Article Ultimate Knowledge Popular 2 min read

What are Artificial Neural Networks?

Artificial Neural Networks are mostly used for specific tasks in the field of deep learning.

Artificial Neural Networks copy the human brain.

A lot of it is in the name: Artificial Neural Networks (ANN, also called "perceptrons") are computing systems that loosely copy the structure of the mammalian brain. They are made up of "artificial neurons," a series of interconnected nodes. These neurons perform two tasks: they analyse data inputs, and they transmit signals for other neurons to process. Originally, ANN were designed to solve problems in a human way, but they've progressively moved on to more specific tasks.

Today, they power deep learning systems which allow machines to recognise images and voices, to drive cars, to handle customer relationships and to reach medical diagnoses, among many applications.

Information travels through layers of neurons

As we explained in our article on deep learning, artificial neural networks are organised in layers: the first layer processes a basic level of information on an image, for instance, and feeds that information to the second layer, which uses it to refine its understanding of the image and sends that refined signal to a third layer, and so on, until the machine comes up with an output (e.g. "This image contains a cat"). But the information does not travel in such a linear way in all ANNs. Less basic networks use a technique called backpropagation, which is essentially a feedback loop that allows them to know when an output doesn't match expectations (e.g. the ANN identified a mouse instead of a cat), and to adjust their layers of neurons accordingly.

A variety of Artificial Neural Networks

There are in fact many different kinds of ANNs: the simple feedforward neural network; the recurrent neural network, in which information travels in multiple directions; the convolutional neural network, most commonly used for image recognition; the Deep Boltzmann machine network, which can learn abstract internal representations that are useful for speech recognition -- to name a few. Depending on the task at hand, the desired type of ANN may vary. In fact, they may copy the structure of the brain, but artificial neural networks really are good at highly-specialised tasks. In that sense, they compliment us well.

Discover the power of

Get started

See in action

Watch a Demo