A neural network model imitates the connections between neurons in the brain. Neural networks can be applied to complex classification or regression problems. A neural network consists of layers of nodes, or neurons, as shown in the following figure. The first layer takes a set of input values and the final layer is the output values. The network can contain any number of intermediate or "hidden" layers. The input to each node (other than the first layer) is the combined weighted output from each node in the previous layer. Each layer also contains a bias unit, whose output is always 1. Each node then applies an activation function to its inputs to generate its output.
The output from the nodes in Layer 2 is given by:
where g(x) is the activation function. The output of the network shown in the preceding figure is:
A neural network is a supervised learning model. In the Teradata Aster implementation, the weights applied to each input are trained based on a training data set using a backpropagation algorithm. The initial weights can be supplied by the user; if none are supplied, a random set of weights is used. For more information about the backpropagation algorithm, see https://en.wikipedia.org/wiki/Backpropagation.
A Neural NetworkIn the preceding figure, the weights are shown as w ijk , where i is the network layer of the origin node, j is the number of the origin node in layer i, and k is the number of the destination node in layer (i+1). In the preceding figure, the input to the nodes in Layer 2 is: