Summary of this chapter

This first chapter has been intended to stimulate some reflective thought on the nature of the problems faced in Pattern Recognition as a prelude to getting into the technicalities. You were invited to put your feet up (except for a short spell in the kitchen), and I hope you did.

The problem of telling men from women by wholly inappropriate measurements of the wrong things is a depressingly accurate paradigm of much pattern classification, and the methods somewhat sketchily outlined are, in the main, those currently in use.

This chapter has surveyed the general ideas used in pattern recognition by looking at very simple examples of the general methodologies. The basic issue of what you measure and how you represent it was discussed in a rather unsatisfactory way, but it was argued that coding objects to be discriminated as points in ${\fam11\tenbbb R}^n$ has more power than alternatives.

Given this choice of coding, pattern classification devolves into the choice of what to measure, which is currently something of a black or at best grey art, and then finding algorithms which can assign a category to a new point. These fall into three major classes, metric, neural net and statistical. Examples of each were sketched, and some rationales for each were discussed. It was pointed out that unsupervised learning was a matter of finding clusters in a space and supervised learning was a matter of fitting a function from a family of functions to a set of data points on which the function values are known.

The problem of dynamic patterns, that is to say patterns with a temporal ordering on them was mentioned in two cases, the case of trajectories in ${\fam11\tenbbb R}^n$, as in speech and on-line character recognition, and the case of trajectories in a space of discrete symbols to which the former problems might be reduced.

The problem of more general relationships between points or symbols was mentioned, and disparaging things were said about philosophers, AI practitioners and the Fuzzy Folk. Many promises were made that the ideas outlined rather sketchily would be explained more fully in subsequent chapters.

Various provocative remarks were made, not intended to irritate you, but to get you to state your position in an email message to me. If you think I am mistaken about something, then it would be a kindness to point out to me the error of my ways. You are doing this course in order to become even better educated than you already are, and the art of putting your case lucidly is developed by practice.