Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Deep Image Reconstruction for Background Subtraction in Heavy-Ion Collisions

Umar Sohail Qureshi, Raghav Kunnawalkam Elayavalli

TL;DR

This work tackles the challenge of large, fluctuating backgrounds in heavy-ion jet reconstruction and the consequent loss of jet substructure information. It introduces DeepSub, a Swin Transformer–based full-event background subtraction framework that outputs a cleaned signal image from a noisy input event image. DeepSub achieves sub-percent to percent accuracy on jet observables such as $p_T$, mass, girth, and ECF, outperforming the conventional ICS method and showing robustness to generator shifts, with faster inference. The approach provides a path toward precision heavy-ion jet measurements and includes public data and code to facilitate broader adoption and future generalization to other processes.

Abstract

Jet reconstruction in an ultra-relativistic heavy-ion collision suffers from a notoriously large, fluctuating thermal background. Traditional background subtraction methods struggle to remove this soft background while preserving the jet's hard substructure. In this Letter, we present DeepSub, the first machine learning-based approach for full-event background subtraction. DeepSub utilizes a model based on Swin Transformer layers to denoise jet images and disentangle hard jets from the heavy-ion background. DeepSub significantly outperforms existing subtraction techniques by reproducing key jet observables such as jet $p_\mathrm{T}$ and mass, and substructure observables such as girth and the energy correlation function, at the sub-percent to percent level. As such, DeepSub paves the way for precision heavy-ion measurements in hitherto inaccessible kinematic regimes.

Deep Image Reconstruction for Background Subtraction in Heavy-Ion Collisions

TL;DR

This work tackles the challenge of large, fluctuating backgrounds in heavy-ion jet reconstruction and the consequent loss of jet substructure information. It introduces DeepSub, a Swin Transformer–based full-event background subtraction framework that outputs a cleaned signal image from a noisy input event image. DeepSub achieves sub-percent to percent accuracy on jet observables such as , mass, girth, and ECF, outperforming the conventional ICS method and showing robustness to generator shifts, with faster inference. The approach provides a path toward precision heavy-ion jet measurements and includes public data and code to facilitate broader adoption and future generalization to other processes.

Abstract

Jet reconstruction in an ultra-relativistic heavy-ion collision suffers from a notoriously large, fluctuating thermal background. Traditional background subtraction methods struggle to remove this soft background while preserving the jet's hard substructure. In this Letter, we present DeepSub, the first machine learning-based approach for full-event background subtraction. DeepSub utilizes a model based on Swin Transformer layers to denoise jet images and disentangle hard jets from the heavy-ion background. DeepSub significantly outperforms existing subtraction techniques by reproducing key jet observables such as jet and mass, and substructure observables such as girth and the energy correlation function, at the sub-percent to percent level. As such, DeepSub paves the way for precision heavy-ion measurements in hitherto inaccessible kinematic regimes.

Paper Structure

This paper contains 6 sections, 2 equations, 3 figures.

Table of Contents

  1. Introduction
  2. Methodology
  3. Samples and Simulation
  4. Machine Learning Architecture
  5. Results
  6. Discussion and Future Work

Figures (3)

  • Figure 1: The DeepSub architecture based on the Swin Transformer for image reconstruction.
  • Figure 2: Distributions of jet $p_\mathrm{T}$ (top-left), mass (top-right), girth (bottom-left), and the energy correlation function (bottom-right) for DeepSub (red-dotted), event-wide ICS (green-dashed), and the ground truth (blue-filled) on Jewel dijet events. The error bars represent the statistical uncertainty in each bin. The bottom panels in each plot show the ratio of the reconstructed value of each observable to their truth values and the yellow-shaded region indicates the statistical uncertainty in the truth.
  • Figure 3: Distributions of jet girth (left), and the energy correlation function (right) for DeepSub (red-dotted), event-wide ICS (green-dashed), and the ground truth (blue-filled) on CoLBT-Hydro dijet events. The error bars represent the statistical uncertainty in each bin. The bottom panels in each plot show the ratio of the reconstructed value of each observable to their truth values and the yellow-shaded region indicates the statistical uncertainty in the truth.