SciPost Submission Page
Quark-Gluon tagging performance at the High-Luminosity LHC using constituent-based transformer models
by F. L. Castillo, J. Levêque
Submission summary
| Authors (as registered SciPost users): | Florencia L. Castillo |
| Submission information | |
|---|---|
| Preprint Link: | https://arxiv.org/abs/2509.14759v1 (pdf) |
| Date submitted: | Sept. 19, 2025, 9:35 a.m. |
| Submitted by: | Florencia L. Castillo |
| 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
Jet constituents provide a more detailed description of a jet's radiation pattern than global observables. In simulations for ATLAS Run-2 data (2015-2018), transformer-based taggers trained on low-level inputs outperformed traditional methods using high-level variables with conventional neural networks for quark-gluon discrimination. With the upcoming High-Luminosity LHC (HL-LHC), which will deliver higher luminosity and energy, the ATLAS detector will be upgraded with an extended Inner Tracker covering the forward region, previously uncovered by a tracking detector. This work studies how these upgrades will improve the accuracy and robustness of quark-gluon jet taggers.
Current status:
Reports on this Submission
Strengths
1- Paper well written 2- checking the robustness against pileup is very interesting, and the results look promising
Weaknesses
Report
Requested changes
1- consider citing an ATLAS detector or reconstruction paper to introduce "topo-clusters" and "topo-towers" (section 2) 2- section 2 : consider replacing ”these are referred as jet constituents" with "these are referred to as jet constituents" or "these are called jet constituents " 3- section 3 : have the text and plot be more coherent with respect to the tagger type. Section 3.1 mentions "ParT" but the associated figure 1 shows only "ParT Const". 4- section 3.3/ Figure 3: consider adding an explanation for the non-monotonous pileup dependency for low pT jets at high rapidity (performance at 200 PU seems better than at 140) 5- there seems to be an extra space between second author name and the "1" below the title
Recommendation
Publish (meets expectations and criteria for this Journal)
Report #2 by Santiago Folgueras (Referee 1) on 2025-11-18 (Invited Report)
Report
Thanks for a nice manuscript, the paper is well written, the results seem promising. I would like to clarify, however, few points with the authors that I belive it will help them improve the readibility of the paper:
1) I miss a longer description of the taggers, Table 1 provides some description but it might be useful to be more clear on what kind of information enters, for example, for a non-ATLAS expert what is a "topo-tower", aren't PFOs made from topo-towers/clusters and tracks?
2) What is the reason of the ParT discriminator increase in gluon jet rejection shown on Figure 1 (left) around 1.7? Similarly when showing the dependence vs number of interactions (Figure 3), why the PU=200 is better than PU=140 at higher rapidity?
Requested changes
1- include a definition of what a topo-tower is? and why are not incorporated into the PFOs? 2-define what low-pt and high-pt the first time it appears, section 3. 3-Provide a more in-detail information on how the ParT discriminator is built, it is not clear how the extra information from the forward tracker is used.
Recommendation
Ask for minor revision