Principled Defenses Against Adversarial Attacks on Deep Learning
ABOUT THIS PROJECT
At a glance
In this project, we propose to provide theoretical foundations for secure learning systems and investigate principled methods to improve model robustness in diverse real-world applications. The scope of the project will cover both data poisoning and evasion attacks. We will contribute several defense strategies to make the system more robust in the BDD Code Repository as well as benchmark datasets for robustness evaluation.
|Dawn Song||Ruoxi Jia|
adversarial machine learning, evasion attacks, data poisoning attacks, defense