Model-independent searches of new physics in DARWIN with deep learning

Abstract We present a deep learning pipeline to perform a model-independent, likelihood-free search for anomalous (i.e., non-background) events in the proposed next-generation multi-ton scale liquid xenon-based direct detection experiment, DARWIN. We train an anomaly detector comprising a variational autoencoder (VAE) and a classifier on high-dimensional simulated detector response data and construct a 1D anomaly score to reject the background-only hypothesis in the presence of an excess of non-background-like events. We use simulated validation data to determine the power of the method to reject the background-only hypothesis in the presence of a WIMP dark matter signal, without any model-dependent assumption about the nature of the signal. We show that our neural networks learn relevant features of the events from low-level, high-dimensional detector outputs, avoiding lossy and computationally expensive compression into lower-dimensional observables. Our approach is complementary to the usual likelihood-based analysis, in that it reduces the reliance on many of the corrections and cuts that are traditionally part of the analysis chain, with the potential of achieving higher accuracy and significant reduction of analysis time. We envisage the methodology presented in this work augmenting or complementing likelihood-based and other data-driven methods currently utilized in the DARWIN (and in the future, XLZD) analysis pipeline.

Authors

Institutions

Publication Details

Journal
The European Physical Journal C
Published
2026-03-26
DOI
https://doi.org/10.1140/epjc/s10052-025-15161-2
Primary Topic
Dark Matter and Cosmic Phenomena
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
article

Model-independent searches of new physics in DARWIN with deep learning

M. Kara, K. Miuchi, F. Lombardi, G. Plante et al.
The European Physical Journal C
Dark Matter and Cosmic Phenomena
article

Model-independent searches of new physics in DARWIN with deep learning

M. Kara, K. Miuchi, F. Lombardi, G. Plante, F Kahlert, Luis Sanchez, M. Babicz, B. Antunović, P. Kavrigin, C. Wittweg, F. Glück, R. Glade-Beucke, D. Bajpai, J. Grigat, J. Qin, Monique Flierman, N. Garroum, C. Capelli, M. Pierre, C. Fuselli, L. Baudis, Y. Mosbacher, S. Lindemann, D. Vorkapic, A. Kopec, P. Di Gangi, Ciaran A. J. O’Hare, A. Elykov, M. Galloway, Y. Biondi, W. Fulgione, Julian Conrad, M. Iacovacci, K. Morå, M. Guida, R Giacomobono, Roberto Trotta, P Kharbanda, A. Razeto, Q. Pellegrini, P. Sánchez-Lucas, J. Schreiner, Y. Xing, S. Mastroianni, A. Deisting, Robin Größle, Rie Miyata, Andre Scaffidi, Kenzo Hiraoka, D. Thers, V. D’Andrea, D. Wenz, G. Zavattini, F. Piastra, Róbert Braun, R. Biondi, S. Milutinovic, S. Liang, Martín F. Adrover, M. Wurm, M. Schümann, Adrian Schwenck, T.C. Luce, P. Urquijo, M. J. Wilson, Kazuo Fujikawa, M. Pandurović, V. Hannen, K. Valerius, L. Grandi, R. Peres, L Principe, N Hargittai, Prof. Elena Aprile, Atsushi Takeda, M. Messina, Ke Qiao, P.-L. Tan, A. Mancuso, Kaixiang Ni, F. Toschi, Frédéric Girard, Romain Gaïor, Amos Breskin, Guido Drexlin, Andrea Molinario, Gabriella Sartorelli, H. Landsman, R. Hammann, F. Semeria, Belina von Krosigk, F. Kuger, A. Stevens, J. Jakob, Owen Stanley, E. Angelino, A. Chauvin, A. Ravindran, M. Balzer, C Di Donato, C. Ferrari, P Gyorgy, J. Long, A. P. Colijn, A. D. Ferella, T. R. Pollmann, F. Tönnies, N. F. Hood, F. Marignetti, M. Doerenkamp, D. Antón Martin, B. Maier, M. Weiss, M. Zhong, J. Ye, G. Trinchero, T. Flehmke, H. Schulze Eißing, B. Andrieu, E. Morteau, M. Rajado, I. Li, C. Hils, Y. Pan, Y. Itow, M. P. Decowski, S. Vecchi, J. A. M. Lopes, C. Boehm, M. Bazyk, L. Scotto Lavina, S. Ouahada, U. Oberlack, M. Selvi, T. Shimada, S. Brommer, C. Cai, A. Bismark, M. Lindner, G. M. Lucchetti, A. Melchiorre, L. Hoetzsch, J. Müller, S. Li, L. C. Daniel Garcia, R. Budnik, P. Gaemers, Y. Kaminaga, F. Joerg, C. Macolino, M. Keller, T. Thümmler, L. Althueser, H. Simgen, Z. Xu, P. Shagin, L. Manenti, E. Barberio, K. M. Weerman, L. LaCascio, J. Loizeau, K. Boese, A. P. Cimental Chavez, R. Singh, R. Engel, B. Kilminster, R. S. James, H. Guan, K. Liu, L. Levinson, S. Diglio, B. Milosovic, D. Ramírez García, S. Y. Shi, V. H. S. Wu, S. Hansmann-Menzemer, S. Vetter, J. Masbou, Y. -T. Lin, E. Masson, H. Fischer, D. W. P. Amaral, W. Wang, M. Steidl, J. Aalbers, C. M. B. Monteiro, M. Solmaz, J. J. Cuenca-García, S. Ahmed Maouloud, D. Xu, S. Wüstling, G. Bruni, D. Koke, Y. Ma, W. Shen, J. Menéndez, K. Martens, M. Kleifges, L. Bellagamba, L. Yang, J. L. Newstead, A. Li, J. Qi, K. Abe, A. Higuera, F. Gao, A. Brown, C. Weinheimer, N. F. Bell, K. Zuber, M. Utoyama, M. Klute, J. Pienaar, J. Mahlstedt, Z. Liang, G. Volta, C. D. Tunnell, K. Eitel, S. Kazama, M. Yamashita, P. Schulte, I. Ostrowskiy, S. Moriyama, J. Wolf, S. Sharma, M. Obradovic, M. Kobayashi, M. Murra, L. Yuan, B. Paetsch, R. F. Lang
article en

