In the new AI era, seeing is no longer be believing. Ensuring the integrity and authentication of digital images is increasingly challenging and vital. This talk gives a brief review of my group’s research efforts in the intersection of deep learning and digital media security -- adversarial deep learning, including both adversarial attacks and adversarial defenses. Deep learning has achieved state-of-the-art performances in many applications. Unfortunately, both digital images and current deep learning models are vulnerable to manipulations and attacks, giving rise to security, privacy and reliability issues in practical applications.

Under the paradigm of adversarial deep learning, as an attacker, we study potential adversarial attacks and explore novel approaches to study potential vulnerabilities of deep learning models, by investigating three fundamental learning tasks: matching, classification and regression. We present novel attacks (both in the digital domain and in the physical domain) for several essential models belonging to the above three dominant tasks: e.g. GAN-generated fake face imagery forensics; multiclass image classification; camera-LIDAR 3d object detection; and single object tracking in videos. We address the related security threats in the above problems and study how to fool deep learning models to make wrong decisions.
 

Speaker: Professor Z. Jane WANG 
Date: 3 November 2021 (Wednesday)
Time: 10:00am – 11:00am
Poster: Click here