Privacy Preserving High Order Expectation Maximization Algorithm for Big Data Clustering with Redundancy Removal

Abstract

Cloud computing has become increasingly prevalent, providing end-users with temporary access to scalable computational resources. At a conceptual level, cloud computing should be a good fit for technical computing users. A heterogeneous cloud, on the other hand, integrates components by many different vendors, either at different levels (a management tool from one vendor driving a hypervisor from another) or even at the same level (multiple different hypervisors, all driven by the same management tool).Nowadays, a large number of heterogeneous data, often referring to big data, is generating from big storage, which requires novel models and technologies to process, especially clustering based computing, for the further promotion the design and applications of big data analytics. However, the heterogeneous data is usually very complex, which is composed of structured data and unstructured data, such as picture, text, pdf and video. In other words, the heterogeneous data contain multimodal between which there are nonlinear relationships. In the existing work, proposed a high-order possibilistic c-means algorithm by extending the conventional possibilistic c-means algorithm from the vector space to the tensor space for multimedia heterogeneous data clustering. Furthermore, employed cloud computing to improve the clustering efficiency for massive heterogeneous data. To protect the private data during clustering on cloud, proposed a privacy-preserving expectation maximization algorithm by using the asymmetric encryption scheme to encrypt the original data. The existing BGV scheme does not support the division operations and exponential operations that are used in the membership matrix updating function of the high-order fuzzy c-means algorithm. To address this problem, use the asymmetric encryption scheme to approximate the membership matrix updating function to a polynomial function. Cloud computing has become increasingly prevalent, providing end-users with temporary access to scalable computational resources. At a conceptual level, cloud computing should be a good fit for technical computing users. A heterogeneous cloud, on the other hand, integrates components by many different vendors, either at different levels (a management tool from one vendor driving a hypervisor from another) or even at the same level (multiple different hypervisors, all driven by the same management tool).Nowadays, a large number of heterogeneous data, often referring to big data, is generating from big storage, which requires novel models and technologies to process, especially clustering based computing, for the further promotion the design and applications of big data analytics. However, the heterogeneous data is usually very complex, which is composed of structured data and unstructured data, such as picture, text, pdf and video. In other words, the heterogeneous data contain multimodal between which there are nonlinear relationships. In the existing work, proposed a high-order possibilistic c-means algorithm by extending the conventional possibilistic c-means algorithm from the vector space to the tensor space for multimedia heterogeneous data clustering. Furthermore, employed cloud computing to improve the clustering efficiency for massive heterogeneous data. To protect the private data during clustering on cloud, proposed a privacy-preserving expectation maximization algorithm by using the asymmetric encryption scheme to encrypt the original data. The existing BGV scheme does not support the division operations and exponential operations that are used in the membership matrix updating function of the high-order fuzzy c-means algorithm. To address this problem, use the asymmetric encryption scheme to approximate the membership matrix updating function to a polynomial function.