E - / π separation with TRD Gennady Ososkov LIT JINR, Dubna Semen Lebedev GSI, Darmstadt and LIT JINR, Dubna Claudia Höhne, Florian Uhlig GSI, Darmstadt.

Slides:



Advertisements
Similar presentations
CBM Calorimeter System CBM collaboration meeting, October 2008 I.Korolko(ITEP, Moscow)
Advertisements

PHYSTAT2007, CERN Geneva, June 2007 ALICE Statistical Wish-List I. Belikov, on behalf of the ALICE collaboration.
E/π identification and position resolution of high granularity single sided TRD prototype M. Târzilă, V. Aprodu, D. Bartoş, A. Bercuci, V. Cătănescu, F.
Kalanand Mishra BaBar Coll. Meeting Sept 26, /12 Development of New SPR-based Kaon, Pion, Proton, and Electron Selectors Kalanand Mishra University.
14 Sept 2004 D.Dedovich Tau041 Measurement of Tau hadronic branching ratios in DELPHI experiment at LEP Dima Dedovich (Dubna) DELPHI Collaboration E.Phys.J.
Multiple Criteria for Evaluating Land Cover Classification Algorithms Summary of a paper by R.S. DeFries and Jonathan Cheung-Wai Chan April, 2000 Remote.
Searching for Single Top Using Decision Trees G. Watts (UW) For the DØ Collaboration 5/13/2005 – APSNW Particles I.
Summary of downstream PID MICE collaboration meeting Fermilab Rikard Sandström.
Off-axis Simulations Peter Litchfield, Minnesota  What has been simulated?  Will the experiment work?  Can we choose a technology based on simulations?
30 March Global Mice Particle Identification Steve Kahn 30 March 2004 Mice Collaboration Meeting.
Particle Identification in the NA48 Experiment Using Neural Networks L. Litov University of Sofia.
Ensemble Learning: An Introduction
Algorithms and Methods for Particle Identification with ALICE TOF Detector at Very High Particle Multiplicity TOF simulation group B.Zagreev ACAT2002,
Energy Reconstruction Algorithms for the ANTARES Neutrino Telescope J.D. Zornoza 1, A. Romeyer 2, R. Bruijn 3 on Behalf of the ANTARES Collaboration 1.
Dec 2005Jean-Sébastien GraulichSlide 1 Improving MuCal Design o Why we need an improved design o Improvement Principle o Quick Simulation, Analysis & Results.
Machine Learning: Ensemble Methods
Experimental Evaluation
Decision Tree Models in Data Mining
Radial Basis Function (RBF) Networks
Tau Jet Identification in Charged Higgs Search Monoranjan Guchait TIFR, Mumbai India-CMS collaboration meeting th March,2009 University of Delhi.
1 A Bayesian Method for Guessing the Extreme Values in a Data Set Mingxi Wu, Chris Jermaine University of Florida September 2007.
RICH Development Serguei Sadovsky IHEP, Protvino CBM meeting GSI, 9 March 2005.
Michigan REU Final Presentations, August 10, 2006Matt Jachowski 1 Multivariate Analysis, TMVA, and Artificial Neural Networks Matt Jachowski
G. Cowan Statistical Methods in Particle Physics1 Statistical Methods in Particle Physics Day 3: Multivariate Methods (II) 清华大学高能物理研究中心 2010 年 4 月 12—16.
Comparison of Bayesian Neural Networks with TMVA classifiers Richa Sharma, Vipin Bhatnagar Panjab University, Chandigarh India-CMS March, 2009 Meeting,
STS track recognition by 3D track-following method Gennady Ososkov, A.Airiyan, A.Lebedev, S.Lebedev, E.Litvinenko Laboratory of Information Technologies.
1 Realistic top Quark Reconstruction for Vertex Detector Optimisation Talini Pinto Jayawardena (RAL) Kristian Harder (RAL) LCFI Collaboration Meeting 23/09/08.
Charmonium feasibility study F. Guber, E. Karpechev, A.Kurepin, A. Maevskaia Institute for Nuclear Research RAS, Moscow CBM collaboration meeting 11 February.
B-tagging Performance based on Boosted Decision Trees Hai-Jun Yang University of Michigan (with X. Li and B. Zhou) ATLAS B-tagging Meeting February 9,
Combining multiple learners Usman Roshan. Bagging Randomly sample training data Determine classifier C i on sampled data Goto step 1 and repeat m times.
TRD and Global tracking Andrey Lebedev GSI, Darmstadt and LIT JINR, Dubna Gennady Ososkov LIT JINR, Dubna X CBM collaboration meeting Dresden, 27 September.
1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 24 Nov 2, 2005 Nanjing University of Science & Technology.
