ICCV 2021 accepted papers on Adversarial robustness and security
57 of the 1,612 papers accepted at ICCV 2021 (3.5%) are on Adversarial robustness and security, compared with 2.8% at ICCV 2019. They fall into 4 topics.
See ICCV 2021 in the interactive map
Submitting to ICCV? Checklist →
Working on something in this area? Describe your idea in scime atlas to see the closest papers by meaning and how crowded the spot has become.
Topics and papers
Certified and empirical robustness · 19 papers
- Learnable Boundary Guided Adversarial Training (141 citations)
- Amplitude-Phase Recombination: Rethinking Robustness of Convolutional Neural Networks in Frequency Domain (136 citations)
- Revisiting Adversarial Robustness Distillation: Robust Soft Labels Make Student Better (128 citations)
- AdvRush: Searching for Adversarially Robust Neural Architectures (54 citations)
- Adversarial Attacks Are Reversible With Natural Supervision (43 citations)
- Batch Normalization Increases Adversarial Vulnerability and Decreases Adversarial Transferability: A Non-Robust Feature Perspective (42 citations)
- Low Curvature Activations Reduce Overfitting in Adversarial Training (39 citations)
- Removing Adversarial Noise in Class Activation Feature Space (38 citations)
- Improving Robustness of Facial Landmark Detection by Defending Against Adversarial Attacks (37 citations)
- Relating Adversarially Robust Generalization to Flat Minima (37 citations)
- Robustness via Cross-Domain Ensembles (32 citations)
- Robustness and Generalization via Generative Adversarial Training (29 citations)
and 7 more in the interactive map.
Attacks on text and graphs · 17 papers
- Naturalistic Physical Adversarial Patch for Object Detectors (207 citations)
- AdvDrop: Adversarial Attack to DNNs by Dropping Information (141 citations)
- Towards Face Encryption by Generating Adversarial Identity Masks (129 citations)
- Knowledge-Enriched Distributional Model Inversion Attacks (114 citations)
- Exploiting Explanations for Model Inversion Attacks (109 citations)
- HIRE-SNN: Harnessing the Inherent Robustness of Energy-Efficient Deep Spiking Neural Networks by Training With Crafted Input Noise (86 citations)
- Adversarial Example Detection Using Latent Neighborhood Graph (66 citations)
- Robust Watermarking for Deep Neural Networks via Bi-Level Optimization (65 citations)
- Improving Robustness Against Common Corruptions With Frequency Biased Models (51 citations)
- Multi-Expert Adversarial Attack Detection in Person Re-Identification Using Context Inconsistency (43 citations)
- Dynamic Divide-and-Conquer Adversarial Training for Robust Semantic Segmentation (38 citations)
- TkML-AP: Adversarial Attacks to Top-k Multi-Label Learning (38 citations)
and 5 more in the interactive map.
Black-box and transfer attacks · 14 papers
- Feature Importance-Aware Transferable Adversarial Attacks (333 citations)
- Admix: Enhancing the Transferability of Adversarial Attacks (325 citations)
- On the Robustness of Vision Transformers to Adversarial Examples (219 citations)
- Meta Gradient Adversarial Attack (101 citations)
- On Generating Transferable Targeted Perturbations (92 citations)
- Parallel Rectangle Flip Attack: A Query-Based Black-Box Attack Against Object Detection (92 citations)
- Data-Free Universal Adversarial Perturbation and Black-Box Attack (75 citations)
- Augmented Lagrangian Adversarial Attacks (38 citations)
- Adversarial Attack on Deep Cross-Modal Hamming Retrieval (34 citations)
- Defending Against Universal Adversarial Patches by Clipping Feature Norms (34 citations)
- Meta-Attack: Class-Agnostic and Model-Agnostic Physical Adversarial Attack (30 citations)
- Aha! Adaptive History-Driven Attack for Decision-Based Black-Box Models (24 citations)
and 2 more in the interactive map.
Backdoor and poisoning · 7 papers
- Invisible Backdoor Attack With Sample-Specific Triggers (516 citations)
- LIRA: Learnable, Imperceptible and Robust Backdoor Attacks (233 citations)
- Rethinking the Backdoor Attacks' Triggers: A Frequency Perspective (207 citations)
- Black-Box Detection of Backdoor Attacks With Limited Information and Data (100 citations)
- ProFlip: Targeted Trojan Attack With Progressive Bit Flips (76 citations)
- Detection and Continual Learning of Novel Face Presentation Attacks (53 citations)
- CLEAR: Clean-Up Sample-Targeted Backdoor in Neural Networks (15 citations)
