Advanced Machine Learning and Statistical Inference Approaches for Big Data Analytics and Information Fusion

A revolution in Big Data Predictive Analytics has been created by the confluence of four major revolutionary technologies, viz. (i) availability of massive datasets, (ii) distributed cluster computing (iii) advances in non-parametric Bayesian Inference and (iv) Markov Chain Monte Carlo (MCMC) methods for fast probability calculations and stochastic searches in high dimensions. The paper presents a historical perspective on the seminal breakthrough developments related to probabilistic reasoning and Bayesian Inference leading up to the current state-of-the-art in data science. This is followed by a discussion of challenges in Big Data Analytics and presentation of a method for Automated Bayesian Machine Learning using a recently developed approach called CrossCat. This approach is based on non-parametric Bayesian Inference and efficient use of MCMC numerical algorithms. Under the DARPA XDATA program, SSCI, MIT and the University of Louisville are developing a Predictive Database System which will have an SQL type front-end and CrossCat backend to facilitate the use of sophisticated machine learning methods by non-experts.