MiniBooNE Event Reconstruction and Particle Identification Hai-Jun Yang University of Michigan, Ann Arbor (for the MiniBooNE Collaboration) DNP06, Nashville,
A.Ayriyan 1, V.Ivanov 1, S.Lebedev 1,2, G.Ososkov 1 in collaboration with N.Chernov 3 1 st CBM Collaboration Meeting, JINR Duba,19-22 May JINR-LIT,
Training of Boosted DecisionTrees Helge Voss (MPI–K, Heidelberg) MVA Workshop, CERN, July 10, 2009.
Decision Trees Binary output – easily extendible to multiple output classes. Takes a set of attributes for a given situation or object and outputs a yes/no.
Electron-hadron separation by Neural-network Yonsei university.
I.BelikovCHEP 2004, Interlaken, 30 Sep Bayesian Approach for Combined Particle Identification in ALICE Experiment at LHC. I.Belikov, P.Hristov, M.Ivanov,
Track reconstruction in TRD and MUCH Andrey Lebedev Andrey Lebedev GSI, Darmstadt and LIT JINR, Dubna Gennady Ososkov Gennady Ososkov LIT JINR, Dubna.
Global Tracking for CBM Andrey Lebedev 1,2 Ivan Kisel 1 Gennady Ososkov 2 1 GSI Helmholtzzentrum für Schwerionenforschung GmbH, Darmstadt, Germany 2 Laboratory.
Particle Identification. Particle identification: an important task for nuclear and particle physics Usually it requires the combination of informations.
Feasibility study of Higgs pair production in a Photon Collider Tohru Takahashi Hiroshima University for S.Kawada, N.Maeda, K.Ikematsu, K.Fujii,Y.Kurihara,,,
Kalanand Mishra BaBar Coll. Meeting February, /8 Development of New Kaon Selectors Kalanand Mishra University of Cincinnati.
Analysis of H  WW  l l Based on Boosted Decision Trees Hai-Jun Yang University of Michigan (with T.S. Dai, X.F. Li, B. Zhou) ATLAS Higgs Meeting September.
Classification and Regression Trees
Combining multiple learners Usman Roshan. Decision tree From Alpaydin, 2010.
Calice Meeting Argonne Muon identification with the hadron calorimeter Nicola D’Ascenzo.
Christian Lippmann (ALICE TRD), DPG-Tagung Köln Position Resolution, Electron Identification and Transition Radiation Spectra with Prototypes.
meeting, Oct. 1 st 2015 meeting, Oct. 1 st Gas Pixel: TRD + Tracker.
SAS TRD Possible TRD configurations for PID up to the TeVs energies fig.s for this talk taken by: B.Dolgoshein Transition radiation detectors -NIM A326(1993)
J/ψ simulations: background studies and first results using a compact RICH detector Alla Maevskaya INR RAS Moscow CBM Collaboration meeting September 2007.
Villa Olmo, Como October 2001F.Giordano1 SiTRD R & D The Silicon-TRD: Beam Test Results M.Brigida a, C.Favuzzi a, P.Fusco a, F.Gargano a, N.Giglietto.
Chapter 11 – Neural Nets © Galit Shmueli and Peter Bruce 2010 Data Mining for Business Intelligence Shmueli, Patel & Bruce.
By Subhasis Dasgupta Asst Professor Praxis Business School, Kolkata Classification Modeling Decision Tree (Part 2)
Converted photon and π 0 discrimination based on H    analysis.
Study of the electron identification algorithms in TRD Andrey Lebedev 1,3, Semen Lebedev 2,3, Gennady Ososkov 3 1 Frankfurt University, 2 Giessen University.
IPHC, Strasbourg / GSI, Darmstadt
Ensemble Classifiers.
Machine Learning: Ensemble Methods
Straw prototype test beam 2017: first glance at the data
Marco Incagli – INFN Pisa
Top Tagging at CLIC 1.4TeV Using Jet Substructure
CMS Preshower: Startup procedures: Reconstruction & calibration
Implementing Boosting and Convolutional Neural Networks For Particle Identification (PID) Khalid Teli .
MiniBooNE Event Reconstruction and Particle Identification
Detector Configuration for Simulation (i)
Data Analysis in Particle Physics
Study of e+e- pp process using initial state radiation with BaBar
Status of CEPC HCAL Optimization Study in Simulation LIU Bing On behalf the CEPC Calorimeter working group.
Gas Pixel TRD/Tracker With the support of the TRT collaboration
Presentation transcript:

e - / π separation with TRD Gennady Ososkov LIT JINR, Dubna Semen Lebedev GSI, Darmstadt and LIT JINR, Dubna Claudia Höhne, Florian Uhlig GSI, Darmstadt Pavel Nevski BNL-CERN 14th CBM Collaboration meeting October , Split, Croatia

