ATLAS Second-Level Electron/Jet Neural Discriminator based on Nonlinear Independent Components

Abstract

The Large Hadron Collider at CERN is designed to collide proton bunches at 40 MHz in the ATLAS detector. This event rate must be reduced to about 200 Hz for storage by selecting the most interesting events for physics analysis. This will be achieved in ATLAS with a three level trigger system. This work proposes the application of the Nonlinear Independent Component Analysis model in the ATLAS second-level trigger to estimate underlying factors of calorimeter information. As a pre-processing step, the calorimeter Region of Interest data are transformed into concentric energy deposition rings. In order to take advantage of the full detector segmentation, the feature extraction procedure is performed at calorimeter layer level. The independent features estimated from each layer were concatenated into a single feature vector, which was used to feed the input nodes of a neural network classifier for highly-efficient electron identification. The proposed approach achieved 97.0% of electron identification for 7.7% of background noise (QCD Jets) acceptance.

Otimização do Sistema Neural de Seleção Online de Eventos num Detector de Partículas através do Processamento Estatístico de Sinais

“…Em seguida,é definido um número de sorteios (neste caso foram 10 sorteios) e também são selecionados os conjuntos de treino, teste e validao de cada subgrupo. Na sequênciaé treinado um discriminador neural para cada sorteio, utilizando os subgrupos definidos anteriormente [12]. O processo de validação cruzadá e exemplificado na Figura 6.…”