sPHENIX tracking discussion

[Zhongling Ji <zji AT physics.ucla.edu>]

    Dear Joe,
      
          Thank you very much for these suggestions and comments. I
      uploaded the updated poster in Invenio:
      
      https://sphenix-invenio.sdcc.bnl.gov/records/26crv-6mm44
      
      
          I used the fonts from the template and thought I cannot change
      them. Now I used a larger font and rearranged the layout. I think
      it looks much nicer now.
      
          I implemented all suggestions from you. For the last figure,
      the units are the phi and z binWidth. I put them in the labels.
          The plotting is embedded in my PyTorch training macro and it
      takes some efforts to export to ROOT macros. I will do this in the
      future.
      
          I am waiting for the approval to print but welcome for further
      suggestions and comments.
      
      Best regards,
      Zhongling

On 8/28/23 9:22 AM, Osborn, Joseph wrote:

Hi Zhongling,

 

Thanks for the nice poster draft. Here are a few comments that will hopefully improve its readability.

 

In general, the text may be a little small (what size font did you use?). Admittedly it is hard to really know how large the text will turn out before printing it, though. Nonetheless, it looks like you have some white space in the various boxes that could be used to enlarge the text.

 

In TPC Event Display section:

 

“…figure shows the TPC hits…” -> “… figure shows TPC hits…” “Due to the high-collision…” -> “Due to the high collision rate at RHIC and thus pile up in the TPC, clustering hits is a difficult task.” “Using neural networks…” -> “This motivates using neural networks to perform clustering in the TPC.”

 

In Traditional clustering section:

“…two sides of TPC under electric field.” -> “…two sides of TPC under the electric field.” “Traditional clustering method” -> The current clustering method…

 

In the delta electrons section:

You may also want to point out that delta electrons can not only produce two peaks but widely distributed charge, which can be difficult to deal with for even the traditional clustering algorithm even in a clean case with no overlapping tracks. Perhaps “delta-electrons can produce two peaks or large clusters, which can make cluster centroid identification difficult for traditional clustering algorithms.” I would change the third sentence to “This motivates using a NN clusterizer to identify and improve cluster position resolution in these cases.”

 

NN Structures section

Each hit has an ADC value… -> Each hit has an ADC value that is proportional to the energy deposited and a phi-z (drift time) coordinate. Output1: Whether having true clusters -> What exactly do you mean here? Part of the output is whether or not the input ADC and coordinates would actually lead to a real reconstructed cluster being formed? If my understanding is correct, then I suggest “Output1: Whether a true cluster was present.” And “Output2: The reconstructed cluster phi and z position.”

 

NN Clustering Performance section

We can apply different cuts on it… -> “Applying different cuts on this value gives different rates of signal (true clusters) to background (background cluster). Increasing this value increases the signal efficiency but also includes more background.” The NN outputs are phi and z positions. -> “The NN outputs are the reconstructed cluster phi and z position.” Suggest changing “predicted” to “reconstructed” here, since in each case (for NN and CCA) the output is some reconstructed cluster position that we want to compare to the truth. These plots need units on the axis labels, and is it possible to make them in root using the sPHENIX style macro? Also presumably the deltaphi is actually deltaRphi, since our clustering resolution is not ~0.3 radians but is something like a few hundred microns (I assume the axes here are in unit mm).

 

Conclusions and Outlook section

Last point – “We aim to improve the performance of the NN clustering with better truth-reconstructed cluster association.”

 

 

___________________________

Joe Osborn, Ph.D

Physics Department

Brookhaven National Laboratory

josborn1 AT bnl.gov

 

 

From: sPHENIX-tracking-l <sphenix-tracking-l-bounces AT lists.bnl.gov> on behalf of Zhongling Ji via sPHENIX-tracking-l <sphenix-tracking-l AT lists.bnl.gov>
Date: Thursday, August 24, 2023 at 7:21 PM
To: sphenix-tracking-l AT lists.bnl.gov <sphenix-tracking-l AT lists.bnl.gov>
Subject: [Sphenix-tracking-l] QM poster

Dear All,

    I uploaded my poster about TPC NN clustering: https://sphenix-invenio.sdcc.bnl.gov/records/jywyf-1h828

    The related note is also uploaded: https://sphenix-invenio.sdcc.bnl.gov/records/tazst-m2y09

    Welcome for your suggestions and comments.

Best regards,
Zhongling