SciPost Submission Page
Precision-Machine Learning for the Matrix Element Method
by Theo Heimel, Nathan Huetsch, Ramon Winterhalder, Tilman Plehn, Anja Butter
Submission summary
Authors (as registered SciPost users): | Tilman Plehn · Ramon Winterhalder |
Submission information | |
---|---|
Preprint Link: | https://arxiv.org/abs/2310.07752v2 (pdf) |
Date submitted: | 2023-10-24 07:42 |
Submitted by: | Winterhalder, Ramon |
Submitted to: | SciPost Physics |
Ontological classification | |
---|---|
Academic field: | Physics |
Specialties: |
|
Approaches: | Computational, Phenomenological |
Abstract
The matrix element method is the LHC inference method of choice for limited statistics. We present a dedicated machine learning framework, based on efficient phase-space integration, a learned acceptance and transfer function. It is based on a choice of INN and diffusion networks, and a transformer to solve jet combinatorics. We showcase this setup for the CP-phase of the top Yukawa coupling in associated Higgs and single-top production.
Current status:
Reports on this Submission
Strengths
1. Explores an innovative approach for extracting fundamental parameters of the underlying Lagrangian from an event sample
2. Contains a concise description of the machine learning methods applied
Weaknesses
1. The description of the method used for generating the event sample is lacking
2. It is unclear whether the results of the developed method relies on training of the network and the check of the extraction relies on both data sets being generated by the same method.
Report
This submission is a well presented study developing machine learning techniques to extract parameters of the underlying Lagrangian from a generated event sample using the matrix element method. The presentation is concise and well written, with only a few typos (e.g. "as well"->"as well as" in the second line of the introduction). Some of the acronyms used are not defined (e.g."cINNs" in the last paragraph of the introduction).
More importantly though, there is no description of how the event sample was generated - this is important, since it is unclear whether the try complexity of the process has been described correctly (and therefore whether the underlying parameter \alpha has been too easy to extract for real-world applications to be relevant). Was the process modelled with a parton shower? At what perturbative precision? (Born or NLO+shower, merged MEPS?). A description of the choices made for the generation would need to be added.
Furthermore, it would seem prudent to check to what extent the conclusions on the usefulness of the method depends on the training and the checks being performed on the same sample (or samples generated with the same choices, so differing only by statistics).
Requested changes
1. describe the method used for generating the event sample(s)
2. check the quality of the extracting for a sample generated with the same choice for couplings, but using a different generator (Herwig/Sherpa/Pythia) or perturbative matching (CKKWL with high-multiplicity matching/MC@NLO etc).