[论文解读] Novel deep learning methods for track reconstruction

Steven Farrell, Paolo Calafiura+12 more|arXiv (Cornell University)|Oct 14, 2018

Medical Imaging Techniques and Applications被引用 44

一句话总结

本文提出两种基于深度学习的方法，用于 HL-LHC 空间点数据的轨道重建：基于 RNN 的轨迹构建和基于 GNN 的命中/段分类，在 ACTS 模拟数据上显示出强大的性能，并突出了相对于基于图像的方法的可扩展性优势。

ABSTRACT

For the past year, the HEP.TrkX project has been investigating machine learning solutions to LHC particle track reconstruction problems. A variety of models were studied that drew inspiration from computer vision applications and operated on an image-like representation of tracking detector data. While these approaches have shown some promise, image-based methods face challenges in scaling up to realistic HL-LHC data due to high dimensionality and sparsity. In contrast, models that can operate on the spacepoint representation of track measurements ("hits") can exploit the structure of the data to solve tasks efficiently. In this paper we will show two sets of new deep learning models for reconstructing tracks using space-point data arranged as sequences or connected graphs. In the first set of models, Recurrent Neural Networks (RNNs) are used to extrapolate, build, and evaluate track candidates akin to Kalman Filter algorithms. Such models can express their own uncertainty when trained with an appropriate likelihood loss function. The second set of models use Graph Neural Networks (GNNs) for the tasks of hit classification and segment classification. These models read a graph of connected hits and compute features on the nodes and edges. They adaptively learn which hit connections are important and which are spurious. The models are scaleable with simple architecture and relatively few parameters. Results for all models will be presented on ACTS generic detector simulated data.

研究动机与目标

• 在 HL-LHC 的高占用下，激励使用深度学习用于轨迹重建，因为传统的组合方法在高占用时表现不佳。
• 提出基于 RNN 的轨迹构建，用以外推并评估类似卡尔曼滤波操作的轨迹候选。
• 提出基于图神经网络的命中和段分类，在空间点图上识别真实连接和轨迹。
• 在 ACTS 模拟的探测器数据上评估所提模型，以评估可扩展性和准确性。

提出的方法

• 使用带有 LSTM 的 RNN 对空间点（命中）序列进行外推以构建轨道，包括产生预测不确定性的高斯输出变体。
• 用均方误差或高斯对数似然损失训练序列命中预测模型，以实现概率性预测。
• 构建并评估相邻探测层上命中点的图表示，应用基于 EdgeNetwork/NodeNetwork 的 GNN 对命中点和段进行分类。
• 实现两项 GNN 任务：(i) 二分类命中分类以识别轨迹命中；(ii) 二分类段分类以区分真实命中对。
• 报告在构造图上的 GNN 分类器的纯度、效率和准确率等性能指标。

实验结果

研究问题

• RQ1 基于 RNN 的模型是否能够有效地外推粒子轨迹并从类似卡尔曼滤波过程的空间点序列中构建轨迹？
• RQ2 图神经网络是否能够在空间点数据图中准确分类命中和段，以恢复轨迹候选？
• RQ3 在类似 HL-LHC 的数据上，RNN 和 GNN 方法在可扩展性和性能方面的比较如何？
• RQ4 将这些方法整合到完整的组合轨迹寻踪流程中的局限性和下一步是什么？

主要发现

• 基于 RNN 的轨迹构建在低占用设置下实现了高命中预测准确性（简单模型 99.93%，高斯模型 99.98%）。
• 高斯 RNN 通过预测协方差提供不确定性估计，拉线分布与预测基本一致。
• GNN 命中分类在部分标注图上达到 99.2% 纯度、97.9% 效率和 99.4% 总体准确率。
• GNN 段分类在构建的段上达到 99.5% 纯度、98.7% 效率和 99.5% 总体准确率。
• GNN 方法在结构化命中数据上显示出强大的可扩展轨迹重建潜力，被认为是在 HL-LHC 条件下最有前景的方法。