A Secure and Efficient Isometric Feature Mapping Outsourcing Framework
Peng Yang, Shaohong Zhang, Lin Yong Zhou, Han Ding · 2022
Data mining outsourcing solutions have received much attention as an effective technique to obtain potentially useful information from massive data. However, privacy issues have become a major impediment to outsourcing data mining. In this paper, we try to guarantee data privacy in outsourcing Isometric Feature Mapping (ISOMAP). To achieve this goal, a secure and efficient ISOMAP outsourcing framework is proposed, which understands the underlying information while protecting the data. Specifically, the neighbor relationships in the low-dimensional space of ISOMAP are the same as the original data. And for classification and clustering downstream tasks, the accuracy of the model built on our framework is almost the same as that of the ISOMAP model based on centralized explicit data. In addition, the proposed framework can achieve almost the same performance as the original data, and customers can gain significant computational savings in efficiency from outsourcing. We show the validity of the proposed framework through theoretical analysis and experiments.