Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Lightweight Deep Learning Framework for Accurate Particle Flow Energy Reconstruction

Yu Wang, Yangguang Zhang, Shengxiang Lin, Xingyi Zhang, Han Zhang

TL;DR

This work tackles energy reconstruction in high-density particle collisions where traditional particle flow methods struggle. It proposes a lightweight yet powerful deep learning framework that blends a multimodal loss (WMSE+SSIM) with a CNN backbone augmented by SE channel attention, 3D convolution-based feature extraction, and offset self-attention, enabling robust interpolation and extrapolation on multi-channel detector images. Empirical results show a 90K-parameter SEBlockCNN can match much larger baselines, while a 25M-parameter 3DCNN achieves state-of-the-art performance on both interpolation and extrapolation tasks; ablation and module-exploration experiments guide practical trade-offs. The approach is open-sourced, promotes reproducibility, and offers actionable guidance for deployment in particle-flow energy reconstruction pipelines with improved accuracy and efficiency.

Abstract

Under extreme operating conditions, characterized by high particle multiplicity and heavily overlapping shower energy deposits, classical particle flow algorithms encounter pronounced limitations in resolution, efficiency, and accuracy. To address this challenge, this paper proposes and systematically evaluates a deep learning reconstruction framework: For multichannel sparse features, we design a hybrid loss function combining weighted mean squared error with structural similarity index, effectively balancing pixel-level accuracy and structural fidelity. By integrating 3D convolutions, Squeeze-and-Excitation channel attention, and Offset self-attention modules into baseline convolutional neural networks, we enhance the model's capability to capture cross-modal spatiotemporal correlations and energy-displacement nonlinearities. Validated on custom-constructed simulation data and Pythia jet datasets, the framework's 90K-parameter lightweight variant approaches the performance of 5M-parameter baselines, while the 25M-parameter 3D model achieves state-of-the-art results in both interpolation and extrapolation tasks. Comprehensive experiments quantitatively evaluate component contributions and provide performance-parameter trade-off guidelines. All core code and data processing scripts are open-sourced on a GitHub repository to facilitate community reproducibility and extension.

Lightweight Deep Learning Framework for Accurate Particle Flow Energy Reconstruction

TL;DR

This work tackles energy reconstruction in high-density particle collisions where traditional particle flow methods struggle. It proposes a lightweight yet powerful deep learning framework that blends a multimodal loss (WMSE+SSIM) with a CNN backbone augmented by SE channel attention, 3D convolution-based feature extraction, and offset self-attention, enabling robust interpolation and extrapolation on multi-channel detector images. Empirical results show a 90K-parameter SEBlockCNN can match much larger baselines, while a 25M-parameter 3DCNN achieves state-of-the-art performance on both interpolation and extrapolation tasks; ablation and module-exploration experiments guide practical trade-offs. The approach is open-sourced, promotes reproducibility, and offers actionable guidance for deployment in particle-flow energy reconstruction pipelines with improved accuracy and efficiency.

Abstract

Under extreme operating conditions, characterized by high particle multiplicity and heavily overlapping shower energy deposits, classical particle flow algorithms encounter pronounced limitations in resolution, efficiency, and accuracy. To address this challenge, this paper proposes and systematically evaluates a deep learning reconstruction framework: For multichannel sparse features, we design a hybrid loss function combining weighted mean squared error with structural similarity index, effectively balancing pixel-level accuracy and structural fidelity. By integrating 3D convolutions, Squeeze-and-Excitation channel attention, and Offset self-attention modules into baseline convolutional neural networks, we enhance the model's capability to capture cross-modal spatiotemporal correlations and energy-displacement nonlinearities. Validated on custom-constructed simulation data and Pythia jet datasets, the framework's 90K-parameter lightweight variant approaches the performance of 5M-parameter baselines, while the 25M-parameter 3D model achieves state-of-the-art results in both interpolation and extrapolation tasks. Comprehensive experiments quantitatively evaluate component contributions and provide performance-parameter trade-off guidelines. All core code and data processing scripts are open-sourced on a GitHub repository to facilitate community reproducibility and extension.

Paper Structure

This paper contains 22 sections, 6 equations, 8 figures.

Table of Contents

  1. Introduction
  2. Simulation experiments and datasets
  3. Design of Deep Neural Networks
  4. Design of Deep Neural Networks
  5. CNN
  6. Channel Attention Module
  7. 3D Convolution-based Feature Extraction Module
  8. Offset Self-Attention Module
  9. Loss Function Design
  10. Experimental Preparation
  11. Training Strategies
  12. Model Evaluation Indicators
  13. General Energy Reconstruction Task Evaluation Indicators
  14. Jet Energy Reconstruction Assessment Metrics and Process
  15. Experiment
  16. ...and 7 more sections

Figures (8)

  • Figure 1: Simulation data display, from left to right, they are respectively Emcal, Hcal, Tracker_n, Tracker_p, and Truth Image. The darkness of the pixel grid color represents the detected energy at the corresponding position.
  • Figure 2:
  • Figure 3:
  • Figure 4:
  • Figure 5:
  • ...and 3 more figures