An artificial neural network (also known as connectionist systems) is a computational model loosely inspired by the behavior observed in their biological counterpart, This consists of a set of units, called artificial neurons, connected together to transmit signals. Input information traverses the neural network.
Each neuron is connected to others through links. In these links the output value of the previous neuron is multiplied by a weight value. These bond weights can increase or inhibit the activation state of adjacent neurons. Similarly, at the output of the neuron, there may be a limiting or threshold function, which modifies the resulting value or imposes a limit that must not be exceeded before propagating to another neuron. This function is known as the activation function.
How can we improve this artificial neural network?
A neural network can be trained and this procedure consists of adjusting each of the weights of the inputs of all the neurons that are part of the neural network, so that the responses of the output layer fit as closely as possible to the data that we know. To do this machine learning, you typically try to minimize a loss function that evaluates the network as a whole. The values of the weights of the neurons are updated seeking to reduce the value of the loss function. This process is done by propagating back.
The goal of the neural network is to solve problems in the same way as the human brain, although neural networks are more abstract. Today’s neural networks typically contain from a few thousand to a few million neural units.