G.Ososkov e/pi separation CBM Collaboration Meeting, TR Production (recall)  dE/dx is described reasonably well  TR is calculated using a model: M. Castellano et al. Computer Physics Communication 61 (1990)  Parameters of the model are chosen to describe the measured data Nr. of foils Foil thickness Gas thickness  Parameters are adjusted only for 1.5 GeV/c  Trunk version of CBM ROOT from SEPT09 was used. Florian Uhlig, CBM Collaboration Meeting, March 2009 π contribution = dE/dx e - contribution= dE/dx+TR See also presentation from S.Lebedev et al on energy loss simulation discussions

G.Ososkov e/pi separation CBM Collaboration Meeting, NIM A326 (1993) From: Dolgoshein TRD overview NIM A326 (1993) …two methods of signal processing have been mainly used : 1) total energy deposition ("Q-method") The main factor limiting the rejection of nonradiating particles in this case is the Landau "tail" of ionization loss which simulates a large energy deposition comparable with that of a radiating particle Inspired by discussion with Pavel Nevski (BNL,CERN) There are important distinctions between the above two methods (Q and N) - for N-method compared to the Q-method thinner foils and detector gas layers are needed; - The readout for the two methods is also different. ADCs (or FADCs) are needed for the Q-method. N-method requires fast discriminators and sealers. With the second method the avalanches produced by X-ray photoelectrons are recorded and counted when they exceed a given threshold (typically 4-5 keV). Nonradiating particles should provide fewer detectable clusters. cluster counting ("N-method”), proposed yet in ) cluster counting ("N-method”), proposed yet in 1980

G.Ososkov e/pi separation CBM Collaboration Meeting, Default method e - /π identification in CBM 30 years passed and new more powerful classifiers appeared Method which is used now on CBM as a defaut (V.Ivanov, S.Lebedev, T.Akishina, O.Denisova) is based on applying a neural network. Energy losses from all 12 TRD layers are normalized i=1,..,12 ordered, then the values of Landau distribution function are calculated to be input to NN Pion suppression result depends on parameters of NN class. For fixed choice of initial weights it is 550

G.Ososkov e/pi separation CBM Collaboration Meeting, Threshold methods on the CBM TRD: (1) photon cluster counting π e-e- 1) Easy algorithm for N counting: - cleare N; - compare with cut=5 KeV, if > 5, then increase N by 1; - repeate for each of 12 TRD layers After 12 ifs N is the photon cluster size to be histogrammed separately for pions and electrons. 2) Then PID algorithm is common: - if N > threshold, then e - is indentified, else – pion is identified \ Pion suppression with cut=5 KeV is 448 After the cut optimization it is 584, although the e - efficiency drops down to 88.3% Boris Dolgoshein (1993): «In general, the cluster counting (N) method, should be the best method due to the distinction of the Poisson distributed number of the ionization clusters produced by nonradiated particles against the Landau tail of dE/dx losses in case of Q-method. But the comparison of the two methods is complicated and should be done for each individual case, because of the different optimum structures required for both methods and the problem of the cluster counting ability of TR chambers.» The main lesson: a transformation needed to reduce Landau tails of dE/dx losses

G.Ososkov e/pi separation CBM Collaboration Meeting, Threshold methods on the CBM TRD: (2) ordered statistics Now, the easy threshold algorithm with corresponding cut on λ 1 gives the pion suppresion 10. Median, which is 6th order statistic λ 6 with the distribution function proportional to [F Landau (x)(1-F Landau (x))] 6, gives pion suppression = 374 π e-e- Such trasformation can be provided by ordering the ΔE i sample i=1,..,12. For instance, the first order statistic λ 1 =ΔE min has the distribution function F(x)=P{λ 1 <x}= P{ΔE 1 <x, ΔE 2 <x,…, ΔE 12 <x}=[F Landau (x)] 12 That means a substantional distribution compression along the horizontal axis. i.e. tails diminishing. The main conclusion: The main conclusion: some of ordered statistics can also be used for pion suppression. However, why don’t use the information of all of them as a neural net input? π e-e- original Landau distribution λ 1 distribution Median distribution

