Uncovering Hidden Systematics in Neural Network Models for High Energy Physics

2026-05-08

AI Summary

This work addresses the heightened sensitivity of neural networks to minute input perturbations in high-energy physics, where conventional systematic uncertainty estimates often underestimate model vulnerability. Inspired by adversarial attacks, the authors propose the first framework capable of quantitatively assessing neural network sensitivity to realistic experimental perturbations. By applying multidimensional, physically motivated perturbations—consistent with experimental uncertainties—while preserving the underlying input distribution, the method probes significant shifts in model outputs. This approach transcends the limitations of traditional error propagation, revealing that models can be substantially misled even within permissible experimental error margins. The framework further offers a practical pathway for evaluating and mitigating systematic uncertainties in typical high-energy physics tasks.

Abstract

Neural networks (NNs) are inherently multidimensional classifiers that learn complex, non-linear relationships among input observables. While their flexibility enables unprecedented performance in high-energy physics (HEP) analyses, it also makes them sensitive to small variations in their inputs. Consequently, the propagation and estimation of systematic uncertainties in NN-based models remain an open challenge. There are indications that uncertainties derived in control regions or from nominal variations of input features can underestimate the true model uncertainty, potentially leaving biases unaccounted for. Inspired by insights from adversarial-attack studies in machine learning, we explore how subtle perturbations, fully consistent with the experimental uncertainties on the input observables, can lead to substantial changes in NN outputs, while keeping the one-dimensional and correlated input distributions nearly unchanged. Using a set of representative HEP tasks, including event classification and object identification, and testing across a variety of network architectures, we demonstrate that networks can be systematically "fooled" at significant rates within the allowed uncertainty envelopes. Building on this observation, we introduce a quantitative framework to probe and measure the hidden sensitivity of neural networks to realistic experimental variations, providing a practical path to evaluate and control their systematic uncertainty in physics analyses.

Problem

• systematic uncertainties
• neural networks
• high-energy physics
• input perturbations
• model sensitivity

Innovation

• systematic uncertainties
• neural networks
• adversarial perturbations
• high-energy physics
• model robustness