An Artial Neural Network for Plug Electron Identication

prudencewooshΤεχνίτη Νοημοσύνη και Ρομποτική

19 Οκτ 2013 (πριν από 4 χρόνια και 24 μέρες)

68 εμφανίσεις

AnArti?alNeuralNetworkfor
PlugElectronIdenti?cation
YvesKemp
,ThomasM¤uller,HartmutStadie,WolfgangWagner
Universit
¤
atKarlsruhe
AnArtialNeuralNetworkforPlugElectronIdenticationp.1/15
Useofplugelectronsforanalysis
eta
-1-0.500.51
n
0
100
200
300
400
500
600

-distributionofelectrons
in4.11versionofsingletop
analysis(afterpreselection)
eta
-5-4-3-2-1012345
n
0
50
100
150
200
250
300
350

-distributionofelectrons
int-channelMC.
ca30%with
jj>1:1
AnArtialNeuralNetworkforPlugElectronIdenticationp.2/15
Selectionofthesamples
Signalsample:
1tightcentralelectronwithtrack
Anotherelectroncandidateinplug(Z-Candidate)
Cutstobeindependentoftriggercuts
2000eventsremain
Backgroundsample:
2balancedjets(1central,1plug)
Severalpreselectioncuts
30000eventsremain
Bothsamplestakenfromdata!(bpel08)
cdfsoft2versions4.9and4.11
AnArtialNeuralNetworkforPlugElectronIdenticationp.3/15
Controlplot:
ET
ofplugelectron
(GeV)
T
E
0102030405060708090100
ev. prob./bin
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
Signal
Background
AnArtialNeuralNetworkforPlugElectronIdenticationp.4/15
VariablesforplugelectronID
Fiducialcut:
1:2<jj<2:8
EmET
(Phoenix-)Track
HadE/EmE(slidingcut)
IsolationRatio
PEM
2
(comparisionwithtestbeamdata
PES5by9u(ShowerproleinPESinudirection)
PES5by9v(ShowerproleinPESinvdirection)
AnArtialNeuralNetworkforPlugElectronIdenticationp.5/15
Selectionvariables
Had/Em
00.010.020.030.040.050.060.070.080.090.1
ev. prob./bin
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
Isolation
00.050.10.150.20.250.30.350.40.450.5
ev. prob./bin
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
0.2
2
χPEM
012345678910
ev. prob./bin
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
PES 5by9 u
0.40.50.60.70.80.911.1
ev. prob./bin
0
0.02
0.04
0.06
0.08
0.1
0.12
PES 5by9 v
0.40.50.60.70.80.911.1
ev. prob./bin
0
0.02
0.04
0.06
0.08
0.1
0.12
Signal
Background
(normalizedto
sameintegral)
AnArtialNeuralNetworkforPlugElectronIdenticationp.6/15
Correlationmatrix
TargetHadEmIsoPEM
2
PES5/9uPES5/9v
Target
100.0-49.4-66.6-64.543.943.2
HadEm
100.052.844.8-24.8-24.2
Iso
100.072.0-38.9-38.5
PEM2
100.0-42.8-43.4
PES5/9u
100.0
45.9
PES5/9v
100.0
Targetis-1forbackground,1forsignal
PES 5/9 u
0.60.650.70.750.80.850.90.951
PES 5/9 v
0.6
0.65
0.7
0.75
0.8
0.85
0.9
0.95
1
Correlationbetweenthe
twoPES5/9variablesdue
tocrosstalkandgeometry
AnArtialNeuralNetworkforPlugElectronIdenticationp.7/15
Arti?cialNeuralNetwork
5inputvariables(+1biasnode)
rankcorrelation
Had/Em54.8
Isolation157
PEM2
314
PES5/9u411
PES5/9v220
Preprocessingofthevariables
5nodesinintermediatelayer
Binaryclassication(-1background,1signal)
50iterations
NeuroBayesANNpackage(seeUlrichstalk)
