Neural network architecture

ANN can be considered as a weighted directed graph in which artificial neurons are nodes. According to the architecture of connections, ANNs can be grouped into two classes (Fig. 2): forward-propagation networks in which the graphs do not have loops, and recurrent networks, or networks with feedback.

Figure 2. Systemization of the networks of direct distribution and recurrent (with feedback).

In the most common family of first-class networks, called the multilayer perceptron, the neurons are arranged in layers and have unidirectional connections between the layers. In fig. 2 shows typical networks of each class. Forward propagation networks are static in the sense that they produce one set of output values ​​for a given input that are independent of the previous network state. Recurrent networks are dynamic, because due to feedbacks in them the inputs of neurons are modified, which leads to a change in the state of the network.