Research On Personalized Privacy-preserving Data Trading Algorithms For Crowd-based Cooperative Computing

Crowd-based cooperative computing has emerged as a means to maximize the crowd intelligence generated by the collaboration of numerous mobile crowd sensing devices in sensing and computing tasks.Nevertheless,participants’ privacy is a pressing concern due to the possibility of data privacy breaches resulting from malicious nodes during data collection and transmission.This thesis focuses on the problem of sensing data trading and model parameter sharing based on personalized privacy preserving in crowd-based cooperative computing,and designs a series of models based on contract theory to organize the data trading process between the demand side and the execution side of crowd cooperative tasks.The main contributions of this thesis are summarized as follows.1.Constructing a personalized privacy-preserving data trading process and proposing an aggregation algorithm to maximize data utility.A contract-based mechanism is presented to address the personalized privacy preserving data trading in crowd-based cooperative computing.The proposed method provides a set of optimal contracts specifying different privacy-preserving levels and data trading prices for selfish data owners.Through rigorous theoretical analysis and extensive experimental validation,the proposed sensing data trading method not only achieves the desired data utility in terms of(α,δ)-accuracy,but also satisfies the properties of individual rationality,incentive compatibility,and budget feasibility.2.Proposing a federated learning model parameter sharing method based on contract theory and local differential privacy.This thesis specifically investigates how to strike a balance between model accuracy and the personalized privacy requirements of each mobile crowd sensing device.A local differential privacy mechanism and a contract-based incentive mechanism are incorporated into the federated learning process.On the one hand,we design a local differential privacy mechanism for each participant in the federated learning parameter sharing process to meet their personalized privacy preserving needs.On the other hand,with the goal of maximizing the accuracy of parameter aggregation results,a set of optimal contracts is designed to achieve the trade-off between personalized privacy control and federated model performance.Finally,comparison experiments are conducted on synthetic dataset and MNIST dataset,respectively.3.Constructing a data trading process that is trustworthy for both data trading platform and participants.This thesis constructs a trustworthy data trading process through a decentralized data trading platform based on blockchain technology,which addresses the risks posed by untrustworthy centralized platforms,such as single point of failure and privacy breaches.The paper also designs a series of smart contracts to prevent fraudulent behavior from participants during the data trading process.In summary,this thesis proposes a method to address the personalized privacy preserving problem of sensing data trading and model parameter sharing in crowd-based cooperative computing,with theoretical analysis and simulation experiments conforming its feasibility and performance advantages.Furthermore,the decentralized and trustworthy data trading platform based on blockchain technology ensures that the data trading process is reliable and sustainable in the long term.