Papers

ICLR 2021 high score papers

1. Geometry-aware Instance-reweighted Adversarial Training
2. Improving Adversarial Robustness via Channel-wise Activation Suppressing
3. How Does Mixup Help With Robustness and Generalization?
4. Towards A Unified Understanding and Improving of Adversarial Transferability
5. When Optimizing $f$-Divergence is Robust with Label Noise
6. A Hypergradient Approach to Robust Regression without Correspondence Does enhanced shape bias improve neural network robustness to common corruptions?
7. you only need adversarial supervision for semantic image synthesis
8. SOAR: Second-Order Adversarial Regularization
9. ARMOURED: Adversarially Robust MOdels using Unlabeled data by REgularizing Diversity
10. Perceptual Adversarial Robustness: Generalizable Defenses Against Unforeseen Threat Models
11. Collective Robustness Certificates
12. Perceptual Adversarial Robustness: Generalizable Defenses Against Unforeseen Threat Models
13. Distributionally Robust Learning for Unsupervised Domain Adaptation
14. On Fast Adversarial Robustness Adaptation in Model-Agnostic Meta-Learning
15. Evaluations and Methods for Explanation through Robustness Analysis
16. Fast Geometric Projections for Local Robustness Certification
17. Heating up decision boundaries: isocapacitory saturation, adversarial scenarios and generalization bounds

Attack

White Box attack

1. Stronger and Faster Wasserstein Adversarial Attacks ~pdf

Poisoning

1. Transferable Clean-Label Poisoning Attacks on Deep Neural Nets ~pdf

Backdoor attack

1. Targeted Backdoor Attacks on Deep Learning Systems Using Data Poisoning~pdf

Attack in other domains

1. DBA: Distributed Backdoor Attacks Against Federated Learning~pdf
2. Adversarial Attacks on Copyright Detection Systems ~pdf

Defense

Adversarial learning variants

1. Theoretically Principled Trade-off between Robustness and Accuracy~pdf
2. You only propagate once: Accelerating adversarial training via maximal principle~pdf
3. Fast is better than free: Revisiting adversarial training~pdf
4. Attacks Which Do Not Kill Training Make Adversarial Learning Stronger~pdf
5. Towards Understanding the Dynamics of the First-Order Adversaries~pdf

Provalable defense

1. Certified Adversarial Robustness via Randomized Smoothing~pdf

Others

1. Adverserial Examples are Not Bugs, They are Features. ~pdf ~Intro
2. Rethinking the security of skip connections in resent-like neural networks. ~pdf ~Intro