Jason Yang

Vin University. Hanoi, Vietnam.

jason.png

I’m Yang Sze Jue (Jason), currently a Research Assistant under the supervision of Prof. Khoa D. Doan from VinUniversity, Hanoi, Vietnam. Prior to that, I was advised by Prof. Chan Chee Seng from Universiti Malaya, Kuala Lumpur, Malaysia.

I am interested in the following topics:

  • Backdoor attacks and defenses
    • I was fascinated by the movie - “The Matrix”, where the main character acts as an ultimate hacker to save the world.Therefore, I am interested in investigating the security vulnerabilities of deep learning models, and its potential mitigations.
  • Watermarking
    • I am interested in Intellectual Property Right (IPR) of deep learning models, and its potential loopholes in verification methods.

news

Dec 6, 2023 A paper titled - “Synthesizing Physical Backdoor Datasets: An Automated Framework Leveraging Deep Generative Models” submitted on arXiv, where we proposed a framework on synthesizing datasets that are comparable as a real dataset to accelerate physical backdoor research.
Aug 31, 2023 A paper titled - “Everyone Can Attack: Repurpose Lossy Compression as a Natural Backdoor Attack” submitted on arXiv, where we identified lossy image compressions as an accesible way of launching backdoor attacks effortlessly.

selected publications

  1. Synthesizing Physical Backdoor Datasets: An Automated Framework Leveraging Deep Generative Models
    Sze Jue Yang, Chinh D. La, Quang H. Nguyen, Eugene Bagdasaryan, Kok-Seng Wong, Anh Tuan Tran, Chee Seng Chan, and Khoa D. Doan
    2023
  2. Everyone Can Attack: Repurpose Lossy Compression as a Natural Backdoor Attack
    Sze Jue Yang, Quang Nguyen, Chee Seng Chan, and Khoa D. Doan
    2023