Discriminating signal from background using neural networks: Application to top-quark search at the Fermilab Tevatron

Abstract

The application of neural networks in high energy physics to the separation of signal from background events is studied. A variety of problems usually encountered in this sort of analysis, from variable selection to systematic errors, are presented. The top-quark search is used as an example to illustrate the problems and proposed solutions. ͓S0556-2821͑96͒06013-4͔ PACS number͑s͒: 14.65. Ha, 02.50.Sk, 13.85.Qk It is well known that neural networks ͑NN's͒ are useful tools for pattern recognition. In high energy physics, they have been used or proposed as good candidates for tasks of signal versus background classification. However, most of the existing studies are somewhat academic, in the sense that they essentially compare the NN performances with other classical techniques of classification using Monte Carlo ͑MC͒ events for that purpose. In realistic applications, real events should be analyzed and compared with simulated events, introducing systematic effects which have to be taken into account and could significantly modify the efficiency of the analysis. We try to give some insight in this direction using the top quark search at the Fermilab Tevatron as illustration. The top quark has been observed by the Collider Detector at Fermilab ͑CDF͒ ͓1͔ and D0 ͓2͔ collaborations. Recently, NN's have been applied to experimental top-quark searches by the D0 Collaboration ͓3͔, for a fixed top-quark mass, concluding that NN's are more efficient than traditional methods, in agreement with previous parton level studies ͓4͔.In this paper we continue and complete the analysis of Ref. ͓4͔ for the top-quark search at the Tevatron. A more realistic study is performed by including parton hadronization and detector simulation with jet reconstruction. In addition, contrary to Ref. ͓4͔ where the top mass was fixed, the present study is valid for a large range of top mass values. Moreover, the number of kinematical variables considered is enlarged and different ways of selecting subsets of the most relevant ones to the process under consideration are discussed. Finally, the influence of systematic errors on the NN results is studied.The analysis is focused on the top-quark search at the pp Fermilab Tevatron operating at ͱsϭ1.8 TeV. The onecharged-lepton channel, pp→tt→l j j j j with lϭe Ϯ , Ϯ , is considered as the signal to look for. The main background is pp→W j j j j→l j j j j. Exact tree-level amplitudes with spin correlations were used to generate MC samples for both signal and background. The latter was evaluated with VECBOS ͓5͔. The CTEQ structure functions ͓6͔ at the scale Qϭm t (Qϭ͗p t ͘) for the top signal ͑background͒ were utilized. The LUND fragmentation model ͓7͔ was used to hadronize the quarks and/or gluons. The obtained events were passed through a fast MC program which simulates the segmentation of a D0-like calorimeter. Jets are reconstructed with a simple algorithm based on the routine used in the LUND package and electrons are defined as isolated clusters with more than 90% electromagnetic energy.Uncorrelate...

Support vector machine classification on a biased training set: Multi-jet background rejection at hadron colliders

“…Positions and angles are expressed in a cylindrical coordinate system, with the z axis along the proton beam, azimuthal angle φ and polar angle θ. The following variables are defined according to these principles 1 : the pseudorapidity η = −ln[tan(θ/2)], the transverse energy E T = E sin θ (as measured by the calorimetry), the transverse momentum p T = p sin θ (as measured by the tracking systems) and the angular distance between two particles in the η−φ space, R = (∆η) 2 + (∆φ) 2 .…”

“…We decided to explore the use of the SVM algorithm (described in Section 2) because of several advantages with respect to other multivariate techniques. For example, Artificial Neural Networks, commonly used in High Energy Physics [2], require to arbitrarily set the complexity of the classifier (i.e. the number of neurons and layers of the net), the training may converge to local minima and, usually, large training sets are needed to finely map the input space.…”

Statistical Pattern Recognition: Application to νμ→ντOscillation Searches Based on Kinematic Criteria

“…During last decade, neural networks have been widely used to solve High Energy Physics problems (see [27] for a introduction to neural networks techniques and applications to HEP). Multi-layer perceptrons efficiently recognize signal features from an, a priori, dominant background environment ( [28,29]). …”

Mixture-of-theories Training: Can We Find New Physics and Anomalies Better by Mixing Physical Theories?

“…Attempts to train a classifier on high-dimensional features to select a data sample where the signal is enhanced have existed since 30 years (or more), see e.g. [1][2][3][4]. Typically statistical tests (goodness of fit of the SM hypothesis, hypothesis test for the BSM hypothesis) are done with a selection on the classifier output, i.e.…”

“…(5,7,2,10) signals correspond to (50K, 70K, 20K, 100K) events8 Specifically for channel 1, we trained ML models on(1,6,11,16,21,26,31,36,37) signals.For channel 2a, we trained ML models on(1,8,15,22,29,36,43,50,51) signals. For channel 2b, we trained ML models on(1,3,5,7,9,11,13,15,17,18) signals and for channel 3, we trained ML models on(1,11,21,31,41,51,61, 71) signals 9. We took two SSAs and five MoTs trained on the same number of events to compare these different schemes.…”