FASERCal evaluation datasets for self-supervised representation learning

Résumé fourni par la source

This dataset contains the simulated FASERCal evaluation datasets used in the study “Towards foundation-style models for energy-frontier heterogeneous neutrino detectors via self-supervised pre-training” (arXiv:2604.07037), accepted for publication in Nature Machine Intelligence. The release comprises two datasets: GENIE: FASERCal test set used to evaluate the models and produce the principal results reported in the paper. After extraction, this dataset contains 122,638 events. NuWro: independent alternative-generator evaluation sample used to study the robustness of the trained models to changes in the neutrino-interaction generator. After extraction, this dataset contains 121,371 events. The events are stored as compressed NumPy (.npz) files containing the reconstructed detector inputs and simulation information used in the analysis. Because of the size of the datasets, each compressed TAR archive has been divided into six parts for deposition on Zenodo. The GENIE archive consists of: fasercal_genie.tar.gz.part-00 fasercal_genie.tar.gz.part-01 fasercal_genie.tar.gz.part-02 fasercal_genie.tar.gz.part-03 fasercal_genie.tar.gz.part-04 fasercal_genie.tar.gz.part-05 The NuWro archive consists of: fasercal_nuwro.tar.gz.part-00 fasercal_nuwro.tar.gz.part-01 fasercal_nuwro.tar.gz.part-02 fasercal_nuwro.tar.gz.part-03 fasercal_nuwro.tar.gz.part-04 fasercal_nuwro.tar.gz.part-05 Reconstructing the archives On Linux/macOS, concatenate the parts in filename order: cat fasercal_genie.tar.gz.part-* > fasercal_genie.tar.gz cat fasercal_nuwro.tar.gz.part-* > fasercal_nuwro.tar.gz Then extract the reconstructed archives with: tar -xzf fasercal_genie.tar.gz tar -xzf fasercal_nuwro.tar.gz The correspondence between the GENIE and NuWro event files is provided in the CSV file genie_nuwro_correspondence.csv. Associated publication:S. Alonso-Monsalve, F. Cufino, U. Kose, A. Mascellani, and A. Rubbia, Towards foundation-style models for energy-frontier heterogeneous neutrino detectors via self-supervised pre-training, arXiv:2604.07037; accepted for publication in Nature Machine Intelligence.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.