SciPost Submission Page
Reconstructing partonic kinematics at colliders with Machine Learning
by David F. Rentería Estrada, R. J. Hernández-Pinto, German F. R. Sborlini and P. Zurita
This Submission thread is now published as
Submission summary
| Authors (as registered SciPost users): | David Rentería · German Sborlini |
| Submission information | |
|---|---|
| Preprint Link: | scipost_202202_00023v3 (pdf) |
| Date accepted: | 2022-08-15 |
| Date submitted: | 2022-07-05 20:24 |
| Submitted by: | Rentería, David |
| Submitted to: | SciPost Physics |
| Ontological classification | |
|---|---|
| Academic field: | Physics |
| Specialties: |
|
| Approaches: | Theoretical, Computational, Phenomenological |
Abstract
In the context of high-energy physics, a reliable description of the parton-level kinematics plays a crucial role for understanding the internal structure of hadrons and improving the precision of the calculations. In proton-proton collisions, this represents a challenging task since extracting such information from experimental data is not straightforward. With this in mind, we propose to tackle this problem by studying the production of one hadron and a direct photon in proton-proton collisions, including up to Next-to-Leading Order Quantum Chromodynamics and Leading-Order Quantum Electrodynamics corrections. Using Monte-Carlo integration, we simulate the collisions and analyze the events to determine the correlations among measurable and partonic quantities. Then, we use these results to feed three different Machine Learning algorithms that allow us to find the momentum fractions of the partons involved in the process, in terms of suitable combinations of the final state momenta. Our results are compatible with previous findings and suggest a powerful application of Machine-Learning to model high-energy collisions at the partonic-level with high-precision.
Author comments upon resubmission
Dear Editor,
We have carefully considered the comments that both referees have provided us. We have made three main modifications in line with their suggestions:
-
Neural networks are defined already at the beginning of Section 4.4. For clarity, we have expanded their definition, incorporated the schematic representation in terms of layers and cite one of the pioneering works on the topic.
-
We included, also in Sec. 4.4 (fourth paragraph), a comment about the cost/loss function used in our implementation. We included the reference to the documentation of the package.
-
We replaced the figures in PNG format by EPS to increase the quality of the figures.
We hope that the modifications implemented serve to clarify the questions and address all the issues identified by the referees.
List of changes
Regarding the specific questions raised by the referees, we proceed to provide answers:
Referee 2:
1. Given that this article is about phenomenology, we consider the details pertaining to the process under study to be relevant for the proper understanding of the work. Therefore, we feel that Secs. 2 and 3 belong in the body of the article rather than in an appendix, as suggested by the referee.
2. Additional explanations about what a neural network is, are included in Sec. 4.4.
3. We find ourselves at loss with the comment about the references used in the introduction. The purpose of the references we included, such as the workshops, was to point out the broad application of ML in particle physics and the significant interest of the community. As such, we feel that the references selected serve to our intent more than citing specific articles. Far from us to claim that these are the only works in HEP. We would be very happy to include more references if the referees were kind enough to point them to us.
4. We included EPS graphs to improve the resolution.
5. In Sec. 4.5 we present an estimation of the reconstruction errors due to the scale uncertainty propagation (at the partonic level). Since K-factors are usually very large for these processes, they dominate over the intrinsic PDF/FF errors (in most of the cases). Thus, we restricted our attention to the scale uncertainty propagation. On the other hand, changing the PDF/FF sets would have led to a much more complex and computationally demanding analysis, that is far beyond the scope of the current work. In fact, it would be another article altogether. Finally, regarding the discussion in App. A, we prefer to keep it as an appendix to avoid overloading the main text with rather technical (although important) discussions.
6. The comment can be found in the second paragraph of Sec. 4, discussing the kinematics of the different collisions.
Referee 1:
1. A comment about the implementation of the loss/cost function was included in Sec. 4.4 (fourth paragraph).
Published as SciPost Phys. Core 5, 049 (2022)
Reports on this Submission
Report #2 by Anonymous (Referee 6) on 2022-8-2 (Invited Report)
- Cite as: Anonymous, Report on arXiv:scipost_202202_00023v3, delivered 2022-08-02, doi: 10.21468/SciPost.Report.5484
Report
Dear Authors,
I thank you for adding the loss information I've requested. Stating the loss function is important because it states the desired goal of your analysis in a statistics language, and therefore, it is necessary for reproducing your results. Using a different loss function for your regression may result in estimators with different statistical characteristics (for example, mean square error vs. mean absolute error.) The physical conclusion may change if the results are getting sensitive to the characteristics.
The authors answered all the issues raised from my side, so I recommend the draft for publication.
Report #1 by Anonymous (Referee 5) on 2022-7-6 (Invited Report)
- Cite as: Anonymous, Report on arXiv:scipost_202202_00023v3, delivered 2022-07-06, doi: 10.21468/SciPost.Report.5341
Report
Thank you for taking into account my feedback. As I said in my last response, I think the paper could be significantly streamlined, but I won't insist.
Sorry, I had a typo in my last post - it should have read "It is NOT necessary to explain what a neural network is in the main body." (as I wrote in my first set of comments). I won't insist on this point, but it is odd to have such pedagogical descriptions in a paper of this type.
For the citations, I really don't think citing a website is appropriate in this case. There are plenty of review articles that would be reasonable to cite. As above, I won't insist, but please think about citing some reviews instead of websites.