Distilling at the Edge: A Local Differential Privacy Obfuscation Framework for IoT Data Analytics
Distilling at the Edge: A Local Differential Privacy Obfuscation Framework for IoT Data Analytics
复制标题
边缘蒸馏：物联网数据分析的本地差异隐私混淆框架
DOI：
10.1109/mcom.2018.1701080
复制
发表时间：
2018-08-01
影响因子：
11.2
通讯作者：
Zhang, Yaoxue
中科院分区：
文献类型：
--
作者：
Xu, Chugui;Ren, Ju;Zhang, Yaoxue
通过将数据处理从云端迁移到网络边缘，边缘计算已成为延迟敏感和上下文感知物联网数据分析的有前景的范例。然而，采用同态加密在边缘服务器实现数据保护和聚合的传统解决方案由于计算开销过大而不可行。如何在边缘计算中保护数据隐私的同时保证数据效用成为物联网数据分析极其重要的问题。在本文中，我们提出了一种用于物联网数据分析的本地差分隐私混淆（LDPO）框架，可以在边缘聚合和提取物联网数据，而不会泄露用户的敏感数据。我们首先介绍LDPO框架的架构和优点，然后介绍保证其性能的一些技术挑战。然后我们提出了 LDPO 框架的初步实现，并使用实际应用程序和数据集验证其在隐私保护级别和数据实用性方面的性能。最后展望了进一步研究的一些未来方向。
Edge computing has emerged as a promising paradigm for delay-sensitive and context-aware IoT data analytics, through migrating data processing from the cloud to the edge of the network. However, traditional solutions adopting homomorphic encryption to achieve data protection and aggregation at edge servers are infeasible because of their heavy computational overhead. How to preserve data privacy while guaranteeing data utility in edge computing becomes an extremely important problem for IoT data analytics. In this article, we propose a local differential privacy obfuscation (LDPO) framework for IoT data analytics to aggregate and distill the IoT data at the edge without disclosing users' sensitive data. We first introduce the architecture and benefits of the LDPO framework, followed by some technical challenges in guaranteeing its performance. Then we present a preliminary implementation of the LDPO framework, and validate its performance in terms of privacy preservation level and data utility using real-world apps and datasets. Some future directions are finally envisioned for further research.