Adversarial Attack
Adversarial attacks aim to deceive machine learning models by subtly altering input data, causing misclassifications or other erroneous outputs. Current research focuses on developing more robust models and detection methods, exploring various attack strategies across different model architectures (including vision transformers, recurrent neural networks, and graph neural networks) and data types (images, text, signals, and tabular data). Understanding and mitigating these attacks is crucial for ensuring the reliability and security of AI systems in diverse applications, from autonomous vehicles to medical diagnosis and cybersecurity.
Papers
Hacking Predictors Means Hacking Cars: Using Sensitivity Analysis to Identify Trajectory Prediction Vulnerabilities for Autonomous Driving Security
Marsalis Gibson, David Babazadeh, Claire Tomlin, Shankar Sastry
Adapters Mixup: Mixing Parameter-Efficient Adapters to Enhance the Adversarial Robustness of Fine-tuned Pre-trained Text Classifiers
Tuc Nguyen, Thai Le
Power in Numbers: Robust reading comprehension by finetuning with four adversarial sentences per example
Ariel Marcus
HGAttack: Transferable Heterogeneous Graph Adversarial Attack
He Zhao, Zhiwei Zeng, Yongwei Wang, Deheng Ye, Chunyan Miao
Bag of Tricks to Boost Adversarial Transferability
Zeliang Zhang, Rongyi Zhu, Wei Yao, Xiaosen Wang, Chenliang Xu
Revealing Vulnerabilities in Stable Diffusion via Targeted Attacks
Chenyu Zhang, Lanjun Wang, Anan Liu
A Generative Adversarial Attack for Multilingual Text Classifiers
Tom Roth, Inigo Jauregi Unanue, Alsharif Abuadbba, Massimo Piccardi