Feedforward networks are a basic and essential class of network. This chapter has helped us study the building blocks of neural networks, and will help illuminate network topics going forward.
Feedforward neural networks are best represented as directed graphs; information flows through in one direction and is transformed by matrix multiplications and activation functions. Training cycles in ANNs are broken into epochs, each of which contains a forward pass and a backwards pass. On the forward pass, information flows from the input layer, is transformed via its connections with the output layers and their activation functions, and is put through an output layer function that renders the output in the form we want it; probabilities, binary classifications, so on. At the end of one of these training cycles, we calculate our error rate based on our loss function; how far...
