ELI5: what is deep learning?

What is deep learning? You’d be forgiven for being unclear on the mechanics.

As one of the more complex branches of AI (already an intricate topic), deep learning is a little-understood field. Though the term is often flung around as a buzzword, its meaning is lost on many of us.

So, here’s an ELI5 overview on deep learning — what it is, and how it works.

What is deep learning?

Deep learning is a subsection of machine learning, which is a type of AI technology. Specifically, it’s a type of machine learning that aims to teach computers to learn by example. And it needs masses of data to learn from.

Deep learning differs from other types of machine learning based on how it works. For a start, deep learning learns from data by itself. It’s the side of machine learning that uses unstructured (messy) data to draw its own conclusions.

So, for instance, deep learning will take thousands of images of lots of different things. Then, it must spot patterns to learn how to classify those images.

The idea behind deep learning, then, is for the computer to ‘learn’ independently. (Without humans telling it what to look for.)

Deep learning and artificial neural networks

Deep learning works by trying to copy the human brain. And to do that, it uses something called artificial neural networks (ANN). So, to answer the ‘what is deep learning’ question, it’s important to understand what an ANN is.

An artificial neural network is a computer system that’s inspired by the way the human brain works. They work a bit like a filter. (Rather than more linear reasoning — going through each process step-by-step).

Artificial neural networks have layers of ‘nodes’. These nodes each analyse the input and then communicate the results with the nodes in the next layer.

So, you give the ANN some data to process. That’s the input. Then, the nodes in the first layer all process that input. Their output then gets processed by the next layer of nodes.

This happens with each layer — allowing the machine to build on the previous layer’s output with added data. These layers are known as ‘hidden layers’.

Finally, you have the output — the answer that the ANN gives to your input. 

Back to deep learning

So, what is deep learning? It’s a massive artificial neural network.

Typically, an ANN only has a few (2-3 max) hidden layers for your input to filter through. Deep learning, meanwhile, has many. In fact, this high number of hidden layers is where the ‘deep’ label comes from.

In short, deep learning is a huge artificial neural network that lets machines learn by example. The more examples it gets, the better it performs.

This has proved useful in teaching machines to identify images and understand speech. Deep learning allows machines to identify key elements of the input, despite extra variables. (For example, snow on a stop sign, a hat on a person, or an accent in speech.)

This means it’s proving useful for tools such as virtual assistants, facial recognition and driverless cars.

Scratching the surface

This is just a scratch on the surface when it comes to the question ‘what is deep learning’. As with most of the terms and fields surrounding AI, you could write a series of books about deep learning and artificial neural networks.

But, if you don’t want to spend time conducting your own deep learning on the subject, simply think of it as a type of machine learning. One that works using lots of layers, to extract increasing ‘understanding’ of raw data over time.

Useful links

What is machine learning? A beginner’s guide

A history of automation: The rise of robots and AI

What is a rule-based system? What is it not?