Session M11: Machine learning and AI

SYMBA: Symbolic Computation of Amplitudes in High-Energy Physics with Machine Learning

While machine learning is widely used for numerical calculations in high-energy physics, analytical calculations using machine learning are still in their infancy. In this talk, we show that a transformer model is able to map correctly interaction amplitudes to their square, averaged over initial and summed over final particle degrees of freedom, for 97.6% and 99% of QCD and QED processes, respectively, at a speed that is up to two orders of magnitude faster than current symbolic computation frameworks. We discuss the performance of the current model, its limitations and possible future directions for this work.   

Simulating 2+1D Lattice Quantum Electrodynamics at Finite Density with Neural Flow Wavefunctions

We present a neural flow wavefunction, Gauge-Fermion FlowNet, and use it to simulate 2+1D lattice compact quantum electrodynamics with finite density dynamical fermions. The gauge field is represented by a neural network which parameterizes a discretized flow-based transformation of the amplitude while the fermionic sign structure is represented by a neural net backflow. This approach directly represents the U(1) degree of freedom without any truncation, obeys Guass's law by construction, samples autoregressively avoiding any equilibration time, and variationally simulates Gauge-Fermion systems with sign problems accurately. In this model, we investigate confinement and string breaking phenomena in different fermion density and hopping regimes. We study the phase transition from the charge crystal phase to the vacuum phase at zero density, and observe the phase separation and the net charge penetration blocking effect under magnetic interaction at finite density. In addition, we investigate a magnetic phase transition due to the competition effect between the kinetic energy of fermions and the magnetic energy of the gauge field. With our method, we further note potential differences on the order of the phase transitions between a continuous U(1) system and one with finite truncation. Our state-of-the-art neural network approach opens up new possibilities to study different gauge theories coupled to dynamical matter in higher dimensions.   

Gauge Equivariant Neural Networks for 2+1D U(1) Gauge Theory Simulations in Hamiltonian Formulation

Gauge Theory plays a crucial role in many areas in science, including high energy physics, condensed matter physics, and quantum information science. In quantum simulations of lattice gauge theory, an important step is to construct a wave function that obeys gauge symmetry. Our work develops gauge equivariant neural network wave function techniques for simulating continuous-variable quantum lattice gauge theories in the Hamiltonian formulation. We have applied the gauge equivariant neural network approach to find the ground state of 2 + 1-dimensional lattice gauge theory with U(1) gauge group using variational Monte Carlo. We have benchmarked our approach against the state-of-the-art complex Gaussian wave functions, demonstrating improved performance in the strong coupling regime and comparable results in the weak coupling regime.   

Unbinned and Profiled Unfolding

Unfolding is an important procedure in particle physics experiments which corrects for detector effects and provides differential cross section measurements that can be used for a number of downstream tasks, such as extracting fundamental physics parameters. Traditionally, unfolding is done by discretizing the target phase space into a finite number of bins and is limited in the number of unfolded variables. Recently, there have been a number of proposals to perform unbinned and high-dimensional unfolding with machine learning. However, none of these methods allow for simultaneously constraining (profiling) nuisance parameters and thus they conflict with the standard approach, where data are unfolded and profiled. We propose a new machine learning-based unfolding method that can process and profile unbinned data. We first demonstrate the method with simple Gaussian examples and then show the impact on a simulated Higgs boson cross section measurement.   

GA Ansatz Optimization on a QAE for HEP Anomaly Detection

The Standard Model (SM) of High Energy Physics (HEP) has proven itself to be one of the most accurate scientific theories ever formulated. It is no surprise that current HEP research progresses slowly to find the correct Beyond Standard Model (BSM) extension. While experiments like the Large Hadron Collider (LHC) can directly probe exotic phenomena, such experiments produce an overwhelming amount of data. Therefore, it becomes imperative to develop techniques to easily sift through the background, SM processes and identify the new, BSM physics. Furthermore, because the BSM deviations could be caused by any number of theoretical BSM processes, current HEP anomaly detection should be as model-agnostic as possible, while remaining highly-sensitive. We propose to further develop anomaly detection algorithms using breakthrough quantum machine learning techniques. To go beyond fixed-ansatz quantum circuits models, we propose a meta-optimization of ansatz using a Genetic Algorithm (GA). The models and optimization are benchmarked on the LHC Olympics 2020 dataset, an already well adopted dataset when it comes to evaluating anomaly detection performance   

Moment Pooling: Gaining Performance and Interpretability Through Physics Inspired Product Structures

As machine learning begins to play an increasingly larger role in high energy physics, it is important to understand and interpret what precisely these models learn. In this work, we propose Moment Pooling architectures, which generalizes the summation in standard Deep Sets architectures to an arbitrary multivariate moments or cumulants. This can be used to drastically decrease latent space sizes, significantly improving the model's interpretability while maintaining performance. We show that this is particularly useful in jet physics, where many existing useful jet observables can be naturally expressed in this form. We then show several examples of how the Moment Pooling architecture may be used in jet tagging.   

Convolutional Visual Networks for Secondary Vertexing in NOvA

NOvA is a Fermilab-based long-baseline neutrino experiment designed primarily to measure electron neutrino appearance and muon neutrino disappearance in a predominantly muon neutrino beam. NOvA also has a broad physics program including neutrino cross-section measurements in the near detector and searches for beyond-standard model phenomena in both detectors. The ability to accurately identify and reconstruct the energies of individual particles within an event is crucial for cross section measurements. NOvA's current reconstruction techniques lack the ability to identify in-detector reinteractions or decays, which can lead to incorrect assignment of energy within an event or misidentification of tracks within the detector. In this contribution, we describe progress toward training a convolutional neural network to provide a "secondary vertexing" capability. We present this methodology in detail and report on the impact this capability will have on track/event reconstruction in NOvA.   

Diffusion-Based Generative Modeling for LArTPC Images

Advances in machine learning have changed how we understand data, especially the recent advent of generative modeling, which allows the creation of novel examples from a given dataset. Seeking to make use of these new methods, I applied a modern diffusion-based generative model (Yang Song & Stefano Ermon, 2019) to the PILArNet public dataset. The data consists of 2D images of simulated particle tracks and showers detected within a Liquid Argon Time Projection Chamber (LArTPC). In this presentation, I will outline the methodology used in the algorithm, demonstrate the quality of the generated images, and provide insight into the future applicability of this approach.