Hello, I'm Jingyuan Zhang (张靖远)

I am a researcher in artificial intelligence and federated learning. I am currently a Research Assistant at Nanyang Technological University (NTU), where I focus on enhancing security and privacy in federated learning and retrieval-augmented generation (RAG). I received my M.Sc. in Artificial Intelligence from NTU and my B.E. in Cyberspace Security from Wuhan University. My research interests include federated domain adaptation, secure federated learning, and long-context retrieval-augmented generation. I have published papers in venues such as ICLR, TKDE, and ACL. Previously, I collaborated with the Alibaba-NTU Global e-Sustainability Lab (ANGEL) on exploring the knowledge boundaries of multimodal large language models.

CV Scholar Github

Publications

Enhancing Federated Domain Adaptation with Multi-Domain Prototype-Based Federated Fine-Tuning

Enhancing Federated Domain Adaptation with Multi-Domain Prototype-Based Federated Fine-Tuning

Jingyuan Zhang, Yiyang Duan, Shuaicheng Niu, Yang Cao, Wei Yang Bryan Lim
ICLR, 2025

MPFT is a novel Federated Domain Adaptation framework that fine-tunes a pre-trained model using multi-domain prototypes enriched with domain-specific local data to significantly improve in-domain and out-of-domain accuracy while converging in a single communication round, reducing computational and communication costs, and ensuring privacy through differential privacy and robustness against feature space hijacking attacks.

Project Page PDF arXiv