面向群智感知隐私保护的联邦典型相关分析方法

A Federal-Based Canonical Correlation Analysis Approach for Privacy Preserving in Crowdsensing

群智感知数据往往携带了丰富的敏感信息,联邦学习技术的兴起为解决感知系统之间隐私安全的协作计算提供了一条可行途径.针对群智感知关联性探测过程中可能导致的隐私泄露问题,提出一种联邦场景下支持隐私保护的典型相关分析方法.该方法通过构造联邦特征并基于非线性随机耦合策略进行特征保护,将典型相关分析求解过程分解为两个完全独立的运算分别由联邦端和服务器端执行,在数据级、运算级和特征级上对感知数据进行多层次保护.实验结果表明,文章方法对联邦场景的适用性和感知数据的保护性都较好.

The crowdsensing data often carries a wealth of sensitive information.The rise of federated learning technology provides a feasible way to solve the privacy preserving problem of cooperative computing between crowdsensing systems.A canonical correlation analysis approach,preserving privacy and supporting federated scenarios,is proposed to solve the problem of privacy disclosure in the process of correlation detection on crowdsensing.By constructing federal eigens and preserving them based on nonlinear random coupling strategy,the proposed approach decomposes the solving process of canonical correlation analysis into two completely independent operations,which are executed on the federals and the server respectively.The privacy of crowdsensing is preserved at the data-level,operation-level and eigen-level effectively.Experimental results verify the effectiveness of the proposed approach in both of applicability to federated scenarios and ability to preserve crowdsensing.