New version page

UW-Madison ECE 539 - Comparison of Cut-based and Artificial Neural Network Selection of Signal B → Kγγ Events

Documents in this Course
Load more

This preview shows page 1-2-3-4-5 out of 14 pages.

View Full Document
View Full Document

End of preview. Want to read all 14 pages?

Upload your study docs or become a GradeBuddy member to access this document.

View Full Document
Unformatted text preview:

Comparison of Cut-based and ArtificialNeural Network Selection of SignalB → Kγγ EventsAda RubinJanuary 6, 2004• Electrons (e-) collided with positrons (e+) at the ’Υ(4S) reso-nance’– Approximately 98% of e+e- collisions result in ’continuum back-ground’e+e-→ q¯q; q = u, d, s, c quarks– Approximately 10−7% of e+e- collisions result in ’signal’ decayse+e- → Υ(4S) → B → Kγγ1• Shape variables– Shapes of continuum and signal events are dramatically differ-ent– Can eliminate events based on differences in distributions ofshape variables for the two different types of events2cos(θT)−1 −0.5 0 0.5 1050100150200250300350signal−1 −0.5 0 0.5 105001000150020002500continuumcos(theataT) Eliminate events such that |cos(θT)| > 0.8 for cut based analysis.3cos(θB)−1 −0.5 0 0.5 1050100150200250300350signal−1 −0.5 0 0.5 1050100150200250300350continuumcos(thetaB) Eliminate events such that |cos(θB)| > 0.75 for cut based analysis.4R20 0.2 0.4 0.6 0.8 101002003004005006007000 0.2 0.4 0.6 0.8 10100200300400500600700800R2 signal continuum Eliminate events such that R2> 0.6 for cut based analysis.5cos(θγLOW,γHIGH)−1 −0.5 0 0.5 10100200300400500600700800−1 −0.5 0 0.5 10200400600800100012001400cos(angle bet. high energy photon & low energy photon) signal continuum Eliminate events such that cos(θγLOW,γHIGH) > 0.6 for cut based anal-ysis.6cos(θK,γLOW)−1 −0.5 0 0.5 1050100150200250300350−1 −0.5 0 0.5 10200400600800100012001400cos(angle bet. K & low energy photon) signal continuum Eliminate events such that |cos(θK,γLOW)| > 0.8 for cut based analysis.7cos(θK,γHIGH)−1 −0.5 0 0.5 1020040060080010001200−1 −0.5 0 0.5 105001000150020002500cos(angle bet. K & low energy photon) signal continuum Eliminate events such that |cos(θK,γHIGH)| > 0.8 for cut based analysis.8Cut-based resultsTraining Set Testing SetCut Signal Background Signal BackgroundBefore cuts 2000 3000 400 400Total Yield 1146 153 226 21• Goal: Reduce background yield leaving the signal yield unchangedusing a single cut based on the ANN output.9ANN Configuration• Used a back propagating multi-layer perceptron type NN withα = 0.1µ = 0.8One hidden layer with 11 neuronsMaximum number of epochs = 10,000Epoch size = 100• Feature vectors normalized:– Mean of feature vector subtracted from all values of featurevector– Values of feature vector scaled from -1 to 110ANN-based results: TrainingRun Cut value Signal yield Background yield1 0.985 1146 552 0.943 1146 483 0.932 1147 614 0.976 1149 555 0.969 1146 616 0.967 1161 677 0.975 1147 698 0.948 1159 689 0.946 1148 6610 0.98 1166 58Average 0.962• The signal yield is the value closest to the signal yield that was achieved viacuts on the shape variables (1146).• The cut on the NN output was varied from 0.9 to 1 in intervals of 0.001.• Since signal events are classified as 1 and continuum background events areclassified as 0, the signal yield is composed of all events for whichneuralnet output > cut is true.• Conversely, the background yield is composed of all events for whichneuralnet output ≤ cut is true.11ANN-based results: TestingApplying the NN cut that was the average of the cuts (cut = 0.962) that gave usa signal yield of approximately 1146 on the training data:Run Signal yield Background yield1 270 172 186 113 244 154 180 65 224 116 262 147 264 208 294 219 136 710 200 11Average 226 1312Discussion of ResultsComparison of results on testing data:Signal Yield Background yieldCut-based 226 21NN-based (Ave. vals.) 226 13• Signal yield has remained unchanged• Background yield has been decreased by 62%• ANN based method of eliminating background events more effec-tive than cuts on shape


View Full Document
Loading Unlocking...
Login

Join to view Comparison of Cut-based and Artificial Neural Network Selection of Signal B → Kγγ Events and access 3M+ class-specific study document.

or
We will never post anything without your permission.
Don't have an account?
Sign Up

Join to view Comparison of Cut-based and Artificial Neural Network Selection of Signal B → Kγγ Events and access 3M+ class-specific study document.

or

By creating an account you agree to our Privacy Policy and Terms Of Use

Already a member?