use of multivariate analysis (mva) technique in data analysis rakshya khatiwada 08/08/2007
TRANSCRIPT
Hunting for Higgs
Use of Multivariate Analysis (MVA) Technique in Data Analysis
Rakshya Khatiwada08/08/2007
Index1. Background2. My work3. MVA4. MLP (Type of MVA techniques)5. Details of work6. Result7. Conclusion and Future8. Acknowledgement
Why study Higgs Boson?Higgs field is supposed to be responsible for
the mass of the elementary particles
Thus, the term Higgs boson
Standard model incomplete without Higgs (not considering Gravity!). Thus need of research on Higgs.
Current Status from CDF and D0Why is it taking so long to detect Higgs?
Not enough Luminosity to detect it. Current limit of 3fb-1 in DØ and CDF but required Luminosity is ????
My work Current : Comparison of Conventional Data Analysis
technique with Multivariate Analysis (MLP Neural Network) using DØ MC P17.
Focused on ZH channel with and backgrounds
Here, I will be discussing only background.
tt bWb
tt
Multivariate Analysis (MVA)Statistical technique used to analyze data
that involves from more than one variable.
MVA package used - Multi Layered Perceptron (MLP)
Feed forward Neural Network (NN) (flow of information in one direction)
Consists of an input layer, two hidden nodes layer and an output layer with one node (gives either signal or background)
MLP NN (Analogous to Brain)NN with two hidden layers
Input layers
Hidden layers
Output layer
Neuron
How does NN work?Works similar to human brain where there
are input and output ports and in between, the processing takes place. Weight is applied to each parameter and processing takes place accordingly. (higher weight, higher priority)
Humans learn by example, in a similar manner, ANN is configured for a specific application such as pattern recognition or data classification through learning process. Thus, it needs to be trained.
Additional information
bbbbZHpp
bbWHpp
,
Signal
Background
tbtbqWZZZbZbbWbttpp ,,,,,,
Channels
xbb
xbb
xbbET
Single lepton
Di-lepton
Missing TE
Variables usedEt
b1 - Transverse Energy of the 1st b jet
Etb2 - Transverse Energy of the 2nd b jet
Ptμ1 - Transverse Momentum of the 1st muon
Ptμ2 - Transverse Momentum of the 2nd muon
Et - Missing Transverse Energy (neutrinos)
Mbb - Mass of bb jets
Mµ+
µ- - Mass of µ+µ-
Ht - Total Transverse Energy of jets
Variable Distributi0ns
Use HT for conventional cut
Applying cuts
Calculating Signal over Root Background (SoRB)
As a function of the cut value
As a function of signal events surviving cut
Output of MLP
SoRB Comparison:As a function of the cut
MVAConventional method
SoRB Comparison:As a function of the number of signal events surviving cut
MVAConventional method
SummaryMVA gives better discrimination of Signal
and Background than conventional analysis.
Signal efficiency (S/√B) significantly higher for MVA.
Less work for us since no need to apply multiple cuts to have good discrimination.
Future planDetailed study of MLP and Bayesian NN
(definition)
Use of real data(not just MC)
Could be useful at LHC if not here for further research in Higgs.
AcknowledgementDr. Pushpa Bhat
Scientist, CMS/DØParticle Physics Division
Michael PogwizdStudent, University of Illonois Urbana
Champaign.
DØ Group
Internship for Physics MajorsFermi National Accelerator Laboratory
Jean, Roger, Erik, Carol and Fermilab family