Privacy Preserving Collaborative Computing: Heterogeneous Privacy Guarantee and Efficient Incentive Mechanism

Abstract

Collaborative computing uses multiple data servers to jointly complete data analysis, e.g., statistical analysis and inference. One major obstruction for it lies in privacy concern, which is directly associated with nodes’ participation and the fidelity of received data. Existing privacy-preserving paradigms for cloud computing and distributed data aggregation only provide nodes with homogeneous privacy protection without consideration of nodes’ diverse trust degrees to different data servers. We propose a two-phase framework that computes the average value while preserving heterogeneous privacy for nodes’ private data. The new challenge is that in the premise of meeting privacy requirements, we should guarantee the proposed framework has the same computation accuracy with existing privacy-aware solutions. In this paper, nodes obtain heterogeneous privacy protection in the face of different data servers through one-shot noise perturbation. Based on the definition of KL privacy, we derive the analytical expressions of the privacy preserving degrees (PPDs) and quantify the relation between different PPDs. Then, we obtain the closed-form expression of computation accuracy. Furthermore, an efficient incentive mechanism is proposed to achieve optimized computation accuracy when data servers have fixed budgets. Finally, extensive simulations are conducted to verify the obtained theoretical results.