Adaptive Introduction (AI)

This chapter introduces the reader to the field of machine learning (ML). Many machine learning algorithms have been created, each suitable for a class of applications. These algorithms have parameters that need to be set if high performance is to be achieved on an application. Choosing the right algorithm and the right set of parameters involves human expertise. Linear and non-linear regression are simple approaches to ML. Given a property (e.g., hours slept before an exam) and an outcome (performance on an exam), these methods provide a formula that can be used to predict an outcome. Linear regression uses a straight line to approximate the data (distance above the line is roughly the same as below the line), while non-linear regression uses a curved or wiggly line. Both methods have limitations on their prediction accuracy. Two general categories of ML algorithms are introduced. In supervised learning you are given properties and outcomes – there’s a teacher who tells you right from wrong. With unsupervised learning, there’s no teacher. Similar cases are grouped together with the expectation that an outcome for one member applies to all. Critical to the success of ML algorithms is access to data – lots of high-quality data for training purposes.