Neural Networks – Feedforward Process and Backpropagation Process
This chapter discusses three simple neural networks in order of complexity. Neural networks were originally designed to attempt to mimic the way neurons in the human brain interact to produce intelligence. The chapter concentrates on the structure of the artificial neural networks. It discusses what is called the feedforward process for these networks. This is the simpler part of the process. The chapter covers some terminology and also refers to layers of the network. A neural network is represented as a graph since it has nodes and edges. The learning in the neural network takes place through the process called backpropagation. Derivatives of the error are obtained with respect to weight. The chain rule tells that the derivatives with respect to upstream weights can be calculated by means of derivatives with respect to weights further downstream.