A new distributed and decentralized stochastic optimization algorithm with applications in Big Data analytics

Abstract

The world is witnessing an unprecedented growth of needs in data analytics. Big Data is distinguished by its three main characteristics: velocity, variety and volume. An open issue and challenge faced by the data community is how to scale up analytic algorithms. To address this issue, optimization of large scale data sets has attracted many researchers in recent years. In this paper, we first present the most recent advances in optimization of Big Data analytics. Further, we introduce a fully distributed stochastic optimization algorithm for decision making over large scale data sets. We also propose the optimal weight design for the proposed algorithm and study its performance by considering a practical application in cognitive networks. Experimental results confirm that the proposed method performs well, proven to be distributed, scalable and robust to missing data and communication failures.