Skip to main content

Lund University Publications

LUND UNIVERSITY LIBRARIES

Improving transferability of generated universal adversarial perturbations for image classification and segmentation

Hashemi, A. S. LU ; Bär, Andreas ; Mozaffari, Saeed and Fingscheidt, Tim (2022) p.171-196
Abstract
Although deep neural networks (DNNs) are high-performance methods for various complex tasks, e.g., environment perception in automated vehicles (AVs), they are vulnerable to adversarial perturbations. Recent works have proven the existence of universal adversarial perturbations (UAPs), which, when added to most images, destroy the output of the respective perception function. Existing attack methods often show a low success rate when attacking target models which are different from the one that the attack was optimized on. To address such weak transferability, we propose a novel learning criterion by combining a low-level feature loss, addressing the similarity of feature representations in the first layer of various model architectures,... (More)
Although deep neural networks (DNNs) are high-performance methods for various complex tasks, e.g., environment perception in automated vehicles (AVs), they are vulnerable to adversarial perturbations. Recent works have proven the existence of universal adversarial perturbations (UAPs), which, when added to most images, destroy the output of the respective perception function. Existing attack methods often show a low success rate when attacking target models which are different from the one that the attack was optimized on. To address such weak transferability, we propose a novel learning criterion by combining a low-level feature loss, addressing the similarity of feature representations in the first layer of various model architectures, with a cross-entropy loss. Experimental results on ImageNet and Cityscapes datasets show that our method effectively generates universal adversarial perturbations achieving state-of-the-art fooling rates across different models, tasks, and datasets. Due to their effectiveness, we propose the use of such novel generated UAPs in robustness evaluation of DNN-based environment perception functions for AVs. (Less)
Please use this url to cite or link to this publication:
author
; ; and
publishing date
type
Chapter in Book/Report/Conference proceeding
publication status
published
host publication
Deep Neural Networks and Data for Automated Driving : Robustness, Uncertainty Quantification, and Insights Towards Safety - Robustness, Uncertainty Quantification, and Insights Towards Safety
pages
171 - 196
publisher
Springer Nature
external identifiers
  • scopus:105003334277
ISBN
978-3-031-01232-7
978-3-031-01233-4
language
English
LU publication?
no
id
17291e08-d4c1-4580-8a06-74a85010d1d7
date added to LUP
2025-01-31 14:06:27
date last changed
2025-08-05 12:09:11
@inbook{17291e08-d4c1-4580-8a06-74a85010d1d7,
  abstract     = {{Although deep neural networks (DNNs) are high-performance methods for various complex tasks, e.g., environment perception in automated vehicles (AVs), they are vulnerable to adversarial perturbations. Recent works have proven the existence of universal adversarial perturbations (UAPs), which, when added to most images, destroy the output of the respective perception function. Existing attack methods often show a low success rate when attacking target models which are different from the one that the attack was optimized on. To address such weak transferability, we propose a novel learning criterion by combining a low-level feature loss, addressing the similarity of feature representations in the first layer of various model architectures, with a cross-entropy loss. Experimental results on ImageNet and Cityscapes datasets show that our method effectively generates universal adversarial perturbations achieving state-of-the-art fooling rates across different models, tasks, and datasets. Due to their effectiveness, we propose the use of such novel generated UAPs in robustness evaluation of DNN-based environment perception functions for AVs.}},
  author       = {{Hashemi, A. S. and Bär, Andreas and Mozaffari, Saeed and Fingscheidt, Tim}},
  booktitle    = {{Deep Neural Networks and Data for Automated Driving : Robustness, Uncertainty Quantification, and Insights Towards Safety}},
  isbn         = {{978-3-031-01232-7}},
  language     = {{eng}},
  pages        = {{171--196}},
  publisher    = {{Springer Nature}},
  title        = {{Improving transferability of generated universal adversarial perturbations for image classification and segmentation}},
  year         = {{2022}},
}