ICLR 2020 accepted papers on Adversarial robustness and security
49 of the 687 papers accepted at ICLR 2020 (7.1%) are on Adversarial robustness and security, compared with 5.8% at ICLR 2019. They fall into 4 topics.
See ICLR 2020 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 · 28 papers
- Fast is better than free: Revisiting adversarial training (576 citations)
- Improving Adversarial Robustness Requires Revisiting Misclassified Examples (285 citations)
- Rethinking Softmax Cross-Entropy Loss for Adversarial Robustness (98 citations)
- MMA Training: Direct Input Space Margin Maximization through Adversarial Training (92 citations)
- Adversarial AutoAugment (87 citations)
- Adversarial Training and Provable Defenses: Bridging the Gap (79 citations)
- MACER: Attack-free and Scalable Robust Training via Maximizing Certified Radius (63 citations)
- Mixup Inference: Better Exploiting Mixup to Defend Adversarial Attacks (63 citations)
- Triple Wins: Boosting Accuracy, Robustness and Efficiency Together by Enabling Input-Adaptive Inference (55 citations)
- Robust Local Features for Improving the Generalization of Adversarial Training (52 citations)
- Towards Stable and Efficient Training of Verifiably Robust Neural Networks (40 citations)
- On Robustness of Neural Ordinary Differential Equations (39 citations)
and 16 more in the interactive map.
Attacks on text and graphs · 11 papers
- FreeLB: Enhanced Adversarial Training for Natural Language Understanding (286 citations)
- Certified Defenses for Adversarial Patches (73 citations)
- Unrestricted Adversarial Examples via Semantic Manipulation (68 citations)
- Prediction Poisoning: Towards Defenses Against DNN Model Stealing Attacks (63 citations)
- Detecting and Diagnosing Adversarial Images with Class-Conditional Capsule Reconstructions (54 citations)
- Fooling Detection Alone is Not Enough: Adversarial Attack against Multiple Object Tracking (54 citations)
- Toward Evaluating Robustness of Deep Reinforcement Learning with Continuous Control (14 citations)
- Biologically inspired sleep algorithm for increased generalization and adversarial robustness in deep neural networks (12 citations)
- Defending Against Physically Realizable Attacks on Image Classification (12 citations)
- White Noise Analysis of Neural Networks (3 citations)
- Optimal Strategies Against Generative Attacks (2 citations)
Black-box and transfer attacks · 8 papers
- Nesterov Accelerated Gradient and Scale Invariance for Adversarial Attacks (230 citations)
- Skip Connections Matter: On the Transferability of Adversarial Examples Generated with ResNets (186 citations)
- Black-Box Adversarial Attack with Transferable Model-based Embedding (66 citations)
- Query-efficient Meta Attack to Deep Neural Networks (48 citations)
- BayesOpt Adversarial Attack (44 citations)
- Sign-OPT: A Query-Efficient Hard-label Adversarial Attack (29 citations)
- Sign Bits Are All You Need for Black-Box Attacks (26 citations)
- Transferable Perturbations of Deep Feature Distributions (18 citations)
