Lab-wide Run 2 Advanced Analysis Group Pushpa Bhat

Lab-wide Run 2 Advanced Analysis Group Pushpa Bhat Fermilab “Keep it simple. As simple as possible. Not any simpler. ” -Einstein DØ Collaboration Meeting Apr. 25, 2001 April 25, 2001 D 0 Collaboration Meeting P. Bhat

Why Lab-wide (CDF/DØ)? • • We may find it necessary to co-operate with the other experiment on important physics analyses Likely scenario for Higgs discovery: – 5 signal at 115 Ge. V/c 2 requires 15 fb-1 at each experiment and combining the results! Run II Higgs study hep-ph/0010338 (Oct-2000) P. C. Bhat, R. Gilmartin, H. Prosper, Phys. Rev. D. 62 (2000) 074022 T. Han, A. S. Turcot , and R. -J. Zhang, Phys. Rev. D 59(1999) April 25, 2001 D 0 Collaboration Meeting P. Bhat 2

Run 2 AAG - Brief History • Desirability of such a group expressed at ACAT 2 K and Top Thinkshop-2 by CDF, DØ physicists • Had discussions with CDF/DØ spokespersons, Directorate • Presentations at a couple of CDF meetings • One-day mini-workshop on Feb. 10, 2001 – ~70 people from CDF/DØ – Opening by Mike Shaevitz – Many talks + tool-kit demo • Identified sub-groups • A few meetings since the workshop • See http: //www-d 0. fnal. gov/~pushpa/run 2 aag/ April 25, 2001 D 0 Collaboration Meeting P. Bhat 3

Run 2 AAG • Focus: Multivariate & Statistical Analysis Issues for Run II • Goals: – Learn and discuss – Explore, develop & adopt tools – Evaluate Methods – Document tools, studies, findings Generate a report • Agenda: – Driven by the participants April 25, 2001 D 0 Collaboration Meeting P. Bhat 4

Run II Advanced Analysis Working Groups • Tools – Provide convenient interfaces to useful algorithms and software, in an experiment-independent manner • Simulation – provide simulated data in a convenient form for evaluation studies • Evaluation – Evaluate methods with a few example studies • Statistics – Address statistical analysis issues April 25, 2001 D 0 Collaboration Meeting P. Bhat 5

Tools Group • Develop methods/tools, discuss ideas, share understanding. – Develop tools/interfaces in such a way that it is not tied to a particular experiment – Usable in ROOT, Python or in C++ – Example Algorithms: Neural Networks (Jetnet, MLPfit, …) Ra. GS (Random Grid Search) PDE (Probability Density Estimation) PCA (Principal Component Analysis) ICA (Independent Component Analysis) Bayesian Fit Markov Chain Monte Carlo (for multivariate Bayesian analysis) April 25, 2001 D 0 Collaboration Meeting P. Bhat 6

Tools • Prior to Run 2 AAG – Tools developed for Run 1 analyses Ra. GS, H-matrix, NN(Jetnet) Interface, Bayes. Fit (Bhat, Prosper, Stewart, …) PDE (Rice Univ. Group) RGSHelper, Sleuth (Knuteson) – A Python tool-kit (based on C++ wrappers) for a number of multivariate and event generator/simulator packages (H. Prosper) – dØMA – a DØ multivariate package of useful tools • Now – Python tool-kit development continues • Will be available from Run 2 AAG web page – dØMA now evolved into Terra. Fer. MA (S. Towers, D. Chakraborty, S. Desai, J. Hobbs) • See http: //www-d 0. fnal. gov/~smjt/ April 25, 2001 D 0 Collaboration Meeting P. Bhat 7

Multivariate Analysis Issues • Dimensionality Reduction – Choosing variables optimally without losing information • Choosing the right method for the problem • Controlling Model Complexity • Testing Convergence • Validation – Given a limited sample what is the best way? • Computational Efficiency April 25, 2001 D 0 Collaboration Meeting P. Bhat 8

…Multivariate Analysis Issues • Correctness of modeling – How do we show that the background is modeled well? Is it sufficient to check the modeling in the mapped variable? Pair-wise correlations? Higher order correlations? – How do we quantify the correctness of modeling? – In conventional analysis, we normally look for variables that are well modeled in order to apply cuts – How well is the background modeled in the signal region? • Robustness of the results – How do we check the robustness? – For each specific case, should we distort our modeling and study the effect on the results? April 25, 2001 D 0 Collaboration Meeting P. Bhat 9

Simulation • We anticipate providing ntuples that contain Monte Carlo simulations of Higgs, SUSY, Single Top events, etc. , together with the most important backgrounds. • Use PGS (Pretty Good Simulator, formerly SHW), developed for SUSY/Higgs Workshop. Provides a fast simulation (thought to be accurate to about 15%) of the response of typical collider experiments to high energy collision events. April 25, 2001 D 0 Collaboration Meeting P. Bhat 10

Evaluation • Use example physics analyses for evaluating methods/tools • Improve several aspects of analysis • Develop methods for sophisticated multivariate cross-checks April 25, 2001 D 0 Collaboration Meeting P. Bhat 11

Statistics • Statistical procedures – Review what we do & the literature – Discuss and study – Develop recommendations • Make information available on the Run 2 AAG web site – Web pages of Ba. BAR and CDF statistical committees will be accessible • Louis is in touch with the Ba. BAR Group • Documentation from CERN workshops – We also receive minutes from CDF statistics committee will meet with DØ counterparts once in three months or so April 25, 2001 D 0 Collaboration Meeting P. Bhat 12

… Statistics • Sample issues: – Systematic uncertainties • What are they? • Can they always be disentangled? • Those connected with theory options – Modeling of errors • Gaussian error on efficiency? – Confidence Levels, Limits – Bayesian approach • Priors – Model comparisons – Combining results April 25, 2001 D 0 Collaboration Meeting P. Bhat 13

Please come and participate! Next meeting: May 24, 2001 http: //www-d 0. fnal. gov/~pushpa/run 2 aag/ (Temporary location) Run 2 AAG web pages need a Web Master from DØ. Tony Vaiculis (CDF) April 25, 2001 D 0 Collaboration Meeting P. Bhat 14

RUNII ADVANCED ANALYSIS METHODS GROUP Home Page Chief Co-ordinator: Pushpa Bhat Mini. Workshop Conveners Pushpa Bhat (D 0/Fermilab) Phil Koehn (CDF/OSU) Louis Lyons (CDF/Oxford) Harrison B. Prosper (D 0/FSU) Goals • Explore and develop advanced analysis methods & tools • Perform example analyses • Serve as a resource to the experiments at the Tevatron Agenda • Study and evaluate Multivariate analysis methods, e. g. : • Random Grid Search, Neural Networks, Probability Density Estimation Methods, Genetic Algorithms, PCA, IDA, . . Next Meeting Working Groups Proceedings Documentati on Interesting Sites • Build analysis tools for use in friendly environments, e. g. : • Python, JPython • Matlab, Mathcad • Address Statistical Analysis Issues • Signal Significance, Confidence Limits, Comparing Hypotheses, Combining Experiments, etc. http: //www-d 0. fnal. gov/~pushpa/run 2 aag/ April 25, 2001 D 0 Collaboration Meeting P. Bhat 15