ABSTRACT
Big data and Internet of Things are playing indispensable roles in human's daily life. In the era of information, all things are linked and enhanced by the Internet, from original virtual data to daily essentials, known as Internet of Things, from national strategy to urban development, known as smart city. This chapter focuses on large-scale data processing methods suitable for big data analysis scenarios in smart city, mainly the large-scale machine learning. It explains several prevalent optimization methods reducing a concrete risk function. Two prevalent convex optimization methods and their parallelization are introduced, namely, the stochastic gradient descent (SGD) and the Newton method. The chapter presents the Petuum system, a specific parallel computing platform for machine learning. It introduces an architecture for decomposing optimization processing and summarises other parallel improvements. The chapter also presents a brief case study for an application in the smart city by using basic classification.
