Neural Nets - Overview

A simple feed-forward neural network has six input nodes, 4 hidden nodes and 2 output nodes. Networks having only two output nodes are commonly used when binary (yes/no or true/false) answers are desired.

Note that each node is connected to all the nodes at the next level. While this level of connectivity is not a requirement for neural nets, most neural nets start out with these connections by default. If a connection's weight drops to zero during the training of the neural net, then in effect the connection no longer exists.

"It's not what you know, it's who you know."

That's the analogous basis of neural network computing, where the connections between the nodes (known as neurons) is more important that the node itself. The connections are numerically weighted: the larger the weight, the more effect that one node will have on the node at the other end of that connection. A neural network typically consists of dozens to thousands of neurons, each usually connected to at least two other neurons.

Neural networks have widespread applications, many of which fit into the category of pattern recognition. The network can be trained to recognize patterns by altering the weights of the connections between its neurons. For instance, neural networks have been trained in such diverse tasks as recognizing blemishes in agricultural produce to identifying aircraft.