Machine Learning in PandaRoot GlueX-Panda Workshop G.Washington - - PowerPoint PPT Presentation

SLIDE 1

Machine Learning in PandaRoot

GlueX-Panda Workshop G.Washington University, May 2019 Ralf Kliemt (GSI)

SLIDE 2

Motivation

• Machine Learning (ML) is about modelling data
• Self-learning algorithms gain knowledge from data to

make predictions Why?

• Gain computation speed in online scenarios
• More precision, e.g. by respecting correlations
• Let algorithms do the tedious tasks of recognising

patterns, structures, principal components etc.

2

SLIDE 3

3

Which type of ML?

SLIDE 4

4

Which type of ML?

SLIDE 5

PANDA FPGA Sim Digi Local Reco Global Reco Event Building Event Selection Analysis Storage Paper Raw Data

• nline
• ffline

sim

Alignm. Calib.

ML Activities at PANDA

5

SLIDE 6

PANDA FPGA Sim Digi Local Reco Global Reco Event Building Event Selection Analysis Storage Paper Raw Data

• nline
• ffline

sim

Alignm. Calib.

ML Activities at PANDA

6

SLIDE 7

Key Concept

7

SLIDE 8

Boosted Decision Tree (BDT)

8

• Break down data by steps of decisions
• Based on features in training data
• Splitting by maximising information gain

—> Typical application: Classification for PID

SLIDE 9

Artificial Neural Networks (ANN)

9

• Transform data meaningfully
• Learn iteratively

SLIDE 10

A lot to choose from…

10

SLIDE 11

11

A lot to choose from…

SLIDE 12

Programming Frameworks & Packages

12

• ROOT / TMVA
• NumPy
• TensorFlow (Deep Learning)
• Keras (on top of TensorFlow with GPUs)
• MLlib Apache Spark
• Sci-Kit Learn
• PyTorch (Deep Learning)
• DL4J (Deep Learning, Java)
• R Implementations


SLIDE 13

Popular Choices

13

SLIDE 14

Machine Learning for Forward Tracking

SLIDE 15

15

Arti tific icia ial l Neural l Netw tworks: s:

Application

• n to the FTS:
• Create all possible combinations of hit pairs (adjacent

layers).

• Train the network to predict if hit pairs are on the

same track or not.

• Input observables:

1) Hit pair positions in x-z projection (vertical layers). 2) Drift radii (Isochrones). 3) Distance between hits.

• Output:

1) Probability that hit pair are on the same track. Connect hits that pass the probability cut (threshold).

e.g. probability(h1-h2)> threshold, and probability(h2-h3)> threshold, so h1, h2, h3 are on the same track.

03.April.2019 Page 9

Ma Machine Learning For Track Finding at PANDA A FTS

Institut für Kernphysik (IKP)

Wa Waleed Esmail, Tobias Stockmanns, and James Ritman On Behalf of the PANDA Collaboration

Contribution at “Connecting the Dots 2019”

SLIDE 16

16

A) first layers B) middle layers inside Magnet C) last layers A B C

ANN Tracking: Pattern Recognition with parallel layers

SLIDE 17

17

ANN Tracking: Residuals including skewed layers

First promising results inside magnetic field

SLIDE 18

18

RNN (LSTM) Tracking: Residuals including skewed layers

Next Step: Track Fitting with RNN - stay tuned

SLIDE 19

Machine Learning for Particle Identification

SLIDE 20

PID: Usual Observables

20

p [GeV/c] 2 4 6 E/p 0.2 0.4 0.6 0.8 1

1 10 10

EMC: E/p vs. p

p [GeV/c] 2 4 6 [rad]

C

Θ 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

1 10 10

• vs. p

C

Θ DRC:

p [GeV/c] 2 4 6 [rad]

C

Θ 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

1 10 10

• vs. p

C

Θ DSC:

p [GeV/c] 2 4 6 dE/dx [a.u.] 2 4 6 8 10 12 14 16 18

1 10 10

STT: dE/dx vs. p

p [GeV/c] 2 4 6 [cm]

iron

L 10 20 30 40 50 60 70 80

10 1 10

) µ

• vs. p (

iron

MUO: L

p [GeV/c] 2 4 6 [cm]

iron

L 10 20 30 40 50 60 70 80

10 1 10

) µ

• vs. p (non-

iron

MUO: L

(charged particles only)

SLIDE 21

PID Approaches

21

P(~ x|h) = L(~ x|h) × P(h) P

h=e,µ,π,K,p

L(~ x|h) × P(h)

<latexit sha1_base64="3ZvoEMUZT2fNdKs/18Cr9fQEUpU=">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</latexit>

L(~ x|h) = Y

k

pk(~ x|h)

