Machine Learning - Lecture 1: Introduction to the Topic

Chris Thornton


What is Machine Learning?

Machine Learning (ML) is the use of data to acquire the rules for a desired behaviour.

Common tasks:

Is it to do with human learning?

Traditionally, ML has involved ideas about how human learning works.

But modern research is increasingly focussed on practical tasks.

What do we mean by `data'?

By `data' we mean sets of variable values, e.g.,

Datapoints

Values are organised in structures called datapoints.

Each datapoint combines a particular set of variables, e.g., age, salary and IQ specifically for the Informatics HoD.

Datapoints are also called vectors in neural-networks, and records in computer science.

A datapoint may also be called a datum.

Tabulation

Data are often presented in a tabulated form, with one datapoint per row, and one variable per column.

The relevant variable name often appears at the head of each column.

  NAME      AGE   SALARY   IQ

  smith     42    36K      130
  bloggs    29    30K      140
  bush      50    60K      120
  ...
A very common task in ML involves predicting one variable value from all the others.

Where this is the aim, it is usual to put the to-be-predicted variable last.

Basic data-types

Data are classified according to the number and character of variables involved.

Explicit and implicit structure

A dataset is a body of data, i.e., a collection of datapoints.

We will be interested in a dataset's structure.

But two meanings for `structure'.

Explicit structure = the actual values seen in the datapoints.

Implicit structure = patterns that are seen across the values.

Example: A-level grades

Dataset containing average A-level grades for the past ten years.

Explicit structure is the year and grade values.

We also see implicit structure---a gradual increase in values over time.

Various ways to model this implicit structure.

We could compute the difference between all years and then average.

This might reveal that grades increase by 0.3% per year on average.

Ways of using the model

The model could then be used for

Why machine learning now?

Machine learning is an increasingly central topic in informatics.

Real-world applications: learning consumer behaviour

Use of CCTV and automatic checkout machines in modern supermarkets enables detailed logs to be kept of purchases made, reductions on offer, counter locations etc.

These logs embody vast quantities of data and are therefore hard to analyse using traditional methods.

Machine Learning can be used to identify patterns in the data.

These may help identify potentially significant patterns of customer behaviour, enabling better management of the supermarket.

Cheese and ice cream

Modeling might reveal that increases in purchases of ice-cream tend to be accompanied by small reductions in purchases of cheese.

The supermarket could make use of this fact in manipulating sales of cheese and ice-cream.

Example: mining financial data

In this application, the data are price fluctuations and the aim is to extract regularities reflecting investment opportunities.

Modeling these patterns can reveal behavioural rules which increase profit.

For example, the discovery that sharp increases in the price of gold tends to be preceded by long periods of price stability might be the basis for an investment rule.

Fraud detection

Predicting fraudulant cases in credit-card transations

Summary

Questions