Jet Tagging studies for VBF Higgs -> WW -> 2l2nu Neural Network for JetTagging

Kind of NN

Up to now three different NN have been tested, each one with pros and cons:
  1. NN on single jet
  2. NN on a pair of jets with some information from the remaining jets
  3. NN on all pair of jet (the first 4 pt ordered pairs of jets)

Analysis steps

Common steps for all NN approaches

Single Jet NN

  1. Creation of a TTree to create input variables for NN
    • Tree producer
    • example:
       JetTagging_NN_SingleTag_Creator 0.4 input.root output.root 
      where 0.4 is the max ΔR for MC matching with MC tagged quarks
  2. Training of the NN TMultiLayerPerceptron (H160WW2l sample)
    • root macro
    • 24k events training (5.5k signal + 18.5k background) + 6k events test
    • output of the training is a set of weights. Otherwise the functions.cxx and functions.h
  3. Test of the NN TMultiLayerPerceptron
  4. PostTraining of the NN TMultiLayerPerceptron
  5. Test of the NN in CMSSW (just a function) -> jet pair with maximum NN output
    • bin code -> comparison between Pt max, Mjj max e NN max methods

Results

img7145484ef32e621cefb310e840c14351.png
H160
imgf066d37bda8d125dc2c76e4e43e71385.png
H170
imgc1aa23dead44959a21328ef29355ff89.png
H200
img45fc6bd6b6f14bb4cd21b4c95e7cc600.png
H500

  • Purity: given a VBF event, number of events with two right jets tagged / total number of VBF events. L2+L3 corrected jets (NN not yet trained, but old trained one reused)
img2423eb76d493c6efa91a56d9788bb3a0.png
H160
img63eb3ee778e387a96e78e308cbcced03.png
H170
imga71e3d7d973f2cbade90a0d3e17df73a.png
H200
imgd9ccd455a210c331de1d5365a234f6ee.png
H500

* Purity: given a VBF event, number of events with two right jets tagged / total number of VBF events. L2+L3 corrected jets (NN re-trained )

img6785e4034cd0e0c12de1a9e2ad633f0a.png
H160
imgff0980c5538e07ac0155e13424922196.png
H170
img5f6d3b5d9755cbb691de4c262234bed5.png
H200
img27b4b17e40e1503a9fe315bfaccb9c89.png
H500

Multi Pair Jet NN

Single Pair Jet NN

  1. Creation of a TTree to create input variables for NN
    • Tree producer
    • example:
       JetTagging_NN_SinglePairTag_Creator 0.4 input.root output.root 
      where 0.4 is the max ΔR for MC matching with MC tagged quarks
  2. Training of the NN TMultiLayerPerceptron (H160WW2l sample)
    • root macro
    • 24k events training + 6k events test
    • output of the training is a set of weights.
  3. Test of the NN TMultiLayerPerceptron
  4. Test of the NN in CMSSW -> jet pair with maximum NN output
    • bin code -> comparison between Pt max, Mjj max e NN max methods

Results

H160WW2l TagReco NNMjjPt Purity.png
H160
H170WW2l TagReco NNMjjPt Purity.png
H170
H200WW2l TagReco NNMjjPt Purity.png
H200
H500WW2l TagReco NNMjjPt Purity.png
H500
    • L2+L3 energy correction samples
Sample Image
H130_tautau_2l/Summer08_IDEAL_V9_v1/GEN-SIM-RECO L2+L3

-- AndreaMassironi - 04 Mar 2009

Edit | Attach | Watch | Print version | History: r8 < r7 < r6 < r5 < r4 | Backlinks | Raw View | WYSIWYG | More topic actions
Topic revision: r8 - 2009-03-16 - AndreaMassironi
 
    • Cern Search Icon Cern Search
    • TWiki Search Icon TWiki Search
    • Google Search Icon Google Search

    Main All webs login

This site is powered by the TWiki collaboration platform Powered by PerlCopyright & 2008-2019 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding TWiki? Send feedback