Published by: Zeng Xiansen Edited by: Wang Xiangyu
WUST News (Correspondent Zeng Xiansen) - A research team led by Professor Lu Jianfeng from the School of Computer Science and Technology at Wuhan University of Science and Technology (WUST) has achieved a significant milestone with their latest paper, PIECE: Incentivizing Personalized Privacy-Preserving for Multi-Version Model Marketplace in Federated Learning, published in IEEE Transactions on Information Forensics and Security (TIFS), a top-tier international journal in cybersecurity. This marks WUST's first publication in TIFS, representing a major breakthrough in the field.
The first author of the paper is Professor Lu Jianfeng (his third TIFS paper as lead author), and the second author is Huang Tao, a 2025 master's graduate from the class of 2025 (now pursuing a Ph.D. at Central South University's School of Computer Science and Engineering). The School of Computer Science and Technology is the primary institution for the paper.
Federated learning, renowned for its collaborative training and privacy-preserving features, shows great potential in advancing the model marketplace. However, challenges such as insufficient training data and market arbitrage hinder its progress. To address these issues, the team proposed PIECE, a personalized privacy-preserving incentive mechanism that integrates sample-level differential privacy techniques and version control strategies. By designing tailored privacy rules and pricing mechanisms, PIECE regulates data owners' privacy strategies and mitigates arbitrage behaviors among model purchasers.
The study frames the market objective as a bi-objective optimization problem that balances social utility and model performance. Through a "privacy choice game," the team analyzed the dynamic relationship between local and global privacy demands, leveraging Nash equilibrium conditions to transform the problem into a socially optimal solution under no-arbitrage constraints. A two-stage pricing strategy based on subadditivity condition relaxation was introduced to achieve this goal.
Experimental results show that PIECE can significantly enhance the operational efficiency of the model marketplace. Under a specified market scale, this mechanism can increase model revenue by at least 8% and improve model performance by up to 16.67% compared to existing advanced baselines, offering important theoretical and technical support for the practical application of federated learning model marketplaces.
IEEE Transactions on Information Forensics and Security is ranked as an A-class journal by the China Computer Federation (CCF), the Chinese Association for Artificial Intelligence (CAAI), the Chinese Association for Cryptologic Research (CACR), and the Chinese Association of Automation (CAA). It is also classified as Q1 journal by the Chinese Academy of Sciences, covering cutting-edge topics like data privacy protection, network security, multimedia security, cryptography applications, and biometric recognition. It is renowned for its rigorous review process and high academic standards, with published results representing the international forefront of the cybersecurity field.
Professor Lu Jianfeng specializes in federated learning, crowdsourced computing, and game theory, achieving a series of influential outcomes.To date, he has published over 60 papers as the first or corresponding author in prestigious venues, including TIFS, JSAC, TMC, TSC, IJCAI, AAAI, TOIT, TII, TVT, TCSS, TCE, and TETCI, with 15 CCF-A class papers and over 20 IEEE/ACM Transactions series publications.
