Bulletin of the American Physical Society
APS April Meeting 2022
Volume 67, Number 6
Saturday–Tuesday, April 9–12, 2022; New York
Session K08: Data Analysis, AI and ML IRecordings Available
|
Hide Abstracts |
Sponsoring Units: DPF GDS Chair: Wei Shi, Stony Brook University Room: Juilliard |
Sunday, April 10, 2022 1:30PM - 1:42PM |
K08.00001: Solving Combinatorial Problems at Particle Colliders Using Machine Learning Anthony Badea, Lawrence Lee High-multiplicity signatures at particle colliders can be given by Standard Model as well as BSM processes. In such signatures, difficulties arise from the large dimensionality of the kinematic space. For final states of indistinguishable particle signatures, this results in a large combinatorial problem that hides underlying kinematic information. We explore using a neural network that includes a Lorentz Layer to effectively extract high-dimensional correlations. We use the case of squark decays in RPV SUSY as a benchmark, comparing the performance to that of classical methods. We demonstrate significant improvements. |
Sunday, April 10, 2022 1:42PM - 1:54PM |
K08.00002: Combining Deep Sets and Dynamic Graph Convolutional Neural Networks for Collider Event Classification Delon Shen, Peter E Onyisi, Jesse D Thaler At experiments such as those occurring at the Large Hadron Collider, classifying between signal and background events is vital for the success of experimental analyses. Methods traditionally employed to classify events suffer from several shortcomings stemming from the fact that traditional machine learning methods restrict us from representing events in a way that both respects the permutation symmetry of the objects like jets or leptons and maximizes the kinematic information the architecture can take as input. In this talk we present novel methods incorporating modern machine learning techniques which allow us to represent events more naturally as variable sized lists of objects like jets or leptons and inherently utilize the permutation symmetry and relational information between objects within an event. We then recreate event classification problems faced by recent experimental analyses of the H→ττ channel and compare the performance of our architectures against more traditional methods to find that these novel architectures lead to a threefold increase in event classification performance over traditional methods employed for experimental analyses. |
Sunday, April 10, 2022 1:54PM - 2:06PM |
K08.00003: Monotonic and Robust Neural Networks Ouail Kitouni, Mike Williams, Niklas Nolte The Lipschitz constant of the map between the input and output space represented by a neural network is a natural metric for assessing the robustness of the model. We present a new method to constrain the Lipschitz constant of dense deep learning models that can also be generalized to other architectures. The method relies on a simple weight normalization scheme during training that ensures the Lipschitz constant of every layer is below an upper limit specified by the analyst. A simple residual connection can then be used to make the model monotonic in any subset of its inputs, which is useful in scenarios where domain knowledge dictates such dependence. Examples can be found in algorithmic fairness requirements or, as presented here, in the classification of the decays of subatomic particles produced at the CERN Large Hadron Collider. Our normalization is minimally constraining and allows the underlying architecture to maintain higher expressiveness compared to other techniques which aim to either control the Lipschitz constant of the model or ensure its monotonicity. We show how the algorithm was used to train a powerful, robust, and interpretable discriminator for heavy-flavor decays in the LHCb real-time data-processing system. |
Sunday, April 10, 2022 2:06PM - 2:18PM |
K08.00004: Moment Unfolding using Deep Learning Krish Desai, Benjamin Nachman, Jesse D Thaler Deconvolving ( ‘unfolding’) detector distortions is a critical step in the comparison of cross section measurements with theoretical predictions. However, most of these approaches require binning while many predictions are at the level of moments. We develop a new approach to directly unfold distribution moments as a function of any other observables without having to first discretize. Our Moment Unfolding technique uses machine learning and is inspired by Generative Adversarial Networks (GANs). We demonstrate the performance of this approach using jet substructure measurements in collider physics. We also discuss challenges with unfolding all moments simultaneously, drawing connections to the renormalization of the partition function. |
Sunday, April 10, 2022 2:18PM - 2:30PM |
