PROJECT TITLE :

Fast and Secure Distributed Nonnegative Matrix Factorization

ABSTRACT:

The nonnegative matrix factorization (NMF) technique has been utilized effectively in a number of different Data Mining activities. Because of the high cost that it incurs when working with large matrices, there has been a recent uptick in interest in the acceleration of NMF. On the other hand, the privacy issue of NMF over federated data is worthy of attention, because NMF is commonly applied in image and text analysis, which may involve leveraging privacy data (for example, medical image and record) across several parties. This raises a concern because the privacy issue of NMF over federated data is worthy of attention (e.g., hospitals). In this paper, we investigate the issues surrounding distributed NMF's acceleration as well as its security. To begin, we present a distributed sketched alternating nonnegative least squares (DSANLS) framework for NMF. This framework makes use of a matrix sketching technique to cut down on the size of nonnegative least squares subproblems and guarantees convergence. Regarding the second issue, we demonstrate that DSANLS, after undergoing some modifications, is capable of being adapted to the security configuration, albeit only for a single or a limited number of iterations. As a result, we propose four efficient distributed NMF methods that can operate in synchronous as well as asynchronous contexts, all while providing a guarantee of data safety. In order to demonstrate the superiority of the methods that we have proposed, we conduct extensive experiments on a variety of real datasets.