<latexit sha1_base64="woYPUKldMCmRspq6e3C9QClfum8=">AC3icbVC7SgNBFJ31GeNr1dJmSBiE3ZjoY0QTKGFRQTzgGxYZmdnk2FnH8zMBsOaXgQLf8TGQhFbP0A7f0acPIqYeODC4Zx7ufceJ2ZUSMP41hYWl5ZXVjNr2fWNza1tfWe3LqKEY1LDEYt40GCMBqSmqSkWbMCQocRhqOXxn6jR7hgkbhtezHpB2gTkg9ipFUkq3nLgtWj+D0ZnDbPTy1Yh65th/b/pRq63mjaIwA54k5Ifny+X3lx3r8rNr6l+VGOAlIKDFDQrRMI5btFHFJMSODrJUIEiPsow5pKRqigIh2OvplA+U4kIv4qpCUfq9ESKAiH6gaM6AyS7YtYbiv95rUR6J+2UhnEiSYjHi7yEQRnBYTDQpZxgyfqKIMypuhXiLuISxVfVoVgzr48T+qlonlULF2pNM7AGBmwD3KgAExwDMrgAlRBDWBwB57AC3jVHrRn7U17H7cuaJOZPfAH2scvmcmfKw=</latexit>

k = MVD dE/dx, DRC thetaC …

Probability that a given track with parameters x corresponds to particle type h Combination of measurements:

Bayes: Machine Learning:

• A. Boosted Decision Tree (BDT)
• B. “Deep Learning” Artificial Neural Network (ANN)

—> gain performance by considering correlations

SLIDE 22

Input Features

22

Input:

• 𝑞𝑞→𝑌𝑌𝑍𝑍,
• where X and Y =

e

∓,𝜌 ∓,𝜈 ∓,k ∓,p ∓

• Beam momentum: 15GeV/c

SLIDE 23

Performance Plots

23

Boosted Decision Tree Artificial Neural Network

SLIDE 24

Confusion Matrices

24

Boosted Decision Tree Artificial Neural Network

SLIDE 25

Pions

25

Boosted Decision Tree Artificial Neural Network

SLIDE 26

Kaons

26

Boosted Decision Tree Artificial Neural Network

SLIDE 27

Machine Learning for the Software Trigger

SLIDE 28

Expected Data Rates

28

• PANDA will run with a continuous beam
• Event rates will be high, some events will overlap
• Storage constraints in size & bandwidth
• Data rate has to be reduced by 1/1000
• No specific trigger on hardware possible
• Event topology of signals similar to background 


—> no “Jets” or similarly obvious features Solution: An online physics filter (“software trigger”)

SLIDE 29

Event Generation

• Signal
• Background

Simulation & Reconstruction Event Filtering

• Combinatorics
• Mass Window Selection
• Trigger Specific Selection 


→ Event Tagging

Global Trigger Tag

29

EvtGen

Physics Channel 1 Physics Channel 2 ... Physics Channel m

DPM

Background

Toy MC Full MC Trigger 1 Trigger n Trigger Decision (Logical OR) Trigger 2 ... Trigger 3

reduce bg. by 1/1000

Software Trigger

SLIDE 30

Software Trigger - Cuts

30

• Cut based approach, optimised on signal & bg. MC
• Many observables taken into account
• Correlations are not respected

Entries 15610

]

2

) [GeV/c

• e

+

m(e 1 2 3 4 5 6 7

200 400 600 800 1000 1200 1400 1600 1800 2000 2200

Entries 15610

= 0.0% ∈

• e

+

e → p p

Entries 25069

]

2

) [GeV/c

• K

+

m(K 0.5 1 1.5 2 2.5 3

100 200 300 400 500

Entries 25069

= 0.8% ∈

• K

+

K → φ

Entries 210195

]

2

) [GeV/c

• π

+

K

S

m(K 1 2 3 4 5 6

500 1000 1500 2000 2500 3000 3500 4000

Entries 210195

= 3.4% ∈

• π

+

K

S

K →

c

η

Entries 15610

]

2

) [GeV/c

• e

+

m(e 1 2 3 4 5 6

200 400 600 800 1000 1200 1400 1600 1800

Entries 15610

= 0.0% ∈

• e

+

e → ψ J/

Entries 6232

]

2

) [GeV/c

• µ

+

µ m( 1 2 3 4 5 6

20 40 60 80 100 120 140 160 180

Entries 6232

= 0.0% ∈ = 21.9%)

t

∈ (

• µ

+

µ → ψ J/

Entries 295962

]

2

) [GeV/c

+

π

• m(K

1 2 3 4 5 6

2000 4000 6000 8000 10000

Entries 295962

= 7.2% ∈

+

π

• K

→ D

Entries 224819

]

2

) [GeV/c

+

π

+

π

• m(K

1 2 3 4 5 6

500 1000 1500 2000 2500 3000 3500 4000 4500

Entries 224819

= 8.6% ∈

+

π

+

π

• K

→

+

D

Entries 74779

]

