Investigating and unmasking feature-level vulnerabilities of CNNs to adversarial perturbations
Not on the map. Why?
Abstract (arXiv)
This study explores the impact of adversarial perturbations on Convolutional Neural Networks (CNNs) with the aim of enhancing the understanding of their underlying mechanisms. Despite numerous defense methods proposed in the literature, there is still an incomplete understanding of this phenomenon. Instead of treating the entire model as vulnerable, we propose that specific feature maps learned during training contribute to the overall vulnerability. To investigate how the hidden representations learned by a CNN affect its vulnerability, we introduce the Adversarial Intervention framework. Experiments were conducted on models trained on three well-known computer vision datasets, subjecting them to attacks of different nature. Our focus centers on the effects that adversarial perturbations to a model's initial layer have on the overall behavior of the model. Empirical results revealed compelling insights: a) perturbing selected channel combinations in shallow layers causes significant disruptions; b) the channel combinations most responsible for the disruptions are common among different types of attacks; c) despite shared vulnerable combinations of channels, different attacks affect hidden representations with varying magnitudes; d) there exists a positive correlation between a kernel's magnitude and its vulnerability. In conclusion, this work introduces a novel framework to study the vulnerability of a CNN model to adversarial perturbations, revealing insights that contribute to a deeper understanding of the phenomenon. The identified properties pave the way for the development of efficient ad-hoc defense mechanisms in future applications.
Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Builds on 29 of its 41 references that are in the map. First posted on arXiv 31 May 2024.
Builds on
29 of 41 references.
References · 29
- 2015Deep Residual Learning for Image Recognitioncited by 52,096
- 2014Explaining and Harnessing Adversarial Examplescited by 7,364
- 2017Towards Deep Learning Models Resistant to Adversarial Attackscited by 6,415
- 2013Intriguing properties of neural networkscited by 5,155
- 2016Towards Evaluating the Robustness of Neural Networkscited by 2,998
- 2020Reliable evaluation of adversarial robustness with an ensemble of diverse parameter-free attackscited by 1,451
- 2021EfficientNetV2: Smaller Models and Faster Trainingcited by 1,135
- 2019Adversarial Examples Are Not Bugs, They Are Featurescited by 953
- 2019HopSkipJumpAttack: A Query-Efficient Decision-Based Attackcited by 333
- 2018Adversarial Robustness Toolbox v1.0.0cited by 215
- 2019Feature relevance quantification in explainable AI: A causal problemcited by 167
- 2019Adversarial Examples Are a Natural Consequence of Test Error in Noisecited by 114
17 more
- 2019When NAS Meets Robustness: In Search of Robust Architectures against Adversarial Attackscited by 87
- 2020Causal Interpretability for Machine Learning -- Problems, Methods and Evaluationcited by 82
- 2019Neural Network Attributions: A Causal Perspectivecited by 77
- 2021Improving Adversarial Robustness via Channel-wise Activation Suppressingcited by 76
- 2021Exploring Architectural Ingredients of Adversarially Robust Deep Neural Networkscited by 74
- 2020Adversarial Machine Learning in Image Classification: A Survey Towards the Defender's Perspectivecited by 59
- 2020Batch Normalization Increases Adversarial Vulnerability and Decreases Adversarial Transferability: A Non-Robust Feature Perspectivecited by 35
- 2019Batch Normalization is a Cause of Adversarial Vulnerabilitycited by 29
- 2021A Review of Adversarial Attack and Defense for Classification Methodscited by 29
- 2021AdvRush: Searching for Adversarially Robust Neural Architecturescited by 28
- 2021CIFS: Improving Adversarial Robustness of CNNs via Channel-wise Importance-based Feature Selectioncited by 21
- 2018Explaining Deep Learning Models using Causal Inferencecited by 18
- 2021Learning Diverse-Structured Networks for Adversarial Robustnesscited by 12
- 2019Towards Understanding Adversarial Examples Systematically: Exploring Data Size, Task and Model Factorscited by 10
- 2021Explaining Adversarial Vulnerability with a Data Sparsity Hypothesiscited by 4
- 2022Layer-wise Regularized Adversarial Training using Layers Sustainability Analysis (LSA) frameworkcited by 2
- 2020On Intrinsic Dataset Properties for Adversarial Machine Learningcited by 1
Built on by
Citing papers per year: 2026: 1.
1 paper.