K08.00005: Using symmetries and transformers to build better latent spaces for di-jet representation learning Radha R Mastandrea, Benjamin Nachman, Barry M Dillon We investigate a method of model-agnostic anomaly detection at the Large Hadron Collider (LHC) through studying jets, collimated sprays of particles produced in high-energy collisions. We train a transformer neural network to encode simulated quark and gluon ("physical-space") di-jets into a high-dimensional ("latent-space") representation. We optimize the representations using the contrastive loss, which tells the transformer to preserve known physical symmetries of the di-jet when making its representations. We then run a signal (consisting of di-jets from simulated new particles) vs. background (di-jets from quarks and gluons) linear classifier test on the latent-space representations and compare the classifier performance with that of a dense binary classifier neural net trained on the physical-space di-jets. We finally explore the possibility of using the transformer net to encode LHC events at the event level -- rather than at the jet level -- into the latent space. Such a representation makes no assumptions on the relationships between particles in a given event while still enforcing known physical symmetries for particle collision events on the whole. This could provide a maximally agnostic event representation that could be used to search for new physics of any type. |
Sunday, April 10, 2022 2:30PM - 2:42PM |
K08.00006: Quark and Gluon Tagging Calibration with the ATLAS detector Haoran Zhao, Shih-Chieh Hsu, Ke Li, Wanyun Su The separation of quark and gluon initiated jets(q/g jets) is crucial to enhance the reach of many new physics searches at the ATLAS experiment in the Large Hadron Collider. A tagger serving as a tool to distinguish quark and gluon jets is developed based on the Boosted Decision Tree using charged-particle track observables associated with the jet. However, quark-versus-gluon jet tagging is difficult to be calibrated due to the difficulty of the hadronization modeling. To improve the performance of the tagger, a "matrix method" is applied to extract the q/g distributions to obtain a scale factor which is a ratio between data and Monte Carlo. The data taken from 2015 to 2018 with an integrated luminosity of 139.0 fb-1 in pp collisions at √s = 13 TeV with the ATLAS detector are used to calibrate the tagger with two control samples to select dijet and gamma+jet events, providing various gluon and quark enriched samples. In this talk, the latest results of calibration and systematic uncertainties will be presented. |
Sunday, April 10, 2022 2:42PM - 2:54PM |
K08.00007: Can You Hear the Shape of A Jet? Rikab Gambhir, Jesse D Thaler, Akshunna S Dogra, Demba Ba \begin{abstract} The identification of interesting substructures within jets is an important tool to search for new physics and probe the Standard Model. In this paper, we present a new set of shape-based observables, $G$-shapeliness, which generalizes the $N$-jettiness from point clusters to any extended shape $G$. We show that the K-Deep Simplex dictionary learning framework is, with some modifications, dual to the 2-Wasserstein metric on energy flows, from which observables can be interpreted as the optimal transport distance to an idealized energy flow distribution. We then use the modified KDS framework to compute $G$-shapeliness values for a toy model in order to discriminate between topology-distinct events, and then finally apply the framework to analyze jet substructure. \end{abstract} |
Sunday, April 10, 2022 2:54PM - 3:06PM |
K08.00008: Jet-level Anomaly Detection in the ATLAS Y->XH Search with a Variational Recurrent Neural Network Gabriel P Matos We present the implementation of machine-learning based anomaly detection to a generic dijet resonance search with LHC proton collision data collected by the ATLAS Experiment. Specifically, we train over data with a novel variational recurrent neural network (VRNN) that identifies anomalous jets solely based on their inconsistency with the background only hypothesis. The VRNN produces a per-jet anomaly score, whose performance is evaluated across a wide variety of new physics topologies to ensure model-independence, across which a selection on the anomaly score is shown to yield between 5-30% increase in significance of signal over background. We also describe the first application of this method to ATLAS data by way of a search for generic new bosons Y and X in association with a Higgs boson. We have utilized the anomaly score to define a model-independent signal region in this analysis, marking the first use of fully unsupervised machine learning in an ATLAS physics search. |
Follow Us |
Engage
Become an APS Member |
My APS
Renew Membership |
Information for |
About APSThe American Physical Society (APS) is a non-profit membership organization working to advance the knowledge of physics. |
© 2024 American Physical Society
| All rights reserved | Terms of Use
| Contact Us
Headquarters
1 Physics Ellipse, College Park, MD 20740-3844
(301) 209-3200
Editorial Office
100 Motor Pkwy, Suite 110, Hauppauge, NY 11788
(631) 591-4000
Office of Public Affairs
529 14th St NW, Suite 1050, Washington, D.C. 20045-2001
(202) 662-8700