LHC Olympics 2020: Simulated Particle Collision Events for Anomaly Detection

Three 'black box' datasets and one background sample each contain 1 million simulated particle collision events from the LHC Olympics 2020 Anomaly Detection Challenge. Each event is represented as a zero-padded array of up to 700 massless particles, with features for transverse momentum (pT), pseudorapidity (eta), and azimuthal angle (phi). The challenge tasks researchers with identifying unknown signal events hidden within the black boxes, using the provided QCD dijet background sample for reference.

Use Cases

• Developing anomaly detection algorithms based on particle kinematic features (pT, eta, phi).
• Benchmarking unsupervised learning methods against a known challenge with published solutions for two black boxes.
• Studying the performance of simulation tools (Pythia8, Delphes) for representing QCD dijet background events.
• Training models to distinguish potential new physics signals from Standard Model background in a controlled, labeled setting.

Strengths

• Contains a large, standardized scale of 1 million events per dataset, enabling statistically significant model training and evaluation.
• Includes ground truth labels (0 for background, 1/2 for signal) for Black Boxes 1 and 3, released after the challenge concluded.
• Data is structured in a consistent array format (Nevents=1M, 2100) with detailed documentation on event selection and simulation tools.

Limitations

• The simulation's physics and detector modeling may not exactly reflect the 'data' in the black boxes, introducing potential systematic differences.
• Column names and a complete data dictionary are not provided in the source metadata, requiring users to consult external documentation.
• The background sample is specific to QCD dijet events selected with a single fat-jet trigger (R=1, pT > 1.2 TeV), limiting generalizability to other event topologies.

Provenance

Source

Simulated data generated for the LHC Olympics 2020 Anomaly Detection Challenge.

Collection Method

Events simulated using Pythia8 and Delphes 3.4.1 fast simulation, with particles recorded in detector coordinates.

Freshness

Last update mentioned is February 11, 2021, for the addition of detector cards and simulation files.