5 papers
Towards Strong Certified Defense with Universal Asymmetric Randomization
Hanbin Hong, Ashish Kundu, Ali Payani +2
Randomized smoothing has become essential for achieving certified adversarial robustness in machine learning models. However, current methods primarily use isotropic noise distribu…
Certifiable Black-Box Attacks with Randomized Adversarial Examples: Breaking Defenses with Provable Confidence
Hanbin Hong, Xinyu Zhang, Binghui Wang +2
Black-box adversarial attacks have demonstrated strong potential to compromise machine learning models by iteratively querying the target model or leveraging transferability from a…
Text-CRS: A Generalized Certified Robustness Framework against Textual Adversarial Attacks
Xinyu Zhang, Hanbin Hong, Yuan Hong +4
The language models, especially the basic text classification models, have been shown to be susceptible to textual adversarial attacks such as synonym substitution and word inserti…
An LLM-Assisted Easy-to-Trigger Backdoor Attack on Code Completion Models: Injecting Disguised Vulnerabilities against Strong Detection
Shenao Yan, Shen Wang, Yue Duan +4
Large Language Models (LLMs) have transformed code completion tasks, providing context-based suggestions to boost developer productivity in software engineering. As users often fin…
Certifying Adapters: Enabling and Enhancing the Certification of Classifier Adversarial Robustness
Jieren Deng, Hanbin Hong, Aaron Palmer +5
Randomized smoothing has become a leading method for achieving certified robustness in deep classifiers against l_{p}-norm adversarial perturbations. Current approaches for achievi…