Title: BJamboree tagger
1B-jet tagging with muons
- Efficiency of jet and muon reconstruction (bb???
X) - Discriminating variables
- Current performance
- Short look into neural networks
- Outlook
2Jet reconstruction efficiency
- 0.7 cone algorithm
- Recod jet compared with mc jet within 0.2 cone
- 75 efficient
3Muon reco efficiency
4Muon reco efficiency (ctd)
- Total muon reconstruction efficiency 70
- Combination of Gtrack efficiency and muon
efficiency - Efficiency for jet? reconstruction 52
5Discriminating variables
- Variables under consideration
- PTRel
- P? / Ejet
- DCA, DCA-significance
- DZ, DZ-significance
- Current implementation only uses PTRel
6PTRel
7P ? / Ejet
8DCA DZ
9DCA DZ significance
10Current tagger performance
bb?? ? X, 1200 evts, 1843 b-jets, 1131 l-jets
11Current tagger performance
bb?? X, 1500 evts, 2373 b-jets, 1446 l-jets
12Tagger Performance
No requirements on ?
13Neural net performance
- Use a 9-input, 18 hidden, 1 output NN
- Inputs are discriminating variables
- Trained NN on 950 bbar events
- Tested NN on 250 bbar events (not the same!)
- Caveat backgrounds (light jets) come from the
same bbar file
14Neural net output
- Solid blue background
- White open signal
15Neural net performance
bb?? ? X, 200 evts, 320 b-jets, 77 l-jets
Grey Values from currently implemented tagger
16Neural net performance
bb?? X, 1500 evts, 2373 b-jets, 1446 l-jets
Grey Values from currently implemented tagger
17Outlook
- Discriminating variables are shaping up
- Neural network is looking promising
- Need more MC, especially background
- cc
- QCD di-jet