What are neural networks?

Artificial neural networks are a model inspired by the functioning of the human brain. It is formed by a set of nodes known as artificial neurons that are connected and transmit signals to each other. These signals are transmitted from the input to generate an output.

What is the purpose of neural networks?

The main objective of this model is to learn by automatically modifying itself so that it can get to perform complex tasks that could not be done through the classic rule-based programming. In this way you can automate functions that initially could only be performed by people.

How do neural networks work?

As mentioned, the functioning of the networks resembles that of the human brain. Networks receive a series of input values ​​and each of these inputs reaches a node called a neuron. The neurons of the network are in turn grouped into layers that form the neural network. Each of the neurons in the network has a weight, a numerical value, which modifies the input received. The new values ​​obtained leave the neurons and continue their way through the network. This operation can be seen schematically in the following image.

Once the end of the network has been reached, an output is obtained which will be the prediction calculated by the network. The more layers the network has and the more complex it is, the more they will also complex the functions that it can perform.

Training of neural networks. Backpropagation or backward propagation

To get a neural network to perform the desired functions, it is necessary to train it. The training of a neural network is carried out by modifying the weights of its neurons so that it can extract the desired results. For this, what is done is to enter training data in the network, depending on the result obtained, the weights of the neurons are modified according to the error obtained and depending on how much each neuron has contributed to said result. This method is known as Backpropagation or backward propagation. With this method it is possible that the network learns, getting a model capable of obtaining very successful results even with data very different from those that have been used during your training.

Although currently its use has become popular, neural networks exist since the 1950s. However, the low power of the equipment of that time and the absence of algorithms that allowed networks to learn in a way Efficiently caused them to stop being used. It has been subsequently, thanks to the creation of the Backpropagation algorithm, the use of GPUs that allow large optimizations for this type of calculations and the greater number of data available for training, when the neural networks have resurfaced and gain prominence in various fields Thanks to these improvements, the emergence of Deep Learning has been possible, which is based on the use of deep neural networks, that is, networks formed by a large number of layers for complex tasks.

Functions of neural networks

The scope of the functions of the neural networks is very wide, due to their operation, they are able to approximate any existing function with sufficient training. Mainly the neural networks are the modifications for prediction and classification tasks. Their range of action is wide and very useful today, they will not only be used for Industry 4.0 applications (recognition of parts and defects that have not been previously introduced for example), if they are not used children in other areas such as economics , in which they can help predict how much prices will vary over the years, or even in medicine where they are helpful in diagnosing various health problems.

Neural networks have become a key piece for the development of Artificial Intelligence, it is one of the main fields of research and the one that is evolving over time, each time more complex and efficient solutions.