Conveners
Computation, Machine Learning, and AI: COMP 1E
- Gordon Watts (University of Washington (US))
- Horst Wahl (Florida State University)
- Aishik Ghosh (University of California Irvine (US))
Computation, Machine Learning, and AI: COMP 2E
- Gordon Watts (University of Washington (US))
- Oleksandr Viazlo (Florida State University (US))
Computation, Machine Learning, and AI: COMP 3E
- Ben Nachman (Lawrence Berkeley National Lab. (US))
Computation, Machine Learning, and AI: COMP 4E
- Ghosh Aishik (Berkeley Lab)
- Ben Nachman (Lawrence Berkeley National Lab. (US))
Computation, Machine Learning, and AI: COMP 5E
- Ghosh Aishik (Berkeley Lab)
- Muge Karagoz (University of Maryland (US))
In this talk, we will introduce a technique to train neural networks into being good event variables, which are useful to an analysis over a range of values for the unknown parameters of a model.
We will use our technique to learn event variables for several common event topologies studied in colliders. We will demonstrate that the networks trained using our technique can mimic powerful,...
To perform theoretical calculations and comparisons with collider data, it must first be corrected for various detector effects, namely noise processes, detector acceptance, detector distortions, and detector efficiency; this process is called “unfolding” in high energy physics (or “deconvolution” elsewhere). While most unfolding procedures are carried out over only one or two binned...
We examine the problem of unfolding in particle physics, or de-corrupting observed distributions to estimate underlying truth distributions, through the lens of Empirical Bayes and deep generative modeling. The resulting method, Neural Empirical Bayes (NEB), can unfold continuous multi-dimensional distributions, in contrast to traditional approaches that treat unfolding as a discrete linear...
As the search for physics beyond the Standard Model widens, 'model-agnostic' searches, which do not assume any particular model of new physics, are increasing in importance. One promising model-agnostic search strategy is Classification Without Labels (CWoLa), in which a classifier is trained to distinguish events in a signal region from similar events in a sideband region, thereby learning...
Excursion is a tool to efficiently estimate level sets of
computationally expensive black box functions using Active Learning.
Excursion uses a Gaussian Process Regression as a surrogate model for
the black box function. It queries the target function (black box) iteratively in order to increase the available information regarding the desired level sets. We implement Excursion using...
Data Quality Monitoring (DQM) is an important process of collecting high quality data for physics analysis. Currently, the workflow of DQM is manpower intensive to scrutinize and certify hundreds of histograms. Identifying good quality and reliable data is necessary to make accurate predictions, simulations, therefore anomalies in the detector must be timely identified to minimize data loss....
Application of machine learning methods in high energy physics has received tremendous success in recent years with rapidly growing use cases. A key aspect in improving the performance of a given machine learning model has been the optimization of its hyperparameters which is usually computationally expensive. A framework has been developed to provide a high-level interface for automatic...
The intelligent Data Delivery Service (iDDS) has been developed to cope with the huge increase of computing and storage resource usage in the coming Large Hadron Collider (LHC) data taking. It has been designed to intelligently orchestrate workflow and data management systems, decoupling data pre-processing, delivery, and main processing in various workflows. It is an experiment-agnostic...
The Reproducible Open Benchmarks for Data Analysis Platform (ROB) is a platform that allows for the evaluation of different data analysis algorithms in a controlled competition-style format [1]. One example for such a comparison and evaluation of different algorithms is the “The Machine Learning Landscape of Top Taggers” paper, which compiled and compared multiple different top tagger neural...
We introduce CaloFlow, a fast detector simulation framework based on normalizing flows. For the first time, we demonstrate that normalizing flows can reproduce many-channel calorimeter showers with extremely high fidelity, providing a fresh alternative to computationally expensive GEANT4 simulations, as well as other state-of-the-art fast simulation frameworks based on GANs and VAEs. Besides...
We put forth a technique to generate images of particle trajectories (particularly electrons and protons) in a liquid argon time projection chamber (LArTPC). LArTPCs are a type of particle physics detector used by several current and future experiments focused on studies of the neutrino. We implement a quantized variational autoencoder and an autoregressive model which produces images...
Athena is the software framework used in the ATLAS experiment throughout the data processing path, from the software trigger system through offline event reconstruction to physics analysis. The shift from high-power single-core CPUs to multi-core systems in the computing market means that the throughput capabilities of the framework have become limited by the available memory per process. For...
We present a novel implementation of classification using boosted decision trees (BDT) on field programmable gate arrays (FPGA). Two example problems are presented, in the separation of electrons vs. photons and in the selection of vector boson fusion-produced Higgs bosons vs. the rejection of the multijet processes. The firmware implementation of binary classification requiring 100 training...
The hls4ml library [1] is a powerful tool that provides automated deployment of ultra low-latency, low-power deep neural networks. We extend the hls4ml library to recurrent architectures and demonstrate low latency by considering multiple benchmark applications. We consider Gated Recurrent Units (GRU) and Long Short Term Memory(LSTM) Models trained using the CERN Large Hadron Collider Top...
