On privacy protection in distributed computation protocols
MetadataShow full item record
With the blooming development of modern communication and networking technologies (e.g. Cloud computing, 3G/4G communication network, etc.), people are changing their ways how they collect data, store it, use it and share it. An increasing amount of individuals and corporates have transferred their data from local machines to online storages maintained by large companies such as Google, Amazon, etc.. Meanwhile, more and more comprehensive projects (e.g. distributed data mining, social network, electronic health system, etc.) that involve more than one party (and its data) have been put to practical use. On one hand, without doubt, these changes vastly improve the efficiency of data utilization, and benefit people's daily lives. On the other hand, these changes cause people's growing privacy concerns. Especially when the data contains private information of the data's owner, storing the data online or participating projects as above increases the risk of data disclosure and thus privacy violations. In this thesis, six widely-used distributed computation protocols and their privacy issues are studied. To protect the privacy of participants in these protocols and to get correct computation results at the same time, privacy-preserving implementations of these protocols are proposed. In particular, this thesis consists of: 1) a privacy-preserving neural network ensemble learning protocol in the semi-honest security model that allows two parties to jointly train a neural network ensemble with the aggregation of their vertically partitioned data, 2) a privacy-preserving perceptron learning protocol in the malicious security model that allows two parties to jointly train a perceptron with the aggregation of their vertically partitioned data, 3) an emergency access system for the online PCHR system that provides stable and secure emergency access functionality, 4) an efficient secure solution to a generalized version of the Yao's Millionaire problem 5) three different secure solutions to the optimal gateway selection in multi-domain wireless networks, 6) a privacy-preserving distributed permutation test that allows two parties to securely perform the test on the aggregation of their vertically partitioned data.