ELI5: what is an artificial neural network?
During an investigation into AI and machine learning, you may have come across the term ‘artificial neural network’ (ANN).
As a core component of deep learning, ANNs are a complex method of computer processing. They’re not easy to understand.
Here, we peel back the complexity and answer the question in simple terms. What is an artificial neural network?
What is an artificial neural network?
An artificial neural network is a simulation of a biological brain, but for your computer. In other words, they’re a way that a machine can process data, that’s inspired by (though not identical to) human and animal brains.
An artificial neural network consists of a collection of connected ‘nodes’ known as ‘artificial neurons.’ These loosely resemble neurons in a biological brain. Each node can process the input, and, using a connection a bit like synapses in a biological brain, communicate their result with the other nodes.
These nodes are typically arranged into layers, known as ‘hidden layers’. Each layer of nodes communicates with the layer next to it. A typical artificial neural network will likely only have two or three hidden layers of nodes. But some big artificial neural networks exist, which have as many as 150 layers.
So, what is an artificial neural network? It’s a computer system of processing that loosely mimics the way humans and animals process information. But, instead of neurons and synapses, think nodes and layers.
What is an artificial neural network? Well, it’s not a rule-based system. Artificial neural networks don’t use task-specific rules or linear reasoning. That is, they aren’t programmed with explicit step-by-step instructions.
Instead, they’re trained using lots of data.
So, the ANN gets lots of input along with the answers it should come up with. For example, if you want the network to be able to identify cats, you will give it lots of pictures of cats, and tell it that the correct answer is ‘cat’.
The network then processes the input and recognises patterns. Then, when it gets fresh input, it compares it to the pattern it found. Over time, nodes get assigned ‘weight’, which means they’re deemed more, or less, important. This improves the accuracy of the output.
A bit like a filter
Instead, a good way to think of the artificial neural networks is to view them a bit like a filter.
So, you give the ANN input — something you want it to process. Each of the nodes in the first layer process that input, and then communicate the results to each node in the next layer. The results from that layer filter into the next layer, and so on. In this way, each layer builds on the layer before it. When all the hidden layers have processed the input, the machine gives you its answer.
When answering what is an artificial neural network, you could say it’s a system that runs your input through a kind of filter to produce an output. In other words, they don’t carefully go through each process step by step.
Peeling back the complexity
Asking ‘what is an artificial neural network’ invites a host of complex answers. Ones which can inundate you not only with technical terminology but biology-related jargon too.
Hopefully, this article has peeled away the complexity, and now you’re a little clearer on this fascinating topic.