2

) [GeV/c

+

π

• K

+

m(K 1 2 3 4 5 6

200 400 600 800 1000 1200 1400 1600

Entries 74779

= 2.8% ∈

+

π

• K

+

K →

+ s

D

Entries 667078

]

2

) [GeV/c

• π

m(p 0.5 1 1.5 2 2.5 3

1000 2000 3000 4000 5000 6000 7000 8000

Entries 667078

= 7.6% ∈

• π

p → Λ

Entries 90669

]

2

) [GeV/c

+

π

• m(pK

1 2 3 4 5 6

200 400 600 800 1000 1200 1400

Entries 90669

= 5.2% ∈

+

π

• pK

→

c

Λ

DPM 2014

SLIDE 31

Software Trigger - Cuts

31

[GeV] s

2 2.5 3 3.5 4 4.5 5 5.5 6 0.1 0.2 0.3 0.4 0.5 0.6 0.7

e+e- φ

c

η (2e) ψ J/ ) µ (2 ψ J/ D

±

D

s

D Λ

c

Λ

Full MC - Efficiencies - all cuts (high suppression)

[GeV] s

2 2.5 3 3.5 4 4.5 5 5.5 6

• 3

10

• 2

10

• 1

10 1

mass cut only all cuts (high supression)

Full MC - Background fraction

• Trade-off between signal efficiency & background

suppression

• Many channels = much feedthrough & cross-tagging

Goal: BG suppression 1/1000 2014

SLIDE 32

Software Trigger - TMVA

32

• First studies with many algorithms
• Dependence on offered observables


—> output performance
 —> calculation speed

[GeV] s

2 2.5 3 3.5 4 4.5 5 5.5 6 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

φ D

±

D

s

D (2e) ψ J/ ) µ (2 ψ J/

c

η

• e

+

e Λ

c

Λ

Efficiency - TMVA application

[GeV] s

2 2.5 3 3.5 4 4.5 5 5.5 6

• 3

10

• 2

10

• 1

10 1

Mass cut only TMVA application

Background fraction

CFMlpANN

Goal: BG suppression 1/1000

Signal efficiency

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Background rejection

0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

MVA Method: BDT SVM RuleFit MLP KNN TMlpANN Likelihood FDA_GA Fisher

Background rejection versus Signal efficiency

2014

SLIDE 33

Software Trigger on GPU?

33

FastSim - Principle Component Analysis - no GPU, yet

pp → D+D− → K−π+π+ D−(incl.) (& cc.)

<latexit sha1_base64="j/nYHFxr7hPc0Pmeo0J07vdFGDI=">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</latexit>

Reaction:

Signal Background Total events 24713 52180 True selection 23555 50350 False selection 1158 1830 True selection rate 0.953 0.965

MC input

SLIDE 34

Software Trigger on GPU

34

FullSim - Artificial Neural Network - training on GTX1080Ti

pp → D+D− → K−π+π+ D−(incl.) (& cc.)

<latexit sha1_base64="j/nYHFxr7hPc0Pmeo0J07vdFGDI=">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</latexit>

Reaction:

MC input All mass spectra normalised to 1

SLIDE 35

Summary

• Machine learning sees a comeback in physics
• Many available libraries with fresh concepts
• ML in algorithms under development:
• Forward tracking
• charged PID
• Software trigger
• Potential parts which may benefit from ML:
• EMC cluster shape analysis
• Event building
• Physics analysis

35

SLIDE 36

Thanks for your attention.

SLIDE 37

SLIDE 38

PandaRoot Communication

• Code Repository: pandaatfair.githost.io
• Issue tracker, including discussions
• WiKi page:

panda-wiki.gsi.de/foswiki/bin/view/Computing/PandaRoot

• Forums: forum.gsi.de & Slack: pandaroot.slack.com
• Bi-weekly online meetings: Thu. 10-11
• Dashboard:

https://cdash.gsi.de/index.php?project=PandaRoot

38

SLIDE 39

39

TString inputGenerator = "psi2s_Jpsi2pi_Jpsi_mumu.dec"; // "dpm" "ftf" or e.g. "box:type(211,1):p(1,1):tht(10,120):phi(0,360)" PndMasterRunSim *fRun = new PndMasterRunSim(); fRun->SetInput(inputGenerator); fRun->SetName("TGeant4"); fRun->SetOptions(""); fRun->SetParamAsciiFile("all.par"); fRun->SetBeamMom(7.0); fRun->SetStoreTraj(kTRUE); fRun->Setup("evtcomplete"); // file name prefix fRun->CreateGeometry(); fRun->SetGenerator(); fRun->AddSimTasks(); fRun->Init(); fRun->Run(1000); //nEvents fRun->Finish();

e.g. in macro/master Short ROOT macros to start simulations