ICML 2022 accepted papers on Adversarial robustness and security
57 of the 1,233 papers accepted at ICML 2022 (4.6%) are on Adversarial robustness and security, compared with 4.2% at ICML 2021. They fall into 4 topics.
See ICML 2022 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 · 40 papers
- Robustness and Accuracy Could Be Reconcilable by (Proper) Definition (25 citations)
- Revisiting and Advancing Fast Adversarial Training Through The Lens of Bi-Level Optimization (19 citations)
- Detecting Adversarial Examples Is (Nearly) As Hard As Classifying Them (11 citations)
- Evaluating the Adversarial Robustness of Adaptive Test-time Defenses (10 citations)
- Improved Certified Defenses against Data Poisoning with (Deterministic) Finite Aggregation (10 citations)
- Building Robust Ensembles via Margin Boosting (7 citations)
- On the Adversarial Robustness of Causal Algorithmic Recourse (6 citations)
- Probabilistically Robust Learning: Balancing Average and Worst-case Performance (6 citations)
- Removing Batch Normalization Boosts Adversarial Training (6 citations)
- Understanding Robust Overfitting of Adversarial Training and Beyond (6 citations)
- Fast and Reliable Evaluation of Adversarial Robustness with Minimum-Margin Attack (5 citations)
- On Collective Robustness of Bagging Against Data Poisoning (4 citations)
and 28 more in the interactive map.
Attacks on text and graphs · 11 papers
- Diffusion Models for Adversarial Purification (76 citations)
- Linear Adversarial Concept Erasure (23 citations)
- Adversarially trained neural representations are already as robust as biological neural representations (19 citations)
- Certified Neural Network Watermarks with Randomized Smoothing (11 citations)
- Certified Robustness Against Natural Language Attacks by Causal Intervention (9 citations)
- Modeling Adversarial Noise for Adversarial Training (2 citations)
- Not All Poisons are Created Equal: Robust Training against Data Poisoning (2 citations)
- Generative Trees: Adversarial and Copycat
- How to Steer Your Adversary: Targeted and Efficient Model Stealing Defenses with Gradient Redirection
- Plug & Play Attacks: Towards Robust and Flexible Model Inversion Attacks
- Generating Distributional Adversarial Examples to Evade Statistical Detectors
Black-box and transfer attacks · 5 papers
- Query-Efficient and Scalable Black-Box Adversarial Attacks on Discrete Sequential Data via Bayesian Optimization (9 citations)
- Diversified Adversarial Attacks based on Conjugate Gradient Method (3 citations)
- Adversarial Attacks on Gaussian Process Bandits (1 citation)
- Rethinking Image-Scaling Attacks: The Interplay Between Vulnerabilities in Machine Learning Systems (1 citation)
- LCANets: Lateral Competition Improves Robustness Against Corruption and Attack
