Prof. ZHANG Jun’s Research on Privacy-Preserving Distributed Learning Published in Nature Communications

Aresearch paper on “Selective Knowledge Sharing for Privacy-Preserving Federated Distillation Without a Good Teacher”, co-authored by Prof. ZHANG Jun, Electronic and Computer Engineering (ECE), and his former PhD student Dr. SHAO Jiawei, was published in leading multidisciplinary journal Nature Communications.

The paper investigates the important topic of privacy-preserving distributed learning. Their proposed method allows multiple clients (e.g. banks, hospitals) to collaboratively train a machine learning model without sharing private data. Compared with existing works, their method substantially reduces communication overhead during the training process, enhances privacy protection, boosts model performance, and allows clients to adopt different models.

The first author, Dr. Shao Jiawei, finished his PhD studies in early 2024 and is now a postdoctoral fellow in Prof. Zhang’s team. Their work was collaborated with Microsoft Research Asia.

Related Link:

HKUST School of Engineering


What to read next