sPHENIX tracking discussion

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

    Dear All,
      
          Attached is my poster abstract for TPC NN clustering.
          Though I haven't gotten a performance advantage over the
      existing clustering method, I will talk about the motivation and
      technical implementation of NN clustering at sPHENIX. In the
      meanwhile before QM, I will improve its performance and hopefully
      I can achieve better performance before QM.
      
      Titile: Clustering hits of Time Projection Chamber by
      machine learning and artificial neural networks at sPHENIX
      Abstract: The Time Projection Chamber (TPC) at sPHENIX
      coverages pseudorapidity $|\eta| <$ 1.1. TPC is the major
      tracking detector and plays a key role in jet and heavy-flavor
      measurements. Charged particles passing through the TPC ionize
      electrons, and their positions are calculated using electron-drift
      time. These ionized electrons produce hits that form clusters for
      track reconstruction. The traditional method of grouping connected
      hits into clusters, known as connected component analysis (CCA),
      becomes less effective in high-multiplicity events, such as Au+Au
      collisions with pileup, due to effects like $\delta$-electrons. A
      neural network (NN) clustering, which uses an NN to predict the
      cluster position based on the distribution of hits, is supposed to
      improve the clustering performance. We simulate high-multiplicity
      events and sPHENIX detector responses and train the NN to predict
      the associated truth cluster position based on the distribution of
      the reconstructed hits. I will show the implementation of NN
      clustering at sPHENIX and our plan to enhance its performance by
      improving truth-information association and fine-tuning the
      parameters of the NN.
      
      Best regards,
      Zhongling

Attachment: sphenix-abs.pdf
Description: Adobe PDF document