Characterizing Natural Adversarial Examples Through Activation Map Analysis

Adversarial examples are an intriguing and critical topic in the field of machine learning. The impact of malignant perturbations on deep learning-based systems, especially in safety-critical applications, highlights a significant security concern. While most research has focused on artificially gen...

Descripción completa

Detalles Bibliográficos
Autores: Pedraza Dorado, Aníbal, Leon , Nerea, Singh, Harbinder, Déniz Suárez, Óscar, Bueno García, María Gloria
Tipo de recurso: artículo
Fecha de publicación:2025
País:España
Institución:Fundación Dialnet. Universidad de La Rioja
Repositorio:RUIdeRA. Repositorio Institucional de la UCLM
OAI Identifier:oai:ruidera.uclm.es:10578/45786
Acceso en línea:https://ietresearch.onlinelibrary.wiley.com/doi/full/10.1049/ipr2.70123
https://hdl.handle.net/10578/45786
Access Level:acceso abierto
Palabra clave:activation map
adversarial examples
deep learning
natural adversarial images
neural networks
id ES_135a8bdcbd027ea994760ca4e0bd02bb
oai_identifier_str oai:ruidera.uclm.es:10578/45786
network_acronym_str ES
network_name_str España
repository_id_str
spelling Characterizing Natural Adversarial Examples Through Activation Map AnalysisPedraza Dorado, AníbalLeon , NereaSingh, HarbinderDéniz Suárez, ÓscarBueno García, María Gloriaactivation mapadversarial examplesdeep learningnatural adversarial imagesneural networksAdversarial examples are an intriguing and critical topic in the field of machine learning. The impact of malignant perturbations on deep learning-based systems, especially in safety-critical applications, highlights a significant security concern. While most research has focused on artificially generated adversarial attacks–crafted through optimization algorithms and constrained perturbations, it is important to note that adversarial examples can also occur naturally, without any artificial manipulation, during the prediction of real-world images. These naturally occurring adversarial examples pose unique challenges, as they are harder to detect and interpret. Despite their importance, the study of natural adversarial examples remains in its early stages. Fundamental questions remain unanswered: Do natural adversarial examples exhibit similar behaviours or properties as artificially generated ones? How should models be adapted to improve their robustness against such natural inputs? To address these questions, this work proposes an in-depth analysis of activation maps to compare the internal behaviour of neural networks when processing clean images, artificially perturbed inputs and natural adversarial examples. A set of quantitative metrics is extracted from activation heatmaps at various network layers, including mean activation intensity, centroid displacement and standard reference image quality metrics. These measurements enable a systematic comparison of how the network attends to different image regions under varying conditions. The experimental results demonstrate that natural adversarial examples exhibit statistically significant differences in activation patterns compared to their artificial counterparts, suggesting that they may require distinct strategies for detection and defence.Wiley202520252025info:eu-repo/semantics/articleapplication/pdfapplication/pdfhttps://ietresearch.onlinelibrary.wiley.com/doi/full/10.1049/ipr2.70123https://hdl.handle.net/10578/45786reponame:RUIdeRA. Repositorio Institucional de la UCLMinstname:Fundación Dialnet. Universidad de La RiojaInglésinfo:eu-repo/semantics/openAccessoai:ruidera.uclm.es:10578/457862026-05-27T07:36:41Z
dc.title.none.fl_str_mv Characterizing Natural Adversarial Examples Through Activation Map Analysis
title Characterizing Natural Adversarial Examples Through Activation Map Analysis
spellingShingle Characterizing Natural Adversarial Examples Through Activation Map Analysis
Pedraza Dorado, Aníbal
activation map
adversarial examples
deep learning
natural adversarial images
neural networks
title_short Characterizing Natural Adversarial Examples Through Activation Map Analysis
title_full Characterizing Natural Adversarial Examples Through Activation Map Analysis
title_fullStr Characterizing Natural Adversarial Examples Through Activation Map Analysis
title_full_unstemmed Characterizing Natural Adversarial Examples Through Activation Map Analysis
title_sort Characterizing Natural Adversarial Examples Through Activation Map Analysis
dc.creator.none.fl_str_mv Pedraza Dorado, Aníbal
Leon , Nerea
Singh, Harbinder
Déniz Suárez, Óscar
Bueno García, María Gloria
author Pedraza Dorado, Aníbal
author_facet Pedraza Dorado, Aníbal
Leon , Nerea
Singh, Harbinder
Déniz Suárez, Óscar
Bueno García, María Gloria
author_role author
author2 Leon , Nerea
Singh, Harbinder
Déniz Suárez, Óscar
Bueno García, María Gloria
author2_role author
author
author
author
dc.subject.none.fl_str_mv activation map
adversarial examples
deep learning
natural adversarial images
neural networks
topic activation map
adversarial examples
deep learning
natural adversarial images
neural networks
description Adversarial examples are an intriguing and critical topic in the field of machine learning. The impact of malignant perturbations on deep learning-based systems, especially in safety-critical applications, highlights a significant security concern. While most research has focused on artificially generated adversarial attacks–crafted through optimization algorithms and constrained perturbations, it is important to note that adversarial examples can also occur naturally, without any artificial manipulation, during the prediction of real-world images. These naturally occurring adversarial examples pose unique challenges, as they are harder to detect and interpret. Despite their importance, the study of natural adversarial examples remains in its early stages. Fundamental questions remain unanswered: Do natural adversarial examples exhibit similar behaviours or properties as artificially generated ones? How should models be adapted to improve their robustness against such natural inputs? To address these questions, this work proposes an in-depth analysis of activation maps to compare the internal behaviour of neural networks when processing clean images, artificially perturbed inputs and natural adversarial examples. A set of quantitative metrics is extracted from activation heatmaps at various network layers, including mean activation intensity, centroid displacement and standard reference image quality metrics. These measurements enable a systematic comparison of how the network attends to different image regions under varying conditions. The experimental results demonstrate that natural adversarial examples exhibit statistically significant differences in activation patterns compared to their artificial counterparts, suggesting that they may require distinct strategies for detection and defence.
publishDate 2025
dc.date.none.fl_str_mv 2025
2025
2025
dc.type.none.fl_str_mv info:eu-repo/semantics/article
format article
dc.identifier.none.fl_str_mv https://ietresearch.onlinelibrary.wiley.com/doi/full/10.1049/ipr2.70123
https://hdl.handle.net/10578/45786
url https://ietresearch.onlinelibrary.wiley.com/doi/full/10.1049/ipr2.70123
https://hdl.handle.net/10578/45786
dc.language.none.fl_str_mv Inglés
language_invalid_str_mv Inglés
dc.rights.none.fl_str_mv info:eu-repo/semantics/openAccess
eu_rights_str_mv openAccess
dc.format.none.fl_str_mv application/pdf
application/pdf
dc.publisher.none.fl_str_mv Wiley
publisher.none.fl_str_mv Wiley
dc.source.none.fl_str_mv reponame:RUIdeRA. Repositorio Institucional de la UCLM
instname:Fundación Dialnet. Universidad de La Rioja
instname_str Fundación Dialnet. Universidad de La Rioja
reponame_str RUIdeRA. Repositorio Institucional de la UCLM
collection RUIdeRA. Repositorio Institucional de la UCLM
repository.name.fl_str_mv
repository.mail.fl_str_mv
_version_ 1869403671255580672
score 15,81155

Similares en LA Referencia