NeurIPS 2023 accepted papers on Adversarial robustness and security
117 of the 3,584 papers accepted at NeurIPS 2023 (3.3%) are on Adversarial robustness and security, compared with 3.6% at NeurIPS 2022. They fall into 4 topics.
See NeurIPS 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 · 55 papers
- Revisiting Adversarial Training for ImageNet: Architectures, Training and Generalization across Threat Models (13 citations)
- DiffAttack: Evasion Attacks Against Diffusion-Based Adversarial Purification (7 citations)
- Efficient Adversarial Contrastive Learning via Robustness-Aware Coreset Selection (7 citations)
- Eliminating Catastrophic Overfitting Via Abnormal Adversarial Examples Regularization (5 citations)
- IPMix: Label-Preserving Data Augmentation Method for Training Robust Classifiers (5 citations)
- Connecting Certified and Adversarial Training (4 citations)
- Enhancing Adversarial Contrastive Learning via Adversarial Invariant Regularization (4 citations)
- Revisiting Adversarial Robustness Distillation from the Perspective of Robust Fairness (4 citations)
- Robust low-rank training via approximate orthonormal constraints (4 citations)
- Generalizable Lightweight Proxy for Robust NAS against Diverse Perturbations (3 citations)
- Hierarchical Randomized Smoothing (3 citations)
- NEO-KD: Knowledge-Distillation-Based Adversarial Training for Robust Multi-Exit Neural Networks (3 citations)
and 43 more in the interactive map.
Attacks on text and graphs · 31 papers
- DecodingTrust: A Comprehensive Assessment of Trustworthiness in GPT Models (62 citations)
- Are aligned neural networks adversarially aligned? (25 citations)
- Content-based Unrestricted Adversarial Attack (15 citations)
- VLATTACK: Multimodal Adversarial Attacks on Vision-Language Tasks via Pre-trained Models (13 citations)
- Diffusion-Based Adversarial Sample Generation for Improved Stealthiness and Controllability (10 citations)
- Scalable Membership Inference Attacks via Quantile Regression (8 citations)
- Enhancing Adversarial Robustness via Score-Based Optimization (6 citations)
- Alignment with human representations supports robust few-shot learning (5 citations)
- Certifiably Robust Graph Contrastive Learning (3 citations)
- GAN You See Me? Enhanced Data Reconstruction Attacks against Split Inference (3 citations)
- Spatial-frequency channels, shape bias, and adversarial robustness (3 citations)
- Temporal Robustness against Data poisoning (3 citations)
and 19 more in the interactive map.
Black-box and transfer attacks · 16 papers
- Boosting Adversarial Transferability by Achieving Flat Local Maxima (20 citations)
- TrojLLM: A Black-box Trojan Prompt Attack on Large Language Models (12 citations)
- Rethinking the Backward Propagation for Adversarial Transferability (8 citations)
- Understanding and Improving Ensemble Adversarial Defense (8 citations)
- A Theory of Transfer-Based Black-Box Attacks: Explanation and Implications (4 citations)
- Improving Adversarial Transferability via Intermediate-level Perturbation Decay (3 citations)
- Label-Only Model Inversion Attacks via Knowledge Transfer (3 citations)
- Vulnerabilities in Video Quality Assessment Models: The Challenge of Adversarial Attacks (3 citations)
- Blurred-Dilated Method for Adversarial Attacks (2 citations)
- Marich: A Query-efficient Distributionally Equivalent Model Extraction Attack (2 citations)
- Towards Evaluating Transfer-based Attacks Systematically, Practically, and Fairly (2 citations)
- BERT Lost Patience Won't Be Robust to Adversarial Slowdown
and 4 more in the interactive map.
Backdoor and poisoning · 15 papers
- Shared Adversarial Unlearning: Backdoor Mitigation by Unlearning Shared Adversarial Examples (12 citations)
- Defending Pre-trained Language Models as Few-shot Learners against Backdoor Attacks (11 citations)
- IBA: Towards Irreversible Backdoor Attacks in Federated Learning (8 citations)
- ParaFuzz: An Interpretability-Driven Technique for Detecting Poisoned Samples in NLP (8 citations)
- FedGame: A Game-Theoretic Defense against Backdoor Attacks in Federated Learning (7 citations)
- Neural Polarizer: A Lightweight and Effective Backdoor Defense via Purifying Poisoned Features (7 citations)
- Black-box Backdoor Defense via Zero-shot Image Purification (5 citations)
- Robust Contrastive Language-Image Pretraining against Data Poisoning and Backdoor Attacks (5 citations)
- Towards Stable Backdoor Purification through Feature Shift Tuning (5 citations)
- BIRD: Generalizable Backdoor Detection and Removal for Deep Reinforcement Learning (4 citations)
- CBD: A Certified Backdoor Detector Based on Local Dominant Probability (4 citations)
- Label Poisoning is All You Need (4 citations)
and 3 more in the interactive map.