Big Data Analytic Paradigms: From Principle Component Analysis to Deep Learning

Abstract

Large sets of data (numerical, textural and image) have been accumulating in all aspects of our lives for a long time. Advances in sensor technology, the Internet, social networks, wireless communication, and inexpensive memory have all contributed to an explosion of “Big Data,” as part of thrust in robust intelligence. Big data is created in many ways in today’s highly inter-connected world. Social networks, system of systems (SoS: complex interoperable) systems and wireless systems are only some of the platforms creating big data. Recent efforts have developed a promising approach, called “Data Analytics”, which uses statistical and computational intelligence (CI) tools such as principal component analysis (PCA), clustering, fuzzy logic, neuro-computing, evolutionary computation, Regression analysis, Bayesian networks, etc. to reduce the size of “Big Data” to a manageable size and apply these tools to (a) extract information, (b) build a knowledge base using the derived data, and (c) eventually develop a non-parametric model for the “Big Data”. This chapter attempts to construct a bridge between SoS and Data Analytics to develop reliable models for such systems. One of the recent most promising data analytic too is “Deep Learning”. Deep learning is the broad term for the recent development and extensions of neural networks in the machine learning community, which has allowed for state of the art results in speech, image, and natural language processing tasks. Hierarchical learning is an area of research which focuses on learning high order representations from low level data. Learning to recognize objects from images, recognizing words or syllables from audio, or recognizing poses and movement from video are all good examples of modern hierarchical learning research, which is a central focus of the “deep learning” movement in the machine learning and computational statistics community. This chapter will give a rather comprehensive look at all the BIG data analytic tools—old and new. Data bases of photovoltaic and wind energy (from US National Renewable Energy Laboratory) as well as transportation data will all be used here.