NeurIPS 2024 accepted papers on Adversarial robustness and security
101 of the 4,560 papers accepted at NeurIPS 2024 (2.2%) are on Adversarial robustness and security, compared with 3.3% at NeurIPS 2023. They fall into 4 topics.
See NeurIPS 2024 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
Attacks on text and graphs · 49 papers
- AgentDojo: A Dynamic Environment to Evaluate Prompt Injection Attacks and Defenses for LLM Agents (83 citations)
- JailbreakBench: An Open Robustness Benchmark for Jailbreaking Large Language Models (78 citations)
- Improving Alignment and Robustness with Circuit Breakers (11 citations)
- Efficient Adversarial Training in LLMs with Continuous Attacks (10 citations)
- Fight Back Against Jailbreaking via Prompt Adversarial Tuning (9 citations)
- Robust Prompt Optimization for Defending Language Models Against Jailbreaking Attacks (9 citations)
- Rainbow Teaming: Open-Ended Generation of Diverse Adversarial Prompts (7 citations)
- Reconstruction Attacks on Machine Unlearning: Simple Models are Vulnerable (7 citations)
- AudioMarkBench: Benchmarking Robustness of Audio Watermarking (6 citations)
- DiffuPac: Contextual Mimicry in Adversarial Packets Generation via Diffusion Model (6 citations)
- Few-Shot Adversarial Prompt Learning on Vision-Language Models (6 citations)
- AdvAD: Exploring Non-Parametric Diffusion for Imperceptible Adversarial Attacks (5 citations)
and 37 more in the interactive map.
Certified and empirical robustness · 33 papers
- Diffusion Models are Certifiably Robust Classifiers (14 citations)
- Curvature Clues: Decoding Deep Learning Privacy with Input Loss Curvature (4 citations)
- On the Scalability of Certified Adversarial Robustness with Generated Data (4 citations)
- SuperDeepFool: a new fast and accurate minimal adversarial attack (4 citations)
- GREAT Score: Global Robustness Evaluation of Adversarial Perturbation using Generative Models (3 citations)
- Training for Stable Explanation for Free (3 citations)
- Verifiably Robust Conformal Prediction (3 citations)
- Collaboration! Towards Robust Neural Methods for Routing Problems (2 citations)
- Constrained Adaptive Attack: Effective Adversarial Attack Against Deep Neural Networks for Tabular Data (2 citations)
- Faster Repeated Evasion Attacks in Tree Ensembles (2 citations)
- PureGen: Universal Data Purification for Train-Time Poison Defense via Generative Model Dynamics (2 citations)
- RAMP: Boosting Adversarial Robustness Against Multiple $l_p$ Perturbations for Universal Robustness (2 citations)
and 21 more in the interactive map.
Backdoor and poisoning · 12 papers
- Mitigating Backdoor Attack by Injecting Proactive Defensive Backdoor (6 citations)
- Data Free Backdoor Attacks (5 citations)
- BAN: Detecting Backdoors Activated by Adversarial Neuron Noise (4 citations)
- SampDetox: Black-box Backdoor Defense via Perturbation-based Sample Detoxification (4 citations)
- SleeperNets: Universal Backdoor Poisoning Attacks Against Reinforcement Learning Agents (4 citations)
- Breaking the False Sense of Security in Backdoor Defense through Re-Activation Attack (3 citations)
- WaveAttack: Asymmetric Frequency Obfuscation-based Backdoor Attacks Against Deep Neural Networks (3 citations)
- Injecting Undetectable Backdoors in Obfuscated Neural Networks and Language Models (1 citation)
- Uncovering, Explaining, and Mitigating the Superficial Safety of Backdoor Defense (1 citation)
- Unveiling and Mitigating Backdoor Vulnerabilities based on Unlearning Weight Changes and Backdoor Activeness (1 citation)
- LT-Defense: Searching-free Backdoor Defense via Exploiting the Long-tailed Effect
- Unelicitable Backdoors via Cryptographic Transformer Circuits
Black-box and transfer attacks · 7 papers
- Boosting the Transferability of Adversarial Attack on Vision Transformer with Adaptive Token Tuning (7 citations)
- Query-Based Adversarial Prompt Generation (6 citations)
- When LLM Meets DRL: Advancing Jailbreaking Efficiency via DRL-guided Search (6 citations)
- Transferability Bound Theory: Exploring Relationship between Adversarial Transferability and Flatness (3 citations)
- Improved Generation of Adversarial Examples Against Safety-aligned LLMs (1 citation)
- Amnesia as a Catalyst for Enhancing Black Box Pixel Attacks in Image Classification and Object Detection
- Ask, Attend, Attack: An Effective Decision-Based Black-Box Targeted Attack for Image-to-Text Models