ICLR 2023 accepted papers on Adversarial robustness and security
57 of the 1,575 papers accepted at ICLR 2023 (3.6%) are on Adversarial robustness and security, compared with 4.9% at ICLR 2022. They fall into 4 topics.
See ICLR 2023 in the interactive map
Working on something in this area? Describe your idea in scime atlas to see the closest accepted papers by meaning and how crowded the spot has become.
Topics and papers
Certified and empirical robustness · 26 papers
- (Certified!!) Adversarial Robustness for Free! (17 citations)
- DensePure: Understanding Diffusion Models for Adversarial Robustness (17 citations)
- Certified Training: Small Boxes are All You Need (10 citations)
- Exploring and Exploiting Decision Boundary Dynamics for Adversarial Robustness (10 citations)
- Data augmentation alone can improve adversarial training (9 citations)
- Why adversarial training can hurt robust accuracy (6 citations)
- One-Pixel Shortcut: On the Learning Preference of Deep Neural Networks (5 citations)
- Self-Ensemble Protection: Training Checkpoints Are Good Data Protectors (5 citations)
- On Explaining Neural Network Robustness with Activation Path (2 citations)
- Part-Based Models Improve Adversarial Robustness (2 citations)
- Rethinking the Effect of Data Augmentation in Adversarial Contrastive Learning (2 citations)
- Squeeze Training for Adversarial Robustness (2 citations)
and 14 more in the interactive map.
Attacks on text and graphs · 17 papers
- Aligning Model and Macaque Inferior Temporal Cortex Representations Improves Model-to-Human Behavioral Alignment and Adversarial Robustness (29 citations)
- Understanding Zero-shot Adversarial Robustness for Large-Scale Models (9 citations)
- Adversarial Training of Self-supervised Monocular Depth Estimation against Physical-World Attacks (7 citations)
- Canary in a Coalmine: Better Membership Inference with Ensembled Adversarial Queries (6 citations)
- Transferable Unlearnable Examples (6 citations)
- Defending against Adversarial Audio via Diffusion Model (5 citations)
- TrojText: Test-time Invisible Textual Trojan Insertion (5 citations)
- Fooling SHAP with Stealthily Biased Sampling (4 citations)
- Indiscriminate Poisoning Attacks on Unsupervised Contrastive Learning (4 citations)
- Toward Adversarial Training on Contextualized Language Representation (4 citations)
- Certified Defences Against Adversarial Patch Attacks on Semantic Segmentation (3 citations)
- Robust Fair Clustering: A Novel Fairness Attack and Defense Framework (2 citations)
and 5 more in the interactive map.
Backdoor and poisoning · 8 papers
- SCALE-UP: An Efficient Black-box Input-level Backdoor Detection via Analyzing Scaled Prediction Consistency (18 citations)
- Distilling Cognitive Backdoor Patterns within an Image (7 citations)
- UNICORN: A Unified Backdoor Trigger Inversion Framework (7 citations)
- Few-shot Backdoor Attacks via Neural Tangent Kernels (5 citations)
- Incompatibility Clustering as a Defense Against Backdoor Poisoning Attacks (1 citation)
- Clean-image Backdoor: Attacking Multi-label Models with Poisoned Labels Only
- Revisiting the Assumption of Latent Separability for Backdoor Defenses
- The Dark Side of AutoML: Towards Architectural Backdoor Search
Black-box and transfer attacks · 6 papers
- Making Substitute Models More Bayesian Can Enhance Transferability of Adversarial Examples (5 citations)
- Bort: Towards Explainable Neural Networks with Bounded Orthogonal Constraint (3 citations)
- Boosting Adversarial Transferability using Dynamic Cues (2 citations)
- On the Perils of Cascading Robust Classifiers
- ROCO: A General Framework for Evaluating Robustness of Combinatorial Optimization Solvers on Graphs
- ILA-DA: Improving Transferability of Intermediate Level Attack with Data Augmentation