Generative-Based Correction of Detector Acceptance in High Energy and Nuclear Physics Experiments
College
College of Sciences
Department
Computer Science
Graduate Level
Doctoral
Graduate Program/Concentration
AI/ML
Presentation Type
Poster Presentation
Abstract
High Energy Physics (HEP) and Nuclear Physics (NP) experiments are essential for probing the fundamental properties and interactions of elementary particles. These experiments rely on high-energy collisions, with resulting interactions captured by sophisticated detector systems that measure key properties such as energy, momentum, and particle type. However, the measured data are inherently affected by detector effects, including smearing, acceptance limitations, inefficiencies, and misidentification, which introduce distortions that obscure the true physical distributions.
In this paper, we focus on correcting detector acceptance effects to recover the underlying phase space of particle interactions. We utilize a machine learning-based approach that learns the detector response and enables extrapolation beyond measured regions. Specifically, we apply fiducial cuts using a parametric model of the CEBAF Large Acceptance Spectrometer (CLAS) at Jefferson Lab, validating our method on Monte Carlo-generated pseudodata for single-pion photoproduction. Our proposed neural network architecture not only models the measured particle distributions within the detector’s acceptance but also reconstructs unmeasured regions, offering a data-driven solution for unfolding phase space distortions. By integrating this approach with smearing corrections, we establish a comprehensive framework for extracting vertex-level scattering information from detector-distorted data, demonstrating the potential of machine learning in enhancing experimental data analysis in HEP and NP.
Keywords
Generative AI, Acceptance Problems, High Energy Physics (HEP), Detector Effects
Generative-Based Correction of Detector Acceptance in High Energy and Nuclear Physics Experiments
High Energy Physics (HEP) and Nuclear Physics (NP) experiments are essential for probing the fundamental properties and interactions of elementary particles. These experiments rely on high-energy collisions, with resulting interactions captured by sophisticated detector systems that measure key properties such as energy, momentum, and particle type. However, the measured data are inherently affected by detector effects, including smearing, acceptance limitations, inefficiencies, and misidentification, which introduce distortions that obscure the true physical distributions.
In this paper, we focus on correcting detector acceptance effects to recover the underlying phase space of particle interactions. We utilize a machine learning-based approach that learns the detector response and enables extrapolation beyond measured regions. Specifically, we apply fiducial cuts using a parametric model of the CEBAF Large Acceptance Spectrometer (CLAS) at Jefferson Lab, validating our method on Monte Carlo-generated pseudodata for single-pion photoproduction. Our proposed neural network architecture not only models the measured particle distributions within the detector’s acceptance but also reconstructs unmeasured regions, offering a data-driven solution for unfolding phase space distortions. By integrating this approach with smearing corrections, we establish a comprehensive framework for extracting vertex-level scattering information from detector-distorted data, demonstrating the potential of machine learning in enhancing experimental data analysis in HEP and NP.