LHC New Physics Dataset

← Back to all benchmarks

Date: 2021-07-05

Name: LHC New Physics Dataset

Domain: High Energy Physics

Focus: Real-time LHC event filtering for anomaly detection using proton collision data

Task Types: Anomaly Detection, Event classification

Metrics: ROC-AUC, Detection efficiency

Models: Autoencoder, Variational autoencoder, Isolation forest

AI/ML Motif: Anomaly Detection

Resources

Benchmark: Visit

Results: Gemini LLM Deep Research

Keywords

anomaly detection proton collision real-time inference event filtering unsupervised ML

Citation

Thea Aarrestad, Ekaterina Govorkova, Jennifer Ngadiuba, Ema Puljak, Maurizio Pierini, and Kinga Anna Wozniak. Unsupervised new physics detection at 40 mhz: training dataset. 2021. URL: https://zenodo.org/record/5046389, doi:10.5281/ZENODO.5046389.

@misc{https://doi.org/10.5281/zenodo.5046389,
  author    = {Aarrestad, Thea and Govorkova, Ekaterina and Ngadiuba, Jennifer and Puljak, Ema and Pierini, Maurizio and Wozniak, Kinga Anna},
  copyright = {Creative Commons Attribution 4.0 International},
  doi       = {10.5281/ZENODO.5046389},
  publisher = {Zenodo},
  title     = {Unsupervised New Physics detection at 40 MHz: Training Dataset},
  url       = {https://zenodo.org/record/5046389},
  year      = {2021}
}

Ratings

CategoryRating

Software

3.00

While not formally evaluated in the previous version, Zenodo and paper links suggest available code for baseline models (e.g., autoencoders, GANs), though they are scattered and not unified in a single repository.

Specification

3.00

The task and context are clearly described, but system constraints and formal inputs/outputs are not fully specified.

Dataset

5.00

Large-scale dataset hosted on Zenodo, publicly available, well-documented, with defined train/test structure. Appears to follow at least 4 FAIR principles.

Metrics

4.00

Uses reasonable metrics (ROC-AUC, detection efficiency) that capture performance but lacks full explanation and standard evaluation tools.

Reference Solution

2.00

Baselines are described across multiple papers but lack centralized, reproducible implementations and hardware/software setup details.

Documentation

3.00

Some description in papers and dataset metadata exists, but lacks a unified guide, README, or training setup in a central location.

Average rating: 3.33/5

Radar plot

$LHC New Physics Dataset radar$

Edit: edit this entry