Improving Robustness of Jet Tagging Algorithms with Adversarial Training (2022-03-25T00:00:00.000000Z)

TL;DR

This work investigates the classifier response to input data with injected mismodelings and probes the vulnerability of flavor tagging algorithms via application of adversarial attacks and presents an adversarial training strategy that mitigates the impact of such simulated attacks and improves the classifiers robustness.

Abstract

Deep learning is a standard tool in the field of high-energy physics, facilitating considerable sensitivity enhancements for numerous analysis strategies. In particular, in identification of physics objects, such as jet flavor tagging, complex neural network architectures play a major role. However, these methods are reliant on accurate simulations. Mismodeling can lead to non-negligible differences in performance in data that need to be measured and calibrated against. We investigate the classifier response to input data with injected mismodelings and probe the vulnerability of flavor tagging algorithms via application of adversarial attacks. Subsequently, we present an adversarial training strategy that mitigates the impact of such simulated attacks and improves the classifier robustness. We examine the relationship between performance and vulnerability and show that this method constitutes a promising approach to reduce the vulnerability to poor modeling.

Authors

Annika Stein

1 papers

X. Coubez

1 papers

S. Mondal

1 papers

TL;DR

Abstract

Authors

References80 items

Improving robustness of jet tagging algorithms with adversarial training: exploring the loss surface

Direct constraint on the Higgs–charm coupling from a search for Higgs boson decays into charm quarks with the ATLAS detector

A new calibration method for charm jet identification validated with proton-proton collision events at √s = 13 TeV

Adversarial methods to reduce simulation bias in neutrino interaction event filtering at liquid argon time projection chambers

DeepAdversaries: examining the robustness of deep learning models for galaxy morphology classification

Measurement of the $c$-jet mistagging efficiency in $t\bar{t}$ events using $pp$ collision data at $\sqrt{s}=13$ TeV collected with the ATLAS detector

Uncertainty-aware machine learning for high energy physics

Feature Engineering and Selection: A Practical Approach for Predictive Models

Deep Learning With PyTorch

A deep neural network to search for new long-lived particles decaying to jets

A search for the standard model Higgs boson decaying to charm quarks

PyTorch: An Imperative Style, High-Performance Deep Learning Library

ATLAS b-jet identification performance and efficiency measurement with $$t{\bar{t}}$$ events in pp collisions at $$\sqrt{s}=13$$ TeV

AI Safety for High Energy Physics

Deep Ensembles: A Loss Landscape Perspective

Feature Engineering and Selection

ATLAS $b$-jet identification performance and efficiency measurement with $t\bar{t}$ events in $pp$ collisions at $\sqrt{s}=13$ TeV

Disentangling Adversarial Robustness and Generalization

Adversarial Attacks and Defences: A Survey

Peeking Inside the Black-Box: A Survey on Explainable Artificial Intelligence (XAI)

Observation of Higgs Boson Decay to Bottom Quarks.

Machine Learning in High Energy Physics Community White Paper

Jet substructure at the Large Hadron Collider

Visualizing the Loss Landscape of Neural Nets

Identification of heavy-flavour jets with the CMS detector in pp collisions at 13 TeV

One Pixel Attack for Fooling Deep Neural Networks

Focal Loss for Dense Object Detection

Towards Deep Learning Models Resistant to Adversarial Attacks

Adversarial Machine Learning at Scale

Learning to Pivot with Adversarial Networks

Robustness of classifiers: from adversarial to random noise

A Survey of Predictive Modeling on Imbalanced Domains

Jet flavor classification in high-energy physics with deep neural networks

Concrete Problems in AI Safety

The Gaussian Distribution

Analysis of classifiers’ robustness to adversarial perturbations

Adam: A Method for Stochastic Optimization

Explaining and Harnessing Adversarial Examples

Unsupervised Domain Adaptation by Backpropagation

Intriguing properties of neural networks

DELPHES 3: a modular framework for fast simulation of a generic collider experiment

FastJet user manual

MadGraph 5: going beyond

RAVE—A Detector-Independent Toolkit to Reconstruct Vertices

The ATLAS Experiment at the CERN Large Hadron Collider

The CMS experiment at the CERN LHC

The anti-$k_t$ jet clustering algorithm

Matplotlib: A 2D Graphics Environment

The relationship between Precision-Recall and ROC curves

PYTHIA 6.4 Physics and Manual

Direct search for the Standard Model Higgs boson

Learning rate schedules for faster stochastic gradient search

Multilayer feedforward networks are universal approximators

Search for Higgs boson decay to a charm quark-antiquark pair in proton-proton collisions at √ s = 13 TeV

AdversarialTraining-for-Jet-Tagging”, GitHub repository (2022)

“CrossEntropyLoss”

Adeelh/pytorch-multi-class-focal-loss: 1.1

advjets-mlhep2020. GitHub repository. https:// github. com/ cshim min/ advje ts- mlhep 2020

“scikit-hep/awkward-1.0: 0.4.5”

and s

University of Birmingham Observation of H bb decays and VH production with the ATLAS detector ATLAS Collaboration

Observation of H ->bb¯ decays and V H production with the ATLAS detector

Observation of H ->bb¯ decays and V H production with the ATLAS detector

Deep Learning

Identification of c-quark jets at the CMS experiment The CMS Collaboration

Understanding Adversarial Training: Increasing Local Stability of Neural Nets through Robust Optimization

Deep Learning

Dropout: a simple way to prevent neural networks from overfitting

I and J

A Review on Ensembles for the Class Imbalance Problem: Bagging-, Boosting-, and Hybrid-Based Approaches

Impact parameter-based btagging algorithms in the 7 TeV collision data with the ATLAS detector: the TrackCounting and JetProb algorithms

Evaluation: From Precision, Recall and F-Factor to ROC, Informedness, Markedness & Correlation

SIE-07-001 December 2007 Evaluation : From Precision , Recall and F-Factor to ROC ,

Influence of cultivation temperature on the ligninolytic activity of selected fungal strains

Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations

delphes-rave