What is the basic architecture of an Artificial Neural Network (ANN)?

A standard ANN model consists of an input layer, which corresponds to the features of the data, an output layer, which corresponds to the target label, and hidden layers, which consist of everything found between the input and output layers. Each layer consists of a certain number of units, or nodes. In a fully connected network, a given unit in a hidden layer receives as input information from all units in the previous layer. For each unit, the weighted combination of the input received is transformed in some manner through an activation function.

It is standard for the input layer to contain the same number of units as data features, and for the output layer to be a single unit corresponding to the target label. In the case of multi-class classification, the output layer might contain as many nodes as classes. However, it is difficult to decipher exactly what is happening within the hidden layers, as if the network is deep enough, some layers are far from the input layer and are difficult to interpret in any manner relative to the original features. The number of hidden layers and units within each layer are hyper-parameters associated with the complexity of the network that are tuned in training/cross validation.