ICLR 2021 accepted papers on Adversarial robustness and security
43 of the 860 papers accepted at ICLR 2021 (5.0%) are on Adversarial robustness and security, compared with 7.1% at ICLR 2020. They fall into 4 topics.
See ICLR 2021 in the interactive map
Submitting to ICLR? 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 · 22 papers
- Improving Adversarial Robustness via Channel-wise Activation Suppressing (57 citations)
- Learning perturbation sets for robust machine learning (41 citations)
- Geometry-aware Instance-reweighted Adversarial Training (31 citations)
- Deep Partition Aggregation: Provable Defenses against General Poisoning Attacks (20 citations)
- How Does Mixup Help With Robustness and Generalization? (20 citations)
- Perceptual Adversarial Robustness: Defense Against Unseen Threat Models (19 citations)
- On Fast Adversarial Robustness Adaptation in Model-Agnostic Meta-Learning (17 citations)
- Evaluations and Methods for Explanation through Robustness Analysis (15 citations)
- How Benign is Benign Overfitting ? (13 citations)
- Effective and Efficient Vote Attack on Capsule Networks (12 citations)
- Bag of Tricks for Adversarial Training (10 citations)
- Orthogonalizing Convolutional Layers with the Cayley Transform (10 citations)
and 10 more in the interactive map.
Attacks on text and graphs · 13 papers
- LowKey: Leveraging Adversarial Attacks to Protect Social Media Users from Facial Recognition (66 citations)
- InfoBERT: Improving Robustness of Language Models from An Information Theoretic Perspective (61 citations)
- Towards Robustness Against Natural Language Word Substitutions (61 citations)
- Witches' Brew: Industrial Scale Data Poisoning via Gradient Matching (34 citations)
- Dataset Inference: Ownership Resolution in Machine Learning (23 citations)
- A Panda? No, It's a Sloth: Slowdown Attacks on Adaptive Multi-Exit Neural Network Inference (22 citations)
- Efficient Certified Defenses Against Patch Attacks on Image Classifiers (21 citations)
- Stabilized Medical Image Attacks (16 citations)
- Online Adversarial Purification based on Self-supervised Learning (12 citations)
- Stochastic Security: Adversarial Defense Using Long-Run Dynamics of Energy-Based Models (9 citations)
- Shape-Texture Debiased Neural Network Training (5 citations)
- Statistical inference for individual fairness (2 citations)
and 1 more in the interactive map.
Black-box and transfer attacks · 5 papers
- Deep Neural Network Fingerprinting by Conferrable Adversarial Examples (22 citations)
- A Unified Approach to Interpreting and Boosting Adversarial Transferability (15 citations)
- Protecting DNNs from Theft using an Ensemble of Diverse Models (6 citations)
- Towards Robust Neural Networks via Close-loop Control (5 citations)
- Policy-Driven Attack: Learning to Query for Hard-label Black-box Adversarial Examples (2 citations)
