Xiyuan Yang 杨熙元

I am a 4th year undergraduate student (Sep. 2021 - Jun. 2025) in School of Computer Science at Wuhan University, supervised by Prof. Mang Ye. My research interest includes Trustworthy Machine Learning (Federated Learning, Differential Privacy, etc.) and Societial Application of Large Language Models.

Currently, I'm a research intern in the Microsoft Research Asia, Social Computing Group, supervised by principal researcher Fangzhao Wu. Previously, I've been a remote research intern at UChicago supervised by Prof. Tian Li.

I'm looking for PhD positions (Fall 2025). If you are interested in collaborating with me or want to have a chat, always feel free to contact me : )

Email  /  Google Scholar  /  Github   

We proposed Federated Parameter-Alignment and Client-Synchronization (FedAS) to handel the prevailing challenge of inter and intra-client inconsistency in the Personalized Federated Learning domain.

We introduced the FedDPA algorithm, leveraging Dynamic Fisher Personalization and Adaptive Constraint to enhance model performance and adaptiveness under Differential Privacy.

We proposed the AugHFL altorithm to address the data corruption issue in heterogeneous FL by data augmentation and weighted aggregation.

We propose a simple technique to improve privacy/utility tradeoffs for federated clustering by randomly rebalancing model updates.

Source code credit to Dr. Jon Barron