Login

Calorimeter Machine Learning

→ Europe/Zurich
Hide

1. Need to investigate the importance of different features used in BDT (especially ECAL moment Y5).

2. Use more samples to train BDT and NN.

3. The output distribution, ROC and the accuracy of NN are not consistent, need further investigation (perhaps by zooming out at 1.0 on output distribution plot).

4. Search in the both BDT and NN hyper-parameter spaces for a fair comparison.

5. The generator doesn't work when MaxNumEvents is actually equal to the total number of events in FileSearch.

There are minutes attached to this event. Show them.
    • 19:00 → 19:20
      Maurizio 20m
      Speaker: Maurizio Pierini (CERN)
    • 19:20 → 19:40
      Ben 20m
      Speaker: Benjamin Henry Hooberman (Univ. Illinois at Urbana-Champaign (US))
    • 19:40 → 20:00
      Amir 20m
      Speaker: Amir Farbin (University of Texas at Arlington (US))
    • 20:00 → 20:20
      Jean-Roch 20m
      Speaker: Dr Jean-Roch Vlimant (California Institute of Technology (US))
    • 20:20 → 20:40
      Ryan 20m
      Speaker: Ryan Reece (University of California,Santa Cruz (US))
    • 20:40 → 21:00
      Matt 20m
      Speaker: Matt Zhang (Univ. Illinois at Urbana-Champaign (US))
      (2017-09-22) BDT Variable Bug Fixes.pdf
    • 21:00 → 21:20
      Taylor 20m
    • 21:20 → 21:40
      Wei Wei 20m
      CaloML-Aug-10-17.pdf
Powered by Indico v3.3.2-pre