficiently rejecting backgrounds. Each segment is 1.176 m in length and has a square cross-sectional area of 0.145 m × 0.145 m, with a photomultiplier tube (PMT) in an acrylic housing on both ends of each segment. 240 Hamamatsu R6594 PMTs and 68 ET 9372KB PMTs were used in the experiment. The detector also contains gamma ray and neutron shielding, along with detector movement elements (in practice the detector was never moved). For a comprehensive review of the many elements of the PROSPECT shielding see 11 . Figure 1: (top) Diagram 14 of the layout for the experiment. Detector center to the reactor core center is approximately 7.84m. (Bottom) Diagram 14 showing cross-section of the PROSPECT detector. PROSPECT collected physics-quality data for 96 days when the HFIR reactor was on and 73 days when the reactor was off, with all data collected at an average baseline distance of 7.84 m from the center of the 85 MW HFIR core. Using traditional analysis techniques, PROSPECT detected a total of 115,852 IBD-like events in the reactor-on data set 11 , with 28,358 ± 18 being contributed by reactor-related, non-neutrino backgrounds, a category that we will describe later in greater detail and will hereafter refer to as ‘accidental’ backgrounds.A total of 30,568 IBD-like signals detected in the reactor-off dataset were contributed overwhelmingly by cosmic ray interactions inside PROSPECT, a background contribution that is largely consistent between reactor-on and reactor-off periods. Following live-time and atmospheric pressure scalings and subsequent subtraction of these backgrounds, we are left with a total of 50,560 ± 406 interactions of electron antineutrinos 11 . The majority come from beta decays of {}^{235}\mathrm{U} fission products but approximately 1 \% come from non-equilibrium isotopes and non-fuel sourced antineutrinos from {}^{28}\mathrm{Al} and {}^{6}\mathrm{He} . For more information on the detector design, experiment layout, experimental results, as well as detailed information on calibration efforts, see 14 and 11 . PROSPECT detects HFIR-produced neutrinos by sensing the products of their interaction with a proton via IBD. In IBD events, the antineutrino interacts with a proton in the detector to produce a positron and a neutron. The positron carries most of the antineutrino energy and quickly annihilates with an electron, producing gamma rays, while the neutron thermalizes and then most often captures on a 6 Li nucleus (approximately 75 \% of the time), which decays into a triton and an alpha particle, depositing energy. Both of these time-correlated events emit visible photons, which are converted into photoelectrons (PEs) and counted by the PMTs at the end of each segment 11 . Thus, IBD candidates are selected by searching for two events close to each other in time and position, with the earlier of these events, referred to as the ”prompt” event, having characteristics of positron annihilation, and the later event (referred to as the ”delayed” event), having characteristics of a neutron capture. Important characteristics used here are the total energy measured in each event, as well as the PMT pulse shape discrimination (PSD) that measures the relative intensities of fast and slow scintillation light at each PMT. The PSD is distinctly different for neutron captures as it is for electromagnetic signatures. A more detailed discussion of the IBD