Winners
NeurIPS 2024 Higgs Uncertainty Challenge has come to an end. We have announced the winners of the competition. After extensive studies on a new hold-out data set, HEPHY and IBRAHIME cannot be separated in a significant way and are declared joint first. HZUME has secured third position.
| Medal | Rank | Team | Avg Coverage | Avg Interval | Avg Quantile Score |
|---|---|---|---|---|---|
 |
1 (Tie) | HEPHY | 0.6683 | 0.4599 | -0.5823 |
|
1 (Tie) | IBRAHIME | 0.6698 | 0.4974 | -0.5761 |
 |
3 | HZUME | 0.6659 | 0.8134 | -2.1650 |
- HEPHY (Lisa Benato, Cristina Giordano, Claudius Krause, Ang Li, Robert Schöfbeck, Dennis Schwarz, Maryam Shooshtari, Daohan Wang) from Vienna’s Institute of High Energy Physics (HEPHY) in Austria will win $2000.
- IBRAHIME (Ibrahim Elsharkawy) from University of Illinois at Urbana-Champaign will win $2000.
- HZUME (Hashizume Yota) from Kyoto University Japan will win $500.
Papers documenting the winning solutions are being prepared and will be linked here. The dataset will be released permanently on Zenodo to serve as a permanent benchmark. All winners will present at Fair Universe HiggsML Uncertainty CERN workshop.
About the Challenge
This NeurIPS 2024 Machine Learning competition is one of the first to strongly emphasise mastering uncertainties in the input training dataset and outputting credible confidence intervals. This challenge explores uncertainty-aware AI techniques for High Energy Physics (HEP).
The context is the measurement of the Higgs Boson signal like in HiggsML challenge on Kaggle in 2014. Participants should design an advanced analysis technique that can not only measure the signal strength but also provide a confidence interval
The confidence interval should include statistical and systematic uncertainties (concerning detector calibration, background levels, etc…). It is expected that advanced analysis techniques that can control the impact of systematics will perform best. This challenge presents an opportunity to push the boundaries of machine learning applications within physics while still focusing on essential ML skills like robust model development and uncertainty quantification.
Organizers
We gratefully acknowledge the efforts of the FAIR Universe Team, composed of researchers dedicated to advancing high energy physics, cosmology, and machine learning for the benefit of the scientific community.
For inquiries, please contact us at: fair-universe@lbl.gov
- Aishik Ghosh
- Benjamin Nachman
- Benjamin Thorne
- Chris Harris
- Daniel Whiteson
- David Rousseau
- Elham E Khoda
- Ihsan Ullah
- Isabelle Guyon
- Jordan Dudley
- Mathis Reymond
- Paolo Calafiura
- Peter Nugent
- Po-Wen Chang
- Ragansu Chakkapai
- Sascha Diefenbacher
- Shih-Chieh Hsu
- Steven Farrell
- Wahid Bhimji
- Yuan-Tang Chou
- Yulei Zhang
Competition Material
Find all the competition resources below:
- Codabench Challenge Page: This serves as the platform to submit entries to the competition.
- Tutorial Slides: These slides will help you register and submit a sample dummy submission.
- Training Data (6.5 GB): Download the training data if you you want to experiment with the data on your local machines.
- Documentation: This contains detailed information about the science behind the challenge, the specifics of the data, and documents the code used to facilitate the evaluation of the competition. It also describes the evaluation metric.
- GitHub Repository: This hosts the code for testing submissions, as well as the starting kit notebook. The starting kit is also available on Google Colab.
- White Paper: This serves as a full breakdown of the competition in detail.