ICLR 2022 accepted papers on Adversarial robustness and security
54 of the 1,095 papers accepted at ICLR 2022 (4.9%) are on Adversarial robustness and security, compared with 5.0% at ICLR 2021. They fall into 4 topics.
See ICLR 2022 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 · 24 papers
- Adversarial Robustness Through the Lens of Causality (19 citations)
- Self-ensemble Adversarial Training for Improved Robustness (19 citations)
- Exploring Memorization in Adversarial Training (14 citations)
- Who Is the Strongest Enemy? Towards Optimal and Efficient Evasion Attacks in Deep RL (13 citations)
- Boosting Randomized Smoothing with Variance Reduced Classifiers (8 citations)
- On the Certified Robustness for Ensemble Models and Beyond (7 citations)
- A Unified Wasserstein Distributional Robustness Framework for Adversarial Training (6 citations)
- Robust Learning Meets Generative Models: Can Proxy Distributions Improve Adversarial Robustness? (6 citations)
- Almost Tight L0-norm Certified Robustness of Top-k Predictions against Adversarial Perturbations (5 citations)
- Evading Adversarial Example Detection Defenses with Orthogonal Projected Gradient Descent (5 citations)
- Finding Biological Plausibility for Adversarially Robust Features via Metameric Tasks (5 citations)
- Generalization of Neural Combinatorial Solvers Through the Lens of Adversarial Robustness (4 citations)
and 12 more in the interactive map.
Attacks on text and graphs · 15 papers
- Reliable Adversarial Distillation with Unreliable Teachers (20 citations)
- Poisoning and Backdooring Contrastive Learning (19 citations)
- RelaxLoss: Defending Membership Inference Attacks without Losing Utility (11 citations)
- On Robust Prefix-Tuning for Text Classification (9 citations)
- Reverse Engineering of Imperceptible Adversarial Image Perturbations (8 citations)
- Increasing the Cost of Model Extraction with Calibrated Proof of Work (7 citations)
- Non-Transferable Learning: A New Approach for Model Ownership Verification and Applicability Authorization (6 citations)
- On the Importance of Difficulty Calibration in Membership Inference Attacks (6 citations)
- Data Poisoning Won’t Save You From Facial Recognition (5 citations)
- Fooling Explanations in Text Classifiers (5 citations)
- Learnability Lock: Authorized Learnability Control Through Adversarial Invertible Transformations (1 citation)
- Real-Time Neural Voice Camouflage (1 citation)
and 3 more in the interactive map.
Black-box and transfer attacks · 8 papers
- Transferable Adversarial Attack based on Integrated Gradients (18 citations)
- How to Robustify Black-Box ML Models? A Zeroth-Order Optimization Perspective (12 citations)
- On Improving Adversarial Transferability of Vision Transformers (11 citations)
- Attacking deep networks with surrogate-based adversarial black-box methods is easy (10 citations)
- Beyond ImageNet Attack: Towards Crafting Adversarial Examples for Black-box Domains (5 citations)
- QUERY EFFICIENT DECISION BASED SPARSE ATTACKS AGAINST BLACK-BOX DEEP LEARNING MODELS (2 citations)
- Demystifying Limited Adversarial Transferability in Automatic Speech Recognition Systems
- Rethinking Adversarial Transferability from a Data Distribution Perspective
Backdoor and poisoning · 7 papers
- Adversarial Unlearning of Backdoors via Implicit Hypergradient (36 citations)
- BadPre: Task-agnostic Backdoor Attacks to Pre-trained NLP Foundation Models (33 citations)
- Few-Shot Backdoor Attacks on Visual Object Tracking (21 citations)
- AEVA: Black-box Backdoor Detection Using Adversarial Extreme Value Analysis (16 citations)
- How to Inject Backdoors with Better Consistency: Logit Anchoring on Clean Data (10 citations)
- Trigger Hunting with a Topological Prior for Trojan Detection (7 citations)
- Post-Training Detection of Backdoor Attacks for Two-Class and Multi-Attack Scenarios (5 citations)
