Privacy on the Edge: Customizable Privacy-Preserving Context Sharing in Hierarchical Edge Computing

• Published in: IEEE transactions on network science and engineering 2020-10, Vol.7 (4), p.2298-2309
• Main Authors: Gu, Bruce, Gao, Longxiang, Wang, Xiaodong, Qu, Youyang, Jin, Jiong, Yu, Shui
• Format: Article
• Language: English
• Subjects: Algorithms; Computational modeling; customizable privacy preservation; Data privacy; Edge computing; End users; Machine learning; Markov decision process; Markov processes; Peer-to-peer computing; Privacy; Quality of service; Reinforcement learning; SARSA reinforcement learning
• ISSN: 2327-4697; 2334-329X
• DOI: 10.1109/TNSE.2019.2933639

The booming of edge computing enables and reshapes this big data era. However, privacy issues arise because increasing volume of data are published per second while the edge devices can only provide limited computing and storage resources. In addition, this has been aggravated by new emerging features of edge computing, such as decentralized and hierarchical infrastructure, mobility, and content-aware applications. Although some existing privacy preserving methods are extended to this domain, the privacy issues of data dissemination between multiple edge nodes and end users is barely studied. Motivated by this, we propose a dynamic customizable privacy-preserving model based on Markov decision process to obtain the optimized trade-off between customizable privacy protection and data utility. We start with establishing a game model between users and adversaries based on a QoS-based payoff function. A modified reinforcement learning algorithm is deployed to derive the exclusive Nash Equilibrium. Furthermore, the model can achieve fast convergence by the reduction of cardinality from n to 2. Extensive experimental results confirm the significance of the proposed model comparing to the existing work both in terms of effectiveness and feasibility.