Machine learning has been a growing area of technology for the last few decades. Now computer scientists are able to write code that allows computers to learn to play games, identify objects in images, and much more besides.
WHAT IS AN ALGORITHM?
An algorithm is a sequence of instructions. For example, I might say: turn right, walk straight to the end of the road, turn left. This would be a simple real-life algorithm to get from place A to place B.
Normally we talk about algorithms in the context of computers. Here the instructions might be more like ‘calculate this’. Computer scientists are now able to design very complex algorithms that can ‘learn’.
WHAT IS A NEURAL NETWORK?
There are lots of different kinds of algorithms that can perform machine learning, but the most basic one is known as a neural network. These are inspired by the way the human brain works. This algorithm takes some data and analyses it, looking for patterns. This is called training data because the neural network ‘trains’ itself by looking at it.
HOW DO NEURAL NETWORKS LEARN?
Consider an example. We want our neural network to be able to identify pictures of handwritten numbers. The ‘training data’ is a set of these images, alongside information telling the computer which number each image actually shows.
The computer feeds these images into the neural network. Say it inputs a picture of a handwritten seven. The network would then output the number it thinks the picture shows, and compare this to the actual answer to see if they match. The network will then alter itself based on whether it got the answer right or wrong, and try again. The aim is to improve its accuracy each time.
After the neural network has done this with all of the data it is given, we say it has been trained.
HOW CAN WE USE THEM?
Once trained, a neural network can be used to look at input data where we don’t know what the output data, the ‘answer’, is. In the example above, we can now give the trained neural network an image of a handwritten number, without telling it what it should expect to see. It should be able to identify the number and tell us what it is.
CONCLUSION
Creating machines that are able to learn can have a big impact. One huge area is medicine. If a computer could be designed to read your symptoms and identify the problem, it could be invaluable in rural and isolated areas where it can be difficult to find a doctor. This is one of many possible applications of this rapidly expanding field.
