Quark, gluon, hadronic tau, and dark jets simulated with COCOA

This record contains samples of proton-proton collisions yielding the following two-body signatures: light quark jets gluon jets hadronic tau decays dark jets Each event is reconstructed in the GEANT4-based COCOA detector simulation [3] and two jets are extracted from each event. The files store jet-level information at truth and reconstructed level, along with the sets of truth particles, tracks, and calorimeter cells associated with each jet. The jets from (1)-(3) are mixed in the files named dijets_tautau_*.root, which are separated into train (x10), val, and test splits. The jet origin is recorded in the jet_id_truth branch. These files contain two branches, JetTree_1 and JetTree_2, which contain two different "views" of the same hard scatter event, under the data augmentation process described in [1]. The pairs of jets enable self-supervised learning using a contrastive loss function, but can also be used separately for supervised learning objectives. The dark jets are generated according to models A, B, C, and D described in [3] and are used to study anomaly detection. To align their kinematic distributions with those of the training dataset, the dark jets are resampled using the indices stored in the darkJets_indices_to_keep_model*.csv files. [1] Self-Supervised Learning Strategies for Jet Physics[2] Configurable calorimeter simulation for AI applications[3] Search for resonant production of dark quarks in the dijet final state with the ATLAS detector