Neural network classification is the process through which computers are able to classify data using circuitry motifs found in biological neural networks. The neural circuit is the most complex circuit known and is capable of more data processing — both in parallel and serially — than any computer in existence as of 2011, and one reason it is so powerful is the adaptive capability of neural circuitry. Connections, synapses and functional logic gates can strengthen and weaken based on previous information and neural firing rates. This same circuit adaptation to efficient information integration and data processing for statistical classification can be exploited by incorporation of these circuit motifs and methods in computer designs.
The brain is an impressive source of processing power combined with naturally occurring nonlinear statistical data modeling tools. The motifs found in neural circuitry vary from simple processing units to complex information integration systems. The biological adaptive systems in neural networks change how they process information based on previous information received. Much in the same way that a person learns to drown out background noise, an artificial neural network can learn to weight separate pieces of information differently, giving more weight to information that the system has learned to designate as "important."
Computational models for neural network classification take advantage of the knowledge gained from studying the naturally occurring neural network processing capabilities, from units within the circuitry to the process through which information is weighted. These motifs can then be better understood, lending further insight into the workings of the brain as well as recreating the motif in silico, which means that computers are used for greater statistical neural network classification processing power. The applications of neural network classification are far-reaching, but the progress made by 2011 had been relatively small, largely because of the nascence of the study and understanding of neural networks.
Information processing methods that incorporates adaptive systems, mimicking the motile circuitry in the brain, such as machine learning based on previous data parameters, allows scientists to process data in a unique and dynamic way. Some people say that this is necessary because data accumulation in science has been growing exponentially and techniques to increase the amount of data taken have been becoming more efficient. Many scientists believe that the bottleneck in scientific discoveries will be assimilation and data processing itself. For statistical analysis, non-linear, machine-based learning methods have become more heavily relied on.
Through a computational model, programmers can create artificial neural networks comprised of artificial neurons to represent a mechanical brain — an in silico information-processing unit that has the capability of altering its statistical analysis dynamically based on previous assessed data. Essentially, basing data processing on neural network classification might allow scientists to create more powerful machines that are capable of learning. A machine capable of dynamically modifying its classification abilities based on previous information is a powerful tool made to assist scientists in the problems produced by trying to analyze large amounts of data.