학술논문
Fast and Flexible Analysis of Direct Dark Matter Search Data with Machine Learning
Document Type
article
Author
Collaboration, LUX; Akerib, DS; Alsum, S; Araújo, HM; Bai, X; Balajthy, J; Bang, J; Baxter, A; Bernard, EP; Bernstein, A; Biesiadzinski, TP; Boulton, EM; Boxer, B; Brás, P; Burdin, S; Byram, D; Carrara, N; Carmona-Benitez, MC; Chan, C; Cutter, JE; Viveiros, L de; Druszkiewicz, E; Ernst, J; Fan, A; Fiorucci, S; Gaitskell, RJ; Ghag, C; Gilchriese, MGD; Gwilliam, C; Hall, CR; Haselschwardt, SJ; Hertel, SA; Hogan, DP; Horn, M; Huang, DQ; Ignarra, CM; Jacobsen, RG; Jahangir, O; Ji, W; Kamdin, K; Kazkaz, K; Khaitan, D; Korolkova, EV; Kravitz, S; Kudryavtsev, VA; Leason, E; Lenardo, BG; Lesko, KT; Liao, J; Lin, J; Lindote, A; Lopes, MI; Manalaysay, A; Mannino, RL; Marangou, N; McKinsey, DN; Mei, D-M; Morad, JA; Murphy, A St J; Naylor, A; Nehrkorn, C; Nelson, HN; Neves, F; Nilima, A; Oliver-Mallory, KC; Palladino, KJ; Rhyne, C; Riffard, Q; Rischbieter, GRC; Rossiter, P; Shaw, S; Shutt, TA; Silva, C; Solmaz, M; Solovov, VN; Sorensen, P; Sumner, TJ; Swanson, N; Szydagis, M; Taylor, DJ; Taylor, R; Taylor, WC; Tennyson, BP; Terman, PA; Tiedt, DR; To, WH; Tvrznikova, L; Utku, U; Vacheret, A; Vaitkus, A; Velan, V; Webb, RC; White, JT; Whitis, TJ; Witherell, MS; Wolfs, FLH; Woodward, D; Xian, X; Xu, J; Zhang, C
Source
Subject
Language
Abstract
We present the results from combining machine learning with the profilelikelihood fit procedure, using data from the Large Underground Xenon (LUX)dark matter experiment. This approach demonstrates reduction in computationtime by a factor of 30 when compared with the previous approach, without lossof performance on real data. We establish its flexibility to capture non-linearcorrelations between variables (such as smearing in light and charge signalsdue to position variation) by achieving equal performance using pulse areaswith and without position-corrections applied. Its efficiency and scalabilityfurthermore enables searching for dark matter using additional variableswithout significant computational burden. We demonstrate this by including alight signal pulse shape variable alongside more traditional inputs such aslight and charge signal strengths. This technique can be exploited by futuredark matter experiments to make use of additional information, reducecomputational resources needed for signal searches and simulations, and makeinclusion of physical nuisance parameters in fits tractable.