Deep Neural Network

Abstract

A particular form of network used in brain imaging, the neural network, is the topic of this chapter. Properties of neural nets are presented together with a detailed description of the structure of neurons and artificial neurons that comprise these networks. A neural net is a conceptual model based upon the human brain. It has an observed input layer that connects to a middle, hidden layer consisting of an unobserved number of synapses and neurons and leads to an observed output layer. There are different functions for measuring the degree or presence of activity between the neurons. The model can be trained in various ways, but usually a design and test set approach is used where the model is given an input and allowed to develop unsupervised and eventually compared to the output (which is often a group) to assess fit using various measures such as mean absolute error and root mean square error. Tips are given on how to Interpret these fit indices. The learning rate, epoch, and batch, which control the amount of learning input data used, are described. A worked example in R is at the end of the chapter looking at stress and anxiety levels in a child and its mother being used to predict the presence, or absence, of hyperactivity.