Abstract

Abstract We present a deep learning pipeline to perform a model-independent, likelihood-free search for anomalous (i.e., non-background) events in the proposed next-generation multi-ton scale liquid xenon-based direct detection experiment, DARWIN. We train an anomaly detector comprising a variational autoencoder (VAE) and a classifier on high-dimensional simulated detector response data and construct a 1D anomaly score to reject the background-only hypothesis in the presence of an excess of non-background-like events. We use simulated validation data to determine the power of the method to reject the background-only hypothesis in the presence of a WIMP dark matter signal, without any model-dependent assumption about the nature of the signal. We show that our neural networks learn relevant features of the events from low-level, high-dimensional detector outputs, avoiding lossy and computationally expensive compression into lower-dimensional observables. Our approach is complementary to the usual likelihood-based analysis, in that it reduces the reliance on many of the corrections and cuts that are traditionally part of the analysis chain, with the potential of achieving higher accuracy and significant reduction of analysis time. We envisage the methodology presented in this work augmenting or complementing likelihood-based and other data-driven methods currently utilized in the DARWIN (and in the future, XLZD) analysis pipeline.

The European Physical Journal CVol. 86(3)
Karlsruhe Institute of Technology (DE), Roma Tre University (IT), Centre National de la Recherche Scientifique (FR), The University of Sydney (AU), Bucknell University (US), St. George's University (GD), Scuola Internazionale Superiore di Studi Avanzati (IT), University of Fribourg (CH), Stockholm University (SE), The University of Melbourne (AU), Tokyo University of the Arts (JP), University of Alabama (US), University of Banja Luka (BA), Chinese University of Hong Kong (HK), Johannes Gutenberg University Mainz (DE), University of Ferrara (IT), University of Zurich (CH), Purdue University West Lafayette (US), Heidelberg University (DE), University of Münster (DE), University of L'Aquila (IT), Istituto Nazionale di Fisica Nucleare, Laboratori Nazionali del Gran Sasso (IT), Polytechnic Institute of Coimbra (PT), Westlake University (CN), University of California San Diego (US), Sorbonne Université (FR), University of Chicago (US), University of Belgrade (RS), National Institute for Subatomic Physics (NL), Istituto Nazionale di Fisica Nucleare, Sezione di Torino (IT), Istituto Nazionale di Fisica Nucleare, Sezione di Napoli (IT), Sorbonne University Abu Dhabi (AE), Istituto Nazionale di Fisica Nucleare, Sezione di Bologna (IT), Gran Sasso Science Institute (IT), Max Planck Institute for Nuclear Physics (DE), Vitenparken (NO), University College London (GB), Imperial College London (GB), Université Paris 1 Panthéon-Sorbonne (FR), Weizmann Institute of Science (IL), Nagoya University (JP), Kobe University (JP), Universitat de Barcelona (ES), Rice University (US), University of Coimbra (PT), Columbia University (US), Technische Universität Dresden (DE), University of Amsterdam (NL), University of Sheffield (GB), University of Bologna (IT), Nantes Université (FR), Tsinghua University (CN)
Openalex Percentile: Top 7%
Dark Matter and Cosmic Phenomena
DOIOpen Access PDFOpenAlex
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.