This talk introduces and shows the simulated performance of an FPGA-based technique to improve fast track finding in the ATLAS trigger. A fast track trigger is being developed in ATLAS for the High Luminosity upgrade of the Large Hadron Collider (HL-LHC), the goal of which is to provide the high-level trigger with full-scan tracking at 100 kHz in the high pile-up conditions of the HL-LHC....
The high collision energy and luminosity of the LHC allow studying jets and hadronically-decaying tau leptons at extreme energies with the ATLAS detector. These signatures lead to topologies with charged particles, which are reconstructed as tracks with the ATLAS inner detector, at an angular separation smaller than the size of a charge cluster in the ATLAS pixel detector, forming merged pixel...
We report on the development of a track finding algorithm for the Fermilab Muon g-2 Experiment’s straw tracker using advanced Deep Learning techniques. Taking inspiration from original studies by the HEP.TrkX project, our algorithm relies on a Recurrent Neural Network with bi-directional LSTM layers to build and evaluate track candidates. The model achieves good performance on a 2D...
One of the key sub-detectors of the CMS experiment, located at the CERN Large Hadron collider, is the electromagnetic calorimeter (ECAL). This homogeneous calorimeter is designed to detect electrons and photons with energies from as low as 500 MeV up to 1 TeV. The ECAL is a homogeneous calorimeter consisting of ~76,000 scintillating crystals arranged around the collision point in an 8m long...
Calibrating the pion energy response is a core component of reconstruction in the ATLAS calorimeter. Deep learning techniques have shown the best energy resolution for a wide range of particle momenta [1]; to further improve the pion energy resolution, a Mixture Density Network (MDN) based deep learning algorithm is explored. In addition to estimating the energy, the MDN also estimates the...
As an unsupervised machine learning strategy, optimal transport (OT) has been applied to jet physics for the computation of distance between collider events. Here we generalize the Energy Mover’s Distance to include both the balanced Wasserstein-2 (W2) distance and the unbalanced Hellinger-Kantorovich (HK) distance. Whereas the W2 distance only allows for mass to be transported, the HK...
Deep learning techniques have gained tremendous attention from researchers in many fields, including particle physics. However such techniques typically do not capture model uncertainty. Bayesian models offer a solid framework to quantify the uncertainty, but they normally come with a high computational cost. A recent paper develops a new theoretical framework casting dropout in Neural...
A framework is presented to extract and understand decision-making information from a deep neural network classifier of jet substructure tagging techniques. The general method studied is to provide expert variables that augment inputs (“eXpert AUGmented” variables, or XAUG variables), then apply layerwise relevance propagation (LRP) to networks that have been provided XAUG variables and those...
As the LHC prepares to enter its third run, analyses are increasingly focused on a drive for precision physics. One of the great tools for precision physics in this field is that of unfolding. This talk describes the development and usage of RooUnfold, RooFitUnfold, and RooUnfoldML in particle physics. Together they form a complete series of statistical software packages for the treatment...
The Mu2e Experiment at Fermilab is looking for neutrino-less conversion of a muon to an electron. The experiment requires an extremely efficient Cosmic Ray Veto (CRV) to detect cosmic muons and ignore them so they cannot be confused with a successful direct conversion. Similarly, noise generated by neutrons and gamma rays from muon beam production/transportation can challenge the operation of...
The IceCube Neutrino Observatory is designed to observe neutrinos interacting deep within the South Pole ice. It consists of 5,160 digital optical modules, which are arrayed over a cubic kilometer from 1,450 m to 2,450 m depth. At the lower center of the array is the DeepCore subdetector, which has a denser configuration that lowers the observable energy threshold to about 5 GeV and creates...
The IceCube Neutrino Observatory detects atmospheric and astrophysical neutrinos using a cubic kilometer of ice instrumented with optical sensors at the South Pole. Neutrinos are detected using these sensors which record the cone of light from Cherenkov radiation, emitted by charged particles moving faster than the speed of light in ice, allowing the event vertex of neutrino interactions to be...
Neutrinos offer a variety of insights into Standard Model physics that are not yet understood, including flavor oscillations and the neutrino mass ordering. One instrument being used to study neutrinos is the IceCube South Pole Neutrino Observatory, a cubic kilometer-scale Cherenkov detector over 1.5 km below the South Pole. An extension, the IceCube-Upgrade, is currently under development and...
In recent years, deep learning has played an emerging role in event reconstruction for neutrino experiments using liquid argon TPCs (LArTPCs), a high-precision particle imaging technology. Several algorithms have been developed to infer the 3D location of charge depositions in the detector. Furthermore, the development of 2D pixel-readouts naturally provides 3D positions. Therefore, there is a...
In high energy physics, Machine Learning (ML) has been applied to a broad range of problems: from jet tagging to particle identification, from the separation of signal over background, to fast simulation of event data, to to name a few. In this presentation, ML algorithms and techniques are explored to form lepton pairs (di-leptons) in a dark fermionic model.
In this model, the final-state...
The increasing number of high-performance computing centers around the globe is providing physicists and other researchers access to heterogeneous systems -- comprising multiple central processing units and graphics processing units per node -- with various platforms. However, it is more often than not the case that domain scientists have limited resources such that writing multiple...
