Datasets:

zixinding
/

CMS-trigger-l1

license: cc-by-4.0
pretty_name: Datasets for Self-Driving Trigger Study at L1
tags:
  - physics
  - high-energy-physics
  - anomaly-detection
  - reinforcement-learning
  - CMS-open-data
size_categories:
  - 1B<n<10B

Datasets for Self-Driving Trigger Study at L1

Mirror of Zenodo record 10.5281/zenodo.17399948 — datasets for Learning to Trigger: Reinforcement Learning at the Large Hadron Collider (arXiv:2606.23993).

Derived from CMS 2016 Open Data for Level-1 (L1) hadronic objects (jets). Each file contains reconstructed jet features and the number of primary vertices (N_PV) per event.

Files

File	Size
`HToAATo4B.h5`	70.9 MB
`MinBias_1.h5`	354.0 MB
`MinBias_2.h5`	368.1 MB
`TT_1.h5`	175.7 MB
`Trigger_food_Data.h5`	79.8 MB
`Trigger_food_MC.h5`	210.1 MB
`data_Run_2016_283408_longest.h5`	173.1 MB
`data_Run_2016_283876.h5`	55.4 MB

MinBias_1.h5 — min-bias MC background (AD autoencoder training).
MinBias_2.h5 — alternate min-bias MC background (control-algorithm studies).
TT_1.h5 — Standard Model t-tbar hadronic signal.
HToAATo4B.h5 — BSM H→AA→4b signal.
data_Run_2016_283876.h5 — real CMS 2016 run (AD training, real background).
data_Run_2016_283408_longest.h5 — longest CMS 2016 run (control-algorithm testing).
Trigger_food_MC.h5 — precomputed control variables (anomaly score, HT, N_PV) for MC.
Trigger_food_Data.h5 — precomputed control variables for real data (matched by N_PV).

Loading

import h5py
from huggingface_hub import hf_hub_download

path = hf_hub_download(repo_id="zixinding/CMS-trigger-l1", filename="Trigger_food_MC.h5",
                       repo_type="dataset")
with h5py.File(path, "r") as f:
    print(list(f.keys()))

Citation

@misc{ding2026learning,
  title         = {Learning to Trigger: Reinforcement Learning at the Large Hadron Collider},
  author        = {Ding, Zixin and Emami, Shaghayegh and Salvi, Giovanna and Tosciri, Cecilia and Gandrakota, Abhijith and Ngadiuba, Jennifer and Tran, Nhan and Herwig, Christian and Miller, David W. and Chen, Yuxin},
  year          = {2026},
  eprint        = {2606.23993},
  archivePrefix = {arXiv},
  primaryClass  = {cs.LG},
  url           = {https://arxiv.org/abs/2606.23993}
}