Towards Effective Multi-Label Recognition Attacks via Knowledge Graph Consistency
Not on the map. Why?
Abstract (arXiv)
Many real-world applications of image recognition require multi-label learning, whose goal is to find all labels in an image. Thus, robustness of such systems to adversarial image perturbations is extremely important. However, despite a large body of recent research on adversarial attacks, the scope of the existing works is mainly limited to the multi-class setting, where each image contains a single label. We show that the naive extensions of multi-class attacks to the multi-label setting lead to violating label relationships, modeled by a knowledge graph, and can be detected using a consistency verification scheme. Therefore, we propose a graph-consistent multi-label attack framework, which searches for small image perturbations that lead to misclassifying a desired target set while respecting label hierarchies. By extensive experiments on two datasets and using several multi-label recognition models, we show that our method generates extremely successful attacks that, unlike naive multi-label perturbations, can produce model predictions consistent with the knowledge graph.
Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 33 of its 44 references that are in the map. First posted on arXiv 11 Jul 2022.
Builds on
33 of 44 references.
References · 33
- 2014Adam: A Method for Stochastic Optimizationcited by 37,240
- 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
- 2016Adversarial examples in the physical worldcited by 1,769
- 2020Reliable evaluation of adversarial robustness with an ensemble of diverse parameter-free attackscited by 1,451
- 2019Natural Adversarial Examplescited by 1,408
- 2018Obfuscated Gradients Give a False Sense of Security: Circumventing Defenses to Adversarial Examplescited by 1,268
- 2016ConceptNet 5.5: An Open Multilingual Graph of General Knowledgecited by 1,067
- 2016Adversarial Machine Learning at Scalecited by 936
- 2015The Limitations of Deep Learning in Adversarial Settingscited by 790
- 2020Fast is better than free: Revisiting adversarial trainingcited by 727
21 more
- 2017Evasion Attacks against Machine Learning at Test Timecited by 689
- 2017Adversarial Examples Are Not Easily Detected: Bypassing Ten Detection Methodscited by 546
- 2019Evading Defenses to Transferable Adversarial Examples by Translation-Invariant Attackscited by 442
- 2018Adversarial Attacks on Neural Networks for Graph Datacited by 397
- 2020Asymmetric Loss For Multi-Label Classificationcited by 381
- 2019Multi-Label Image Recognition with Graph Convolutional Networkscited by 322
- 2019Fooling automated surveillance cameras: adversarial patches to attack person detectioncited by 268
- 2018Physical Adversarial Examples for Object Detectorscited by 163
- 2020TResNet: High Performance GPU-Dedicated Architecturecited by 119
- 2017Universal Adversarial Perturbations Against Semantic Image Segmentationcited by 92
- 2020Attention-Driven Dynamic Graph Convolutional Network for Multi-Label Image Recognitioncited by 80
- 2019An Alternative Surrogate Loss for PGD-based Adversarial Testingcited by 45
- 2015Exploit Bounding Box Annotations for Multi-label Object Recognitioncited by 39
- 2018Universal Perturbation Attack Against Image Retrievalcited by 32
- 2018Distributionally Adversarial Attackcited by 26
- 2021LAFEAT: Piercing Through Adversarial Defenses with Latent Featurescited by 24
- 2019Multi-Label Adversarial Perturbationscited by 18
- 2021Multi-Label Image Classification with Contrastive Learningcited by 18
- 2019Collaboration based Multi-Label Learningcited by 11
- 2020Characterizing the Evasion Attackability of Multi-label Classifierscited by 5
- 2021Attack Transferability Characterization for Adversarially Robust Multi-label Classificationcited by 2
Built on by
Not cited yet.