[论文解读] Ultra-High-Resolution Detector Simulation with Intra-Event Aware GAN and Self-Supervised Relational Reasoning

Hosein Hashemi, N. M. Hartmann+3 more|arXiv (Cornell University)|Mar 7, 2023

Medical Imaging Techniques and Applications被引用 24

一句话总结

本文提出 IEA-GAN，一种具备事内事件关系推理和自监督损失的 GAN，用于生成超高分辨率的探测器响应，在 Belle II PXD 数据上实现最先进的保真度（FID = 1.50），并实现存储量降低。

ABSTRACT

Simulating high-resolution detector responses is a computationally intensive process that has long been challenging in Particle Physics. Despite the ability of generative models to streamline it, full ultra-high-granularity detector simulation still proves to be difficult as it contains correlated and fine-grained information. To overcome these limitations, we propose Intra-Event Aware Generative Adversarial Network (IEA-GAN). IEA-GAN presents a Relational Reasoning Module that approximates an event in detector simulation, generating contextualized high-resolution full detector responses with a proper relational inductive bias. IEA-GAN also introduces a Self-Supervised intra-event aware loss and Uniformity loss, significantly enhancing sample fidelity and diversity. We demonstrate IEA-GAN's application in generating sensor-dependent images for the ultra-high-granularity Pixel Vertex Detector (PXD), with more than 7.5 M information channels at the Belle II Experiment. Applications of this work span from Foundation Models for high-granularity detector simulation, such as at the HL-LHC (High Luminosity LHC), to simulation-based inference and fine-grained density estimation. To our knowledge, IEA-GAN is the first algorithm for faithful ultra-high-granularity full detector simulation with event-based reasoning.

研究动机与目标

• 推动粒子物理中超高分辨率探测器响应的快速、节省存储的仿真。
• 应对具有高空间分辨率、粒度细致且在事件内相关的探测器图像的挑战。
• 开发一种能够捕捉事件内相关性和类别间关系的GAN架构。
• 提出新的损失项和关系推理模块，以提升保真度和多样性。

提出的方法

• 引入 Intra-Event Aware GAN (IEA-GAN) 及其 Relational Reasoning Module (RRM)，形成一个情境感知的事件图。
• 使用自监督的 2C 损失来建模二元类别之间的关系，以及 Uniformity 损失以防止模式坍缩。
• 用表示事件级可变性的随机自由度 (Rdof) 装饰生成器嵌入。
• 将嵌入投射到单位超球面以稳定训练并实现关系推理。
• 用对抗、2C、IEA 和 Uniformity 损失的组合进行训练，以捕捉事件内相关性与上下文。
• 使用图像级指标（FID）和物理级指标（occupancy、螺旋参数分辨率）进行评估。

(a) Rdof stands for Random degrees of freedom, which decorates the generator’s sensor/layer embedding with an event-level learnable embedding responsible for the generator’s intra-event correlation. The Relational Reasoning Modules (RRM) in the generator and the discriminator do the intra-event reas

实验结果

研究问题

• RQ1 如何忠实地生成具有事件内相关性的超高分辨率、类似探测器的图像？
• RQ2 关系推理和自监督损失是否能在高维探测数据上超越最先进的 GAN，在保真度和多样性方面带来提升？
• RQ3 生成的样本是否能再现与 Geant4 可比的真实 occupancy、能量分布和轨迹参数分辨率？
• RQ4 通过在线生成而非离线背景叠加，是否可能实现显著的存储减少？

主要发现

• IEA-GAN 在所评估模型中取得最低的 FID，1.50±0.16，超过 WGAN-gp、BigGAN-deep、ContraGAN 和 PE-GAN。
• IEA-GAN 捕捉到事件内相关性和层级相关上下文信息，提升样本多样性与真实感。
• 该模型在像素级分布上与 Geant4 有强一致性，并在螺旋参数分辨率方面保持物理行为。
• Uniformity 损失和关系推理模块有助于缓解模式坍缩，并更好地表示双峰的 occupancy 分布。
• 应用于 Belle II PXD（7.5M 通道）显示出通过实现在线生成替代存储背景叠加而带来的显著存储节省。
• 该方法对 HL-LHC 规模的探测器仿真及其他细粒度密度估计任务具有潜在适用性。

(b) The Relational Reasoning Module for the generator (left) and the discriminator (right)