Secure Cross-device Federated Recommender Systems

The School of EECS is hosting the following PhD Milestone 3 Thesis Review Seminar:

Secure Cross-device Federated Recommender Systems

Speaker: Wei Yuan
Host: Prof Hongzhi Yin

Abstract: The demand for recommender systems has been rapidly growing across various online applications (e.g., e-commerce, short videos, and social media) due to their ability to effectively mitigate the problem of information overload. Typically, recommender systems are developed by service providers and trained on vast amounts of user data collected and stored on centralized cloud servers. However, this centralized paradigm poses significant privacy risks, leading to potential data leakage concerns.

To solve these privacy issues, researchers have integrated federated learning, a privacy-preserving training framework, into recommender systems, namely federated recommender systems (FedRecs), in which a central server coordinates users to train a recommender model while keeping their personal data confined to local devices. Due to the privacy-preserving advantages, FedRecs have garnered considerable attention in recent years. However, existing studies predominantly focus on improving performance and training efficiency, with limited exploration of security concerns in FedRecs.

This research highlights the complex security challenges in FedRecs arising from their open and decentralized nature. It offers a comprehensive investigation of FedRecs’ security from two perspectives: privacy protection and system robustness. Specifically, our research addresses the following three key topics (user data privacy, service model privacy, and system poisoning robustness), advancing the privacy-preserving and robust capabilities of existing FedRecs.

Bio: Wei Yuan is a Ph.D. candidate from the School of EECS at The University of Queensland under the supervision of Prof. Hongzhi Yin and Dr. Miao Xu. He received his master’s degree in software engineering at Nanjing University. His research interests include secure and trustworthy recommender systems, federated learning, and urban computing.