Weiyu Sun

Hello! My name is Weiyu Sun.

My interests are different kinds of health-related algorithms (e.g., video-based heart rate information process) and their practical applications (e.g., using these techniques for healthcare), and Trustworthy AI (e.g., the reliability of different healthcare-related algorithms).

I earned my B.E. and M.E. degrees in Electornic & Computer Engineering at Nanjing University, where I mainly dealt with healthcare applications (e.g., non-contact health information processing, such as heart rate and ECG).

I also conducted remote teamwork as a visiting student with the Hong Kong University of Science and Technology (Guangzhou).

Currently I'm going to chase my Ph.D degree at Georgia Institute of Technology this Fall! Wishing me a smooth journey!

Email / Orcid / Google Scholar

It offers a generlizable backdoor behavior mitigation method for the LLM community by effectively identifying and neutralizing backdoor fingerprints in the embedding space, without relying on assumptions about trigger characteristics or locations.

We propose a bi-level optimization way to generate the effective trigger pattern to enhance the robustness and effectiveness of the backdoor attack in self-supervised learning scenarios.

We reveal two characteristic domain conflict issues in the DNN rPPG (remote photoplethysmography) model training, and propose a new training framework (DOHA) to deal with these conflicts and enhance the performance of rPPG models.

We propose a self-supervised training strategy to efficiently train the rPPG DNN model.

Academic Life

Awards and Expertise

Internship

Works & Publishment

Internship

Works & Publishments

Internship

Works & Publishments

Thanks for you reading!

flowers: