PROJECT TITLE :

Cloud-Based Outsourcing for Enabling Privacy-Preserving Large-Scale Non-Negative Matrix Factorization

ABSTRACT:

It is inescapable and self-evident that clients with limited resources will find it necessary and obvious to outsource complicated and labor-intensive tasks to public cloud vendors in order to achieve their primary goal of reducing costs. Unfortunately, it's rare to be able to trust the companies that provide public cloud services. They could leak the data accidentally, misuse the user's data, compromise the user's privacy, or intentionally corrupt computational results in order to make the system unreliable. All of these things could happen. It is therefore important to find a way to prevent this from occurring while taking advantage of the computational power offered by public cloud vendors. Non-negative matrix factorization, also known as NMF, is a significant method for reducing the dimensions of data that has been utilized extensively in the field of large-scale data processing. NMF, however, cannot be conducted efficiently using local computation resources because of the non-polynomial hardness of the problem. This is especially true when dealing with large amounts of data. Motivated by this issue, we present a novel outsourced scheme for NMF (O-NMF), which aims to lessen the computing burden of clients and tackle secure problems faced by outsourcing NMF. This is done in response to the motivation provided by this issue. In particular, O-NMF uses Paillier homomorphism as a means to protect data privacy by relying on two servers that do not collude with one another. In addition, O-NMF enables a verification mechanism that clients can utilize for assistance in verifying the accuracy of results that have been returned. This work also provides a security analysis and experimental evaluation, both of which demonstrate that O-NMF is both valid and practical. Both of these aspects are provided in this work.