In 2017 the ATLAS experiment implemented an ensemble of neural networks (NeuralRinger algorithm) dedicated to improving the performance of filtering events containing electrons in the high-input rate online environment of the Large Hadron Collider at CERN, Geneva. The ensemble employs a concept of calorimetry rings. The training procedure and final structure of the ensemble are used to minimize fluctuations from detector response, according to the particle energy and position of incidence. This reposiroty is dedicated to hold all tuning scripts for each subgroup in the ATLAS e/g trigger group.
WARNING: Do not change anything in the master branch.
Proliferation of big data applications is an outcome of the technological breakthrough during the Digital Era. Rare events of interest immersed in a large amount of data ensure in a particular scenario where online decision taking is needed to discard irrelevant information while maintaining potentially useful events. Offline analysis provides the final decision through the scrutiny of those events. In such scenarios, the rare incidence of interesting events does not allow the creation of a bias in the observations by the online filtering process. At the same time, a high filtering efficiency is required in order to occupy the output bandwidth only with events potentially interesting for the offline analysis.
High energy physics experiments were early pioneers of dealing with the big data applications and their legacy is currently being continued by the experiments at the Large Hadron Collider (LHC) at CERN, Geneva, Switzerland. The LHC is the currently leading edge collider, providing high energy collisions through two opposite way circular beams that allow to study rare physics processes. One successful example is the Higgs boson discovery at the LHC in 2012, only half century after its theoretical prediction, which resulted in the laureation of P. Higgs and F. Englert with the Physics Nobel prize. ATLAS, which is the largest experiment at the LHC, played an important role in this discovery.
To allow the observation of rare physics processes like the Higgs boson production, the protons at the LHC are placed along the ring to collide typically every 25ns, therefore providing high-event input rate (over 30MHz) to the experiments. It is not viable to record and process all these events due to the large amount of information generated, i.e. the ATLAS experiment alone would require a bandwidth of ~70GB/s while the data taking period can expand for decades. Therefore, online filtering, performed by the ATLAS Trigger and Data Acquisition System (TDAQ), is required to select interesting events to reduce the recorded rate to viable levels for storage and further offline processing.
The trigger system relies on pattern recognition to identify physics objects of interest to the ATLAS analyses. These objects are filtered in two sequential levels in order to achieve the low latency driven by the LHC collision rate. The first level (L1) is based on hardware and has a latency of less than 2.5ms, while reducing the input rate to, at most, 100kHz. The second level, called the High Level Trigger (HLT), has a software-based filtering with a mean target latency of 550ms and an average output rate of 1kHz.
Each physics object has its own filtering features. The physics objects studied here are focused on electron-based channels, which are found in many interesting physics phenomena, for example the decays of the Higgs boson. Electron pattern recognition relies on discriminating information of the ATLAS calorimeter system for energy measurement and its inner tracking Detector for signal patterns through particle tracks, which involves image processing-like algorithms. The latter requires higher processing resources, therefore, as a way to achieve lower latency for electron triggering, early discrimination evaluates only calorimetry information.
In order to collect more data required for conclusive experimental results on even rarer physics processes, the LHC has increased the number of collisions. One way to achieve this, is to squeeze the beam, which results in higher number of collisions per bunch-crossing (pile-up). This generates higher pressure on the trigger system, where not only there is more information to be processed, requiring more bandwidth and processing resources, but also the decision taking process is harder to perform once signals overlap deteriorating the distinctive patterns used for particle identification. This is also the case for electron identification patterns that are affected by the ever-increasing pile-up at the LHC.
To account for this effect, the ATLAS experiment upgraded in 2017 the initial selection performed in the HLT electron filtering to an ensemble of neural networks algorithm (NeuralRinger). The ensemble is fed only from calorimetry information contained in concentric rings of energy deposition, which compacts the information through physics expert knowledge, while keeping the discriminating patterns.
zee: Dedicated branch to support all tuning developments for Zee decay and electron triggers higher than 15 GeV for the fast calo step (assigned to João);zee_el: Dedicated branch to support all tuning developments for Zee decay and electron triggers higher than 15 GeV for the trigger electron step (assigned to João);jpsi: Dedicated branch to support all tuning developments for Jpsiee decay and electron triggers below 15 GeV for the fast calo step (assigned to Micael);zrad: Dedicated branch to support all tuning developments for Z radiative decay and photon triggers for the fast calo step (assigned to Juan).
saphyra: Tuning package used to derive the ringer (see here);prometheus: The ATLAS analysis framework to test the tuning (see here);
- Dr. João Victor da Fonseca Pinto, UFRJ/COPPE, CERN/ATLAS ([email protected]) [maintainer, developer]
- Dr. Werner Freund, UFRJ/COPPE, CERN/ATLAS ([email protected]) [developer]
- Msc. Micael Veríssimo de Araújo, UFRJ/COPPE, CERN/ATLAS ([email protected]) [developer]
WARNING: This is a public repository.