摘要:

After the upgrade of the HERA machine at DESY in 2001, an increase in the luminosity of a factor 5 is expected. Since the data output rate of the L2 trigger should be kept at the pre-upgrade level, a smarter way of preprocessing data has been developed, extracting the most physically relevant information in order to optimize the neural networks. We describe here the new neural preprocessor DDB2 (Data Distributed Board) and focus especially on the algorithmic principles. A general overview of the hardware implementations of such algorithms is then discussed, notably the use of the current, fast FPGA technology to combine parallelism and speed. ©2001 American Institute of Physics.

ISSN:0094-243X    

DOI:10.1063/1.1405265

出版商:AIP    

年代:1901

数据来源: AIP

摘要:

Neural networks implemented in hardware can perform pattern recognition very quickly, and as such have been used to advantage in the triggering systems of certain high energy physics experiments. Typically, time constants of the order of a few microseconds are required. In this paper, we present a new system. MAHARADJA, for evaluating MLP and RBF neural network paradigms in real time. The system is tested on a possible ATLAS muon triggering application suggested by the Tel Aviv ATLAS group, consisting of a 4-8-8-4 MLP which must be evaluated in 10 microseconds. The inputs to the net are dx/dz,x(z=0),dy/dz, andy(z=0),whereas the outputs give pt, tan(phi), sin(theta), and q, the charge. With a 10 MHz clock, MAHARADJAcalculates the result in 6.8 microseconds; at 20 MHz, which is readily attainable, this would be reduced to only 3.4 microseconds. The system can also handle RBF networks with 3 different distance metrics (Euclidean, Manhattan and Mahalanobis), and can simulate any MLP of 10 hidden layers or less. The electronic implementation is with FPGA’s, which can be optimized for a specific neural network because the number of processing elements can be modified. ©2001 American Institute of Physics.

ISSN:0094-243X    

DOI:10.1063/1.1405266

出版商:AIP    

年代:1901

数据来源: AIP

摘要:

Since 1998 feed-forward networks have been applied for the separation of hadronic WW-decays from background processes measured by the DELPHI collaboration at different center-of-mass energies of the Large Electron Positron collider at CERN. Prior to the publication of the 189 GeV results (1) intensive studies of systematic effects and uncertainties were performed. The methods and results will be discussed and compared to standard selection procedures. ©2001 American Institute of Physics.

ISSN:0094-243X    

DOI:10.1063/1.1405267

出版商:AIP    

年代:1901

数据来源: AIP

摘要:

We present a search for electroweak production of single top quarks in the DO&slash; detector at the Tevatron collider. After initial selections, the signal forms less than one percent of the background, and requires a powerful analysis tool to separate it from background. For this purpose, we employ the neural network package MLPfit, and train it on Monte Carlo (MC) models of two processes for signal, and on data and MC models of five processes for background. Based on an analysis of singularities in the Feynman diagrams for single-top production, we choose an optimal set of kinematic variables as inputs to the networks. We use separate networks for each signal-background pair. For the dominant backgrounds, we use sequential nets. ©2001 American Institute of Physics.

ISSN:0094-243X    

DOI:10.1063/1.1405268

出版商:AIP    

年代:1901

数据来源: AIP

摘要:

A neural mapping is developed to improve the overall performance of Tilecal, which is the hadronic calorimeter of the ATLAS detector. Feeding the input nodes of a multilayer feedforward neural network with the energy values sampled by the calorimeter cells in beam tests, it is shown that the original energy scale of pion beams is reconstructed over a wide energy range and linearity is significantly improved. As it happens for classical methods, a compromise between nonlinearity correction and the optimization of the energy resolution of the detector has to be accomplished. A hybrid training method for the neural mapping is proposed to achieve this design goal. Using the back-propagation algorithm, the method intercalates an epoch of training steps, for which the neural mapping mainly focus on linearity correction, with another block of training steps, in which the original energy resolution obtained by linearly combining the calorimeter cells becomes the main target. ©2001 American Institute of Physics.

ISSN:0094-243X    

DOI:10.1063/1.1405269

出版商:AIP    

年代:1901

数据来源: AIP

摘要:

A neural electron/jet discriminator based on calorimetry is developed for the second-level trigger system of the ATLAS detector. As preprocessing of the calorimeter information, a principal component analysis is performed on each segment of the two sections (electromagnetic and hadronic) of the calorimeter system, in order to reduce significantly the dimension of the input data space and fully explore the detailed energy deposition profile, which is provided by the highly-segmented calorimeter system. It is shown that projecting calorimeter data onto 33 segmented principal components, the discrimination efficiency of the neural classifier reaches 98.9&percent; for electrons (with only 1&percent; of false alarm probability). Furthermore, restricting data projection onto only 9 components, an electron efficiency of 99.1&percent; is achieved (with 3&percent; of false alarm), which confirms that a fast triggering system may be designed using few components. ©2001 American Institute of Physics.

ISSN:0094-243X    

DOI:10.1063/1.1405270

出版商:AIP    

年代:1901

数据来源: AIP

摘要:

We have investigated the possibility of employing neural networks for identification of electrons and &tgr; leptons with the DO&slash; detector in the upcoming run of the Tevatron collider at Fermilab. Preliminary results based on Monte Carlo simulations indicate that for any acceptable level of signal efficiency, neural networks consistently outperform covariance matrices so far employed for the same purpose. Using a subset of variables used by a covariance matrix, a properly trained neural network offers 2 times better background rejection for taus, and 10 times for electrons, at 90&percent; signal efficiency. Similar enhancements can be expected for identification of other objects (such as muons,borcjets, quark vs gluon jets, neutrinos, etc). ©2001 American Institute of Physics.

ISSN:0094-243X    

DOI:10.1063/1.1405271

出版商:AIP    

年代:1901

数据来源: AIP

摘要:

An unconventional solution for finding the location of event creation is presented. It is based on two feed-forward neural networks with fixed architecture, whose parameters are chosen so as to reach a high accuracy. The interaction point location is a parameter that can be used to select events of interest from the very high rate of events created at the current experiments in High Energy Physics. The system suggested here is tested on simulated data sets of the ZEUS Central Tracking Detector, and is shown to perform better than conventional algorithms. ©2001 American Institute of Physics.

ISSN:0094-243X    

DOI:10.1063/1.1405272

出版商:AIP    

年代:1901

数据来源: AIP

摘要:

The main application of neural networks (NN) in Higgs physics so far has been to optimize the signal over background ratio. The positive result obtained imply that the use of NN will lead to a big reduction in the integrated luminosity required for the discovery of the Higgs in RunII. Neural Networks have also been recently used in Higgs physics to set up tagging algorithms to identify the heavy flavor content of jets. Whereas in the previous studies the NN b-tagging methods used are channel-independent, a channel-dependent method has been used in the present work. The signalpp¯→WH→l&ngr;bb¯has been studied against the dominant backgroundpp¯→Wbb¯,in an attempt to improve the signal over background ratio by trying to push the invariant mass of the background events further away from the signal. This result would get the equivalent effect of an improved mass resolution. ©2001 American Institute of Physics.

ISSN:0094-243X    

DOI:10.1063/1.1405273

出版商:AIP    

年代:1901

数据来源: AIP

摘要:

A major goal of high energy physicists over the next few years is to reduce the uncertainty in our knowledge of the top quark mass. Neural network based methods may play a useful role in this regard, as is borne out by the preliminary results reported here. ©2001 American Institute of Physics.

ISSN:0094-243X    

DOI:10.1063/1.1405274

出版商:AIP    

年代:1901

数据来源: AIP