ProtoDUNE pi0 ML

Tuesday 26 Mar 2019, 11:00 → 12:00 America/Los_Angeles

https://stanford.zoom.us/j/6509264799

2019-03-06
==========
https://indico.slac.stanford.edu/event/135/

Status of ProtoDUNE Pi0 ML (Patrick)
------------------------------------
Update on converting larsoft data to larcv format.
Demonstrate how ML can help to remove ghost space points.
Show a corner case where time-based chi2 doesn't perform well.

-  Mike: Does cluster3D consider charge conservation of the wires? (p.5)
-  Patrick: No. Only time info is taken for the chi2 calculation
-  Mike admired the ML technique to handle the de-ghosting problem.
   It would be nice if we could demonstrate a full analysis in this direction.  
   He also asked how Cluster3D compare to others.
-  Tracy clarified that Cluster3D aims to keep the efficiency over purity,
   while others (SpacePointSolver, WireCell) do the opposite.
   The strategy is to keep as many 3D points as possible, and let the ML
   algo. decides. He will discuss with Chris Backhouse in person during
   SBND workshop.

AOB
---
-  Kazu is visiting CSU next week.
-  Mike & Kazu will give a summary on the visit (in next or the following  meeting).
-  Meeting adjourned at 11:25 PDT.

 

11:00 → 11:20 Stauts of ProtoDUNE Pi ML

Speaker: Patrick Tsang (SLAC)

2019-03-26 Status of ProtoDUNE Pi0 ML.pdf