实现时空数据的共享流通及协同分析能够挖掘数据潜在价值、助力地理信息产业发展，但私有数据的隐私泄露抑制了时空数据的共享。为了在进一步推动时空数据共享程度、优化共享效果的同时兼顾参与方经济效益及合法权益，提出了一种时空数据隐私保护共享的群体学习(spatio-temporal data privacy preserving sharing swarm learning，STDPPS-SL)方法。首先，构建基于群体学习的多参与方时空数据共享网络，保护参与方数据所有权，实现数据内容不泄漏且参与方权益平等的时空数据共享;其次，提出基于t分布的差分隐私随机梯度下降算法，防止共享过程中因隐私泄露导致参与方私有数据保密性被破坏，从而造成参与方经济损失;最后，设计打分系统量化参与方可信程度，保证数据共享结果可信。理论分析证明，本文所提方法(STDPPS-SL)满足严格差分隐私，能够保护参与方的私有数据所有权。在公开数据集上的对比实验表明，该方法(STDPPS-SL)能够实现参与方隐私保护的时空数据共享，并且兼顾安全性与可用性。
Research on Spatio-Temporal Data Privacy Preserving Sharing Swarm Learning
Achieving the sharing and collaborative analysis of spatio-temporal data can explore the potential value of data and boost the development of geographic information industry,but the privacy leakage of private data inhibits the sharing of spatio-temporal data.In order to further promote the degree of spatio-temporal data sharing and optimize the sharing effect while taking into account the economic benefits and legitimate rights of the participants,this paper proposes a spatio-temporal data privacy preserving sharing swarm learning(STDPPS-SL)method.Firstly,a multi-participant spatio-temporal data sharing network based on swarm learning is proposed in order to protect the ownership of participant's data and enable the process of spatio-temporal data sharing without revealing the contents of the data,while keep the equal rights of participants.Secondly,a differential privacy stochastic gradient descent algorithm based on the t-distribution is proposed in order to prevent the confidentiality of the participant's private data from being destroyed due to the privacy leakage during the sharing process,and in order to avoid economic losses to the participants.Finally,a scoring system is designed to quantify the credibility of the participants,in order to ensure the credibility of the data sharing results.Theoretical analysis proves that the proposed method(STDPPS-SL)can protect the private data ownership of the participants by satisfying strict differential privacy.Comparative experiments on open datasets show that the proposed method(STDPPS-SL)is able to realize the spatio-temporal data sharing process with the protection of participant's privacy,and the method balances the security and usability.
data sharingspatio-temporal big dataswarm learningdistributed learning
万方数据
维普
