CSE Department Speaker Series - Adversary for Social Good: Turning Adversarial Attacks into Applications by Lingwei Chen Ph.D.

Abstract:

Despite their remarkable inference ability, machine learning models are faced with the inherent learning-security challenge of lacking adversarial robustness. In other words, they are vulnerable to adversarial attacks that can easily fool the models into misclassification by adding small perturbations to the input data. Accordingly, understanding machine learning security in adversarial settings has significantly emerged as a mainstream topic in AI domain. Most of the current research works in this line follow either adversarial or defensive perspectives to analyze methodologies of each other and develop strategies to overcome the opponents. However, few of them go beyond adversary, and turn these attacks into applications for social good. This talk will cover some of our representative works that advance adversarial vulnerability of machine learning for supporting real-world tasks. 

If unable to attend in person, sign up for the webinar. Register Here