SLIDE 1
1
Fiducialization in DEAP-3600 using Machine Learning algorithms with robust validation
Connor Stone Queen’s University May 30 2017
2
Fiducialization in DEAP-3600 using Machine Learning algorithms with - - PowerPoint PPT Presentation
Fiducialization in DEAP-3600 using Machine Learning algorithms with - - PowerPoint PPT Presentation
Fiducialization in DEAP-3600 using Machine Learning algorithms with robust validation Connor Stone Queens University May 30 2017 1 Big Thanks To Everyone At DEAP-3600! 2 Thats Me! 3 Introduction to DEAP-3600 and Machine Learning
SLIDE 2
SLIDE 3
3
That’s Me!
SLIDE 4
4
Introduction to DEAP-3600 and Machine Learning Using Machine Learning to improve the Dark Matter discovery potential Validating the event classification Validating the event classification
SLIDE 5
5
Argon Volume Acrylic Vessel, Light Guides Wavelength Shifter (TPB) Photo-Multiplier Tube (PMT) Fiducial Volume
SLIDE 6
6
Position reconstruction can be achieved using relative PMT charges
SLIDE 7
7
Position reconstruction can be achieved using relative PMT charges
More light Less light
SLIDE 8
8
The MBLikelihood algorithm does a good job of creating a fiducial volume
1 tonne fiducial volume Surface event leakage Uniform in R3
SLIDE 9
9
Machine Learning is well poised to assist in creating a fiducial volume
Machine Learning
Class 2 Class 1 Measurement 1 Measurement 2 Decision Boundary Measurement 2 Measurement 1
SLIDE 10
10 Logistic Regression (LR) Linear Support Vector Machine (LSVM)
SLIDE 11
11 K-Nearest Neighbors (KNN) Naive Bayes (NB)
SLIDE 12
12 Support Vector Machine (SVM) Neural Network (NN)
SLIDE 13
13
Using Machine Learning to improve the Dark Matter discovery potential
SLIDE 14
14
Training data can be taken from the real detector for Machine Learning, but not for MBLikelihood
Both agree these are surface events Both agree these are surface events
SLIDE 15
15
Some Machine Learning algorithms perform well, but don’t beat MBLikelihood
MBLikelihood radius is the best
SLIDE 16
16
MBLikelihood and Machine Learning algorithms can be combined (Boosted)
Bad Bad Bad Good
SLIDE 17
17
The boosted analysis has reduced surface event leakage
Near order of magnitude improvement! LSVM+ at 2.5*10-4, MBR at 1.8*10-3
+
SLIDE 18
18
Validating the event classification Validating the event classification
SLIDE 19
19
The boosted analysis behaves well
- n real neutron data
The boosted analysis behaves well
- n real neutron data
SLIDE 20
20
The boosted analysis has an expected energy dependence The boosted analysis has an expected energy dependence
SLIDE 21
21
The boosted analysis is robust against changes in the optical model The boosted analysis is robust against changes in the optical model
SLIDE 22
22
Conclusion: The boosted analysis can increase the fiducial volume by 50%!
SLIDE 23
23
Extra
SLIDE 24
24
SLIDE 25
25
Runtime is very different for each algorithm
SLIDE 26
26
Preprocessing the data allows for easier surface event classification
PMT2 PMT1 PMT2 PMT1
SLIDE 27
27
Machine Learning is robust against changes in the optical model
SLIDE 28
28
Some optical properties don’t affect MBLikelihood or the boosted analysis
SLIDE 29
29
Support Vector Machine Logistic Regression
SLIDE 30
30
Neural Network Naive Bayes
SLIDE 31
31
SLIDE 32
32