This paper presents a scheme for privacy-preserving clustering in a three-party scenario, focusing on cooperative training of multivariate mixture models. With modern-day big data often collected and stored across multiple independent parties, preservation of private data is an important issue during cross-party communications when carrying out statistical analyzes of the joint data. We consider the situation where the data are horizontally distributed among three parties and that each data owner wants to learn the global parameters while data from other parties are kept private. The inter-party communications must not expose any information that may potentially disclose details of the private data, including how the data are partitioned across the parties. In addition, unlike most existing methods, the proposed scheme does not require a special trusted party to be involved. Clustering plays an important role in statistical learning and is one of the most widely used data mining methods. We shall illustrate our scheme using a Gaussian mixture model (GMM) based cluster analysis.
CITATION STYLE
Leemaqz, K. L., Lee, S. X., & McLachlan, G. J. (2017). Private distributed three-party learning of gaussian mixture models. In Communications in Computer and Information Science (Vol. 719, pp. 75–87). Springer Verlag. https://doi.org/10.1007/978-981-10-5421-1_7
Mendeley helps you to discover research relevant for your work.