G.Ososkov e/pi separation CBM Collaboration Meeting, The idea of ordering signals to be input to NN Distributions of all 12 dE/dx after their ordering NN with input of 12 ordered ΔE gives pion suppression = 685

G.Ososkov e/pi separation CBM Collaboration Meeting, Idea: input to NN probabilities of ΔE Plot of cumulative distributions of ΔE calculated from previous histogramms Note: all ΔE-s must be scaled to interval [0,1] to be input to NN. The excellent guess – input to NN not ΔE-s, but their probabilities calculated individually for each ΔE i One can calculate these probabilities either by pion distribution or by electron one Pion suppression for the first case = When e - distribution is used, it is = 786

G.Ososkov e/pi separation CBM Collaboration Meeting, A thought: -let us try some other classifiers from TMVA (Toolkit for MultiVariate data Analysis with ROOT)

G.Ososkov e/pi separation CBM Collaboration Meeting, Decision trees in Particle Identification (From Ososkov’s lecture on CBM Tracking workshop, 15 June 2009) data sample Single Decision Tree Root Node Branch Leaf Node ● Go through all PID variables, sort them, find the best variable to separate signal from background and cut on it. ● For each of the two subsets repeat the process. ● This forking decision pattern is called a tree. ● Split points are called nodes. ● Ending nodes are called leaves. 1)Multiple cuts on X and Y in a big tree (only grows steps 1-4 shown) However, a danger exists - degrading of classifier performance by demanding perfect training separation, which is called “overtraining” all cuts for the decision tree

G.Ososkov e/pi separation CBM Collaboration Meeting, How to boost Decision Trees weights of misclassified events ● Given a training sample, boosting increases the weights of misclassified events (background wich is classified as signal, or vice versa), such that they have a higher chance of being correctly classified in subsequent trees. ● Trees with more misclassified events are also weighted, having a lower weight than trees with fewer misclassified events. ● Build many trees (~1000) and do a weighted sum of event scores from all trees 1-1 (score is 1 if signal leaf, -1 if background leaf). The renormalized sum of all the scores, possibly weighted, is the final score of the event. High scores mean the event is most likely signal and low scores that it is most likely background. Boosted Decision Trees (BDT) Boosting Algorithm has all the advantages of single decision trees, and less susceptibility to overtraining. Many weak trees (single-cut trees) combined (only 4 trees shown) boosting algorithm produces 500 weak trees together

G.Ososkov e/pi separation CBM Collaboration Meeting, e-/ π separation with boosted decision tree BDT output Result for the BDT classifier: pion supression is 2180 for 90% electron efficiency Cut = 0,77

G.Ososkov e/pi separation CBM Collaboration Meeting, Summary and outlook Comparative study of e/pi separation methods was accomplished for - 1D cut methods - photon cluster counting - ordered statistics of dE/dx - default Neural Net classifier - Neural Net classifiers with input of - ordered statistics - probabilities of ordered statistics - Boosted Decision Tree classifier The BDT shows the best performance Outlook: - - Correct simulation parameters in order to obtain better correspondence to experimental results - Facilitate the input for NN and BDT by approximations of cumulative distributions - Stability and robustness study for NN and BDT classifiers - Test other classifiers from TMVA ( Toolkit for MultiVariate data Analysis with ROOT)

G.Ososkov e/pi separation CBM Collaboration Meeting, P.Nevski’s comment related to the practical aspects of pi/e rejection: experimental factors For experiments like CBM one should consider not only a rejection procedure, as it is, but it is necessary to take into account its robustness to such experimental factors as calibration of measurements, pile up of signals etc. Since these factors are different for each station, measurements are taken in different conditions and, inevitably, are heterogeneous. That leads to serious violations of all neural network methods That leads to serious violations of all neural network methods. only parameter - threshold Cluster counting methods, as it was shown in practice, is quite stable to its only parameter - threshold and, therefore, it is very robust. However that is the subject for more detailed study. The final question of a mathematician who is not experienced in TRD design and elecronics: - if pion supression, as 500, is enough; - if photon cluster counting is cheaper to carry it out than existing approach, then why do not consider the “N-method” as a real alternative to “Q-methods” despite of all improvements shown above?

G.Ososkov e/pi separation CBM Collaboration Meeting, Thanks for your attention!