AnArtialNeuralNetworkforPlugElectronIdenticationp.8/15
Arti?cialNeuralNetworkresults
Network output
-1-0.8-0.6-0.4-0.2-00.20.40.60.81
events
0
200
400
600
800
1000
1200
StandardCuts:
Signalefciency84%
Backgroundeff.5.3%
ANNcut
>
0.65:
Samesignalefciency
Backgroundeff.3.5%
34%lessbackground
ANNcut
>
0.40:
Samebackgroundeff.
Signalefciency91%
8%moresignal
AnArtialNeuralNetworkforPlugElectronIdenticationp.9/15
Independenttests
MET (GeV)
010203040506070
0
100
200
300
400
500
600
700
800
900
MissingET
NNcut:5679ev.<25GeV
standard:6483ev.<25Gev
ANN
cuts
sample:Allevents
withoneplugcan-
didate+track
W (GeV)
T
M
020406080100120140
0
50
100
150
200
250
300
350
400
TransverseW-Mass
(MET>15GeV)
NNcut:5738events
standard:5533events
AnArtialNeuralNetworkforPlugElectronIdenticationp.10/15
Backgroundestimation:4-sectormethod
Isolation
00.050.10.150.20.250.30.350.40.450.5
MET (GeV)
0
10
20
30
40
50
60
70
AB
CD
sample:Alleventswithoneplug
candidate+track
Allstandardcuts
exceptisolation
applied
A:
(MET<15GeV,Iso>0.2:501
B:
(MET<15GeV,Iso<0.1:4015
C:
(MET>20GeV,Iso>0.2:106
D:
(MET>20GeV,Iso<0.1:4156
)
NQCDinD
NTotalinD
=20:4%
ThismethoddoesnotworkforANN
cut
AnArtialNeuralNetworkforPlugElectronIdenticationp.11/15
QCDbackgroundestimation:Fitmethod
Idea:
FitbackgroundandMCtodata
Background:
InvertANNcut:<-0.95
MC:
W+jetssample
Fit:
Scalingfactorsforbackgroundand
MC,tMETininterval15-30GeV
ANN>0.5:
)
NQCDinMET>20GeV
NTotalinMET>20GeV
=15:3%
StandardCuts:
)
NQCDinMET>20GeV
NTotalinMET>20GeV
=17:9%
sample:Alleventswithoneplugcandidate+track
MET (GeV)
1618202224262830
ev. per 1 GeV
0
50
100
150
200
250
300
350
MET (GeV)
010203040506070
ev. per 2 GeV
0
100
200
300
400
500
600
700
800
-QCD
-DataminusQCD
++Data
-QCD
++Data
-Fit
-MC
AnArtialNeuralNetworkforPlugElectronIdenticationp.12/15
Jetmultiplicity
Exclusive jet multiplicity
0123456789
Events per jet bin
1
10
102
103
104
++QCD(ANN<-0.95)
-DataminusQCD
Jets
Data
QCD
Data-QCD
0
39693
2581
37112
1
7507
4204
3303
2
1976
1083
893
3
402
240
262
4
53
44
11
WithCDFstandardcuts
0
37960
2990
35870
2
1926
1255
673
sample:Alleventswithoneplugcandidate+track
Additionalrequirement:MET>20GeV
(realisticscenarioforW+jetsanalysis)
AnArtialNeuralNetworkforPlugElectronIdenticationp.13/15
Exclusive2-JetBin
data-QCD
190019201940196019802000202020402060
QCD events
950
1000
1050
1100
1150
1200
1250
1300
VaryingANNcutsuchthat
S=B
or
S=p
B
isoptimal

20%lessbackground

5.7%moresignal
AnArtialNeuralNetworkforPlugElectronIdenticationp.14/15
Conclusion,Outlook
Plugelectronswillgivebettersingletoplimit
IdenticationwithArticialNeuralNetworkuseful
20%lessbackgroundin2-jet-bin
5.7%moresignalin2-jet-bin
Plantoprovideeasy-to-usegenerictool
WaitingforGen5plugTopNtuples
Determineacceptanceandefciencies
Integrateintonextroundofsingletopanalysis
AnArtialNeuralNetworkforPlugElectronIdenticationp.15/15