Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

The Pareto Frontier of Resilient Jet Tagging

Rikab Gambhir, Matt LeBlanc, Yuanchen Zhou

Abstract

Classifying hadronic jets using their constituents' kinematic information is a critical task in modern high-energy collider physics. Often, classifiers are designed by targeting the best performance using metrics such as accuracy, AUC, or rejection rates. However, the use of a single metric can lead to the use of architectures that are more model-dependent than competitive alternatives, leading to potential uncertainty and bias in analysis. We explore such trade-offs and demonstrate the consequences of using networks with high performance metrics but low resilience.

The Pareto Frontier of Resilient Jet Tagging

Abstract

Classifying hadronic jets using their constituents' kinematic information is a critical task in modern high-energy collider physics. Often, classifiers are designed by targeting the best performance using metrics such as accuracy, AUC, or rejection rates. However, the use of a single metric can lead to the use of architectures that are more model-dependent than competitive alternatives, leading to potential uncertainty and bias in analysis. We explore such trade-offs and demonstrate the consequences of using networks with high performance metrics but low resilience.

Paper Structure

This paper contains 9 sections, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Monte Carlo simulated event samples
  4. Model architectures surveyed
  5. Results
  6. Pareto Frontier
  7. Knowledge Distillation
  8. Case Study: determining q/g fractions
  9. Concluding remarks

Figures (2)

  • Figure 1: The Pareto frontier for (a) q/g tagging, (b) top tagging tasks. The AUC of models trained with Pythia samples is plotted vs. the resilience, defined as the percent difference in AUC evaluated on the Pythia and Herwig test samples. The various markers denote different classifiers: DNN (triangles), EFN (circles), PFN (pluses), ParT (stars), Angularities (crosses) and Multiplicities (diamonds). The shaded grey region is Pareto-excluded.
  • Figure 2: (a) Results of training two student DNNs via distillation from a teacher PFN. (b) Summary of distillation training from a teacher PFN to all DNNs and EFNs in the study.