Neural processing originally referred to the way the brain works, but the term is more typically used to describe a computer architecture that mimics that biological function. In computers, neural processing gives software the ability to adapt to changing situations and to improve its function as more information becomes available. Neural processing is used in software to do tasks such as recognize a human face, predict the weather, analyze speech patterns, and learn new strategies in games.
The human brain is composed of approximately 100 billion neurons. These neurons are nerve cells that individually serve a simple function of processing and transmitting information. When the nerve cells transmit and process in clusters, called a neural network, the results are complex – such as creating and storing memory, processing language, and reacting to sudden movement.
Artificial neural processing mimics this process at a simpler level. A small processing unit, called a neuron or node, performs a simple task of processing and transmitting data. As the simple processing units combine basic information through connectors, the information and processing becomes more complex. Unlike traditional computer processors, which need a human programmer to input new information, neural processors can learn on their own once they are programmed.
For example, a neural processor can improve at checkers. Just like a human brain, the computer learns that certain moves by an opponent are made to create traps. Basic programming might allow the computer to fall for the trap the first time. The more often a certain trap appears, however, the greater attention the computer pays to that data and begins to react accordingly.
Neural programmers call the increasing attention that the computer pays to certain outcomes “weight.” Traditional processing would provide the computer only with the basic rules of the game and a limited number of strategies. Neural processing, by gathering data and paying greater attention to more important information, learns better strategies as time goes on.
The power of neural processing is in its flexibility. In the brain, information is presented as an electrochemical impulse – a small jolt or a chemical signal. In artificial neural processing, the information is presented as a numeric value. That value determines whether the artificial neuron goes active or stays dormant, and it also determines where it sends its signal. If a certain checker is moved to a certain square, for instance, the neural network reads that information as numeric data. That data is compared against a growing amount of information, which in turn creates an action or output.