ICML 2021 accepted papers on Adversarial robustness and security
50 of the 1,184 papers accepted at ICML 2021 (4.2%) are on Adversarial robustness and security, compared with 4.5% at ICML 2020. They fall into 4 topics.
See ICML 2021 in the interactive map
Submitting to ICML? 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 · 27 papers
- Measuring Robustness in Deep Learning Based Compressive Sensing (50 citations)
- Globally-Robust Neural Networks (29 citations)
- To be Robust or to be Fair: Towards Fairness in Adversarial Training (23 citations)
- Maximum Mean Discrepancy Test is Aware of Adversarial Attacks (15 citations)
- CIFS: Improving Adversarial Robustness of CNNs via Channel-wise Importance-based Feature Selection (11 citations)
- Efficient Training of Robust Decision Trees Against Adversarial Examples (11 citations)
- Improved OOD Generalization via Adversarial Training and Pretraing (10 citations)
- Double-Win Quant: Aggressively Winning Robustness of Quantized Deep Neural Networks via Random Precision Training and Inference (9 citations)
- Improved, Deterministic Smoothing for L_1 Certified Robustness (9 citations)
- Improving Gradient Regularization using Complex-Valued Neural Networks (9 citations)
- Towards Better Robust Generalization with Shift Consistency Regularization (9 citations)
- Training Adversarially Robust Sparse Networks via Bayesian Connectivity Sampling (9 citations)
and 15 more in the interactive map.
Attacks on text and graphs · 11 papers
- Label-Only Membership Inference Attacks (97 citations)
- Learning from History for Byzantine Robust Optimization (55 citations)
- Adversarial Purification with Score-based Generative Models (32 citations)
- Knowledge Enhanced Machine Learning Pipeline against Diverse Adversarial Attacks (21 citations)
- Towards Defending against Adversarial Examples via Attack-Invariant Features (15 citations)
- Grey-box Extraction of Natural Language Models (8 citations)
- Model-Targeted Poisoning Attacks with Provable Convergence (7 citations)
- A General Framework For Detecting Anomalous Inputs to DNN Classifiers (6 citations)
- Learning to Generate Noise for Multi-Attack Robustness (6 citations)
- Making Paper Reviewing Robust to Bid Manipulation Attacks (6 citations)
- Watermarking Deep Neural Networks with Greedy Residuals
Black-box and transfer attacks · 9 papers
- Weight-covariance alignment for adversarially robust neural networks (8 citations)
- RNNRepair: Automatic RNN Repair via Model-based Analysis (7 citations)
- Progressive-Scale Boundary Blackbox Attack via Projective Gradient Estimation (5 citations)
- A Zeroth-Order Block Coordinate Descent Algorithm for Huge-Scale Black-Box Optimization (4 citations)
- Uncovering the Connections Between Adversarial Transferability and Knowledge Transferability (2 citations)
- PopSkipJump: Decision-Based Attack for Probabilistic Classifiers (1 citation)
- Query Complexity of Adversarial Attacks (1 citation)
- SPADE: A Spectral Method for Black-Box Adversarial Robustness Evaluation (1 citation)
- Neural Tangent Generalization Attacks
