This post is the first in a series on machine learning.
Machine learning is changing our world – at breathtaking pace. It has allowed computers to translate news articles as well as humans, beat humans at Go, detect malware and flu outbreaks, drive cars, make phone calls, and recommend what movie you might want to watch next.
This is all very well, but what is machine learning and how about some simple examples of machine learning in action?
Machine Learning is the task of training a computer system so that it can make useful predictions about new data.
For example, a system can be trained by giving it lots of pictures of cats and dogs and for each picture telling the system whether the picture is of a cat or a dog. Once trained, the system can be given a picture it hasn’t seen before and it will say whether it looks like a cat or a dog. This is an example of something called supervised learning.
Another example is to make a computer system play itself at a game like Chess or Go. When one side in the game outplays the other, the decisions that the winning side made are encouraged in future and those that the losing side made are discouraged in future. By playing many thousands of games, the system increases in skill, until it might become better than any human at the game. This example is a type of reinforcement learning.
As a final example, a system can be trained by inspecting large quantities of normal traffic on a network so it learns what “normal” looks like. It can then be deployed to a network within a data centre that might be running a cloud computing service. If an attacker breaches that network, the traffic generated by the attacker will look different from “normal” and the system should raise an alert. This example is called unsupervised learning.
In my next post, I’ll start describing some of the most fundamental concepts in machine learning.