Skip to Content.
Sympa Menu

sphenix-physics-l - [Sphenix-physics-l] Physics Roundup - January 2022

sphenix-physics-l AT lists.bnl.gov

Subject: sPHENIX discussion of physics

List archive

Chronological Thread  
  • From: "Perepelitsa, Dennis" <dvp AT bnl.gov>
  • To: "sphenix-physics-l AT lists.bnl.gov" <sphenix-physics-l AT lists.bnl.gov>
  • Subject: [Sphenix-physics-l] Physics Roundup - January 2022
  • Date: Tue, 25 Jan 2022 13:55:07 +0000

Dear sPHENIX collaborators,

At the end of the January Collaboration Meeting, during the Topical Group and MDC-II reports, we started a bit of a discussion about the use Machine Learning (ML) methods for physics measurements. These are widely used (and uncontroversial) in, for example, reconstruction of heavy flavor decays.

However, the particular focus of the discussion was on improving jet measurements, such as the jet resolution at low-pT. In this case, there is a significant potential bias from the training set (for example, made using Pythia) looking particularly different than the real data (due to physics quenching effects) which needs to be studied.

Here I wanted to collect some relevant literature and invite people to continue that discussion.

The main paper mentioned was this one: https://inspirehep.net/literature/1698396 . It proposes an ML-based method to better estimate the pT of a jet in HI background based on the properties of the particles in the jet cone, and thus reduce the experimental resolution. The argument is that one can then push to lower-pT and/or larger-R than in a more typical subtraction method. 

Here is a nice pair of seminars in 2020 from ALICE collaborators on this topic: 

Raymond Ehlers (ORNL): https://indico.phy.ornl.gov/event/10/attachments/236/790/2020-08-27.rehlers.ORNL_nuclear_physics_seminar.v4.pdf  
Hannah Bossi (Yale): https://indico.phy.ornl.gov/event/10/attachments/236/788/HBossi_ORNLSeminar_2020.pdf 

These talks also describe some studies to assess the sensitivity of the results to the assumed fragmentation pattern in the training sample.

I would be curious to hear from folks who are knowledgable about this particular ALICE study, or the area more broadly - is there interest in doing this kind of thing in sPHENIX?

Dennis

P.S. Here are some more general resources: 

Review of Modern Physics on ML and the physical sciences: https://journals.aps.org/rmp/abstract/10.1103/RevModPhys.91.045002  
The Particle Data Group review of Machine Learning - a new installment in the recent PDG version: https://pdg.lbl.gov/2021/web/viewer.html?file=../reviews/rpp2021-rev-machine-learning.pdf  
A “living review” of ML in HEP hosted on a GitHub page and actively updated: https://github.com/iml-wg/HEPML-LivingReview  




  • [Sphenix-physics-l] Physics Roundup - January 2022, Perepelitsa, Dennis, 01/25/2022

Archive powered by MHonArc 2.6.24.

Top of Page