Architecture of ANNs

The units present in each layer and the connections between the units of the different layers define the architecture of an ANN, as shown in the following diagram:

In the previous diagram we can identify three layers: input, hidden, and output. The layers exchange information with each other through connections. Each connection has a weight associated with it that can be tuned based on experience, making the ANNs adaptive to inputs and capable of learning. The input-output link, in other words, the network transfer function, isn't programmed but is simply obtained by a learning process based on empirical data obtained during the training phase. A neural network is structured using a layered approach, which is the same approach used by the Keras framework.