Bayesian Statistics and Marketing, 2nd Edition

8A Bayesian Perspective on Machine Learning

Abstract

Machine Learning (ML) has emerged in the last few decades as a practical, scalable and impactful discipline that offers practical solutions to estimation problems for large scale data. In this chapter we examine the interplay between ML and the Bayesian paradigm. In particular, we look at the isomorphisms in the two frameworks and discuss how novel ML ideas can be used to enrich the Bayesian toolkit.

8.1 INTRODUCTION

The topic of artificial intelligence (AI) pertains to the idea that machines can emulate, and perhaps at some point exceed, human intelligence. In order for machine to exhibit such intelligence it needs to learn how to accomplish tasks. This aspect of training a machine to learn and perform tasks is broadly defined as machine learning or ML. The ML toolkit is large and often encompasses a number of topics that we would traditionally associate with statistics and optimization. Having said that, there are a number of new constructs that are particular to ML that have had an impact both on academia and on practice. Some of these constructs are broad ideas (e.g., Regularization, Bagging, Boosting) while others are practical toolkits (e.g., Deep Learners, Random Forests). The field of ML is too broad to cover in this chapter so we will focus on a subset of these topics. Our particular goal will be to discuss the interplay between ideas from the Bayesian paradigm and some of these novel ML constructs.

Machine learning ...