SciPost Submission Page
Advancing the CMS Level-1 Trigger: Jet Tagging with DeepSets at the HL-LHC
by Stella Schaefer, Christopher Brown, Duc Hoang, Sioni Summers, Sebastian Wuchterl
This is not the latest submitted version.
Submission summary
| Authors (as registered SciPost users): | Stella Felice Schaefer |
| Submission information | |
|---|---|
| Preprint Link: | https://arxiv.org/abs/2509.24371v1 (pdf) |
| Code repository: | https://github.com/CMS-L1T-Jet-Tagging/TrainTagger |
| Date submitted: | Sept. 30, 2025, 7:42 a.m. |
| Submitted by: | Stella Felice Schaefer |
| Submitted to: | SciPost Physics Proceedings |
| Proceedings issue: | The 2nd European AI for Fundamental Physics Conference (EuCAIFCon2025) |
| Ontological classification | |
|---|---|
| Academic field: | Physics |
| Specialties: |
|
| Approach: | Experimental |
Abstract
At the High Luminosity LHC, selecting important physics processes such as (di-) Higgs production will be a high priority. The Phase-2 Upgrade of the CMS Level-1 Trigger will reconstruct particle candidates and use pileup mitigation for the 200 simultaneous proton-proton interactions. A fast cone algorithm will reconstruct jets from these particles, providing access to jet constituents for the first time. We introduce a new multi-class jet tagger with a small, quantized DeepSets neural network. The tagger, trained on a mix of simulated CMS events, predicts various hadronic and leptonic classes. We present the tagger, its performance, and its improvements for triggering on (di-) Higgs events.
Current status:
Reports on this Submission
Strengths
processing, is always appreciated and valuable.
Weaknesses
- Model design procedure is not detailed.
- FPGA deployment/synthesis workflow is not provided.
- The data itself is not described.
- The above weaknesses are most likely resulting from the page constraints. However, there is enough room to improve the data description.
Report
Requested changes
1- Introduction: Capitalise "particle flow (PF)" -> "Particle Flow (PF)" 2- All sections and subsection: These are not titles, thus only the first word is to be capitalised. 3- Section 2: You have mentioned that 10% of the data is reserved for testing and validation. What are the portions, individually? As testing and validation sets cannot be mixed. If the considered test set is rather small, it will not mitigate the risk of overfitting. 4- Section 2: Regarding the padding, is it constant or variable? 5- Section 2.1: "hardware and latency constraints" What are these constraints? Provide numbers. 6- Section 2.1: "DeepSet" is written as two separate words in the rest of the paper. 7- Section 3: "Fig. 2a" You have abbreviated figure term for sub-figures. Needs to be adjusted in LaTeX packages. 8- Section 3, paragraph 2: A new paragraph has to start indented. 9- Section 3: The class balance statistics of the dataset is not provided. How many classes and what are the proportions? 10- Generally speaking, plots are too small to be legible. 11- Section 5: Provide a reference or a footnote to the FPGA's specification page/document from the manufacturer. 12- Section 5: "models" -> "model's" 13- A code repository is linked to the SciPost submission. This should be added to the paper either as a reference (preferred if the link is permanent), or a footnote.
Recommendation
Ask for minor revision