Artificial Neural Networks
(Software that Thinks)
Definition:
| Artificial Neural Networks (ANNs) use many simple
computer processors acting together to tackle computational problems too
complex for even large single processors. The processor units are connected
by communication channels which carry numeric but no symbolic data. Each
processor unit can work independently on its own local data and inputs
via the connections. The parallel processing and the high connectivity
are common representations of ANNs.
The DARPA Neural Network Study (1988, AFCEA International Press, p. 60) definition goes on to say "...the function is determined by network structure, connection strengths, and the processing performed at computing elements or nodes." This approach to computing also includes the characteristic of adaptive learning. This learning is evolved from the ability to extract meaning from examples and to generalize information from data that may be either imprecise or too complicated. Once the data is analyzed the ANN becomes an expert that can provide projections, analyze trends, or solve situations based on "what if" questions. The efficient information processing capabilities of ANNs allow for greater knowledge of complex problems and therefore improved decision making. |
History
| The idea of incorporating artificial
intelligence(AI) in machines has been around since man first developed
machines to perform tasks. The belief that artificial neural networks could
provide the most powerful way to harness the capabilities of computers
sprang up almost as soon as the first electronic computer. An early pioneer
in this field, von Neuman, made one of the first analogies between computing
and the operation of the human mind(a neural network). He likened the interaction
of networked computers and circuit elements to the flow of neurons in the
thought process of living animals.
In 1942, Norbert Weiner and his colleagues began formulating a set of related ideas that would later become known as "Cybernetics". The central theme of these ideas is the belief that biological mechanisms can be studied and modeled from an engineering and mathematical perspective. The underlying premise is the use of feedback as a learning method. Just as humans learn though experience and can draw on this experience to understand and interpret situations which may or may not be similar to previously faced situations, properly programed computers can learn from their past experiences without interference from external sources. |
Applications:
|
Examples:
| Several applications of ANNs
have proven very successful, among these are credit card fraud detection
and medical diagnosis. Neural networks help combat credit card fraud by
recognizing fraudulant use based on past charge patterns. A number of financial
institutions have deployed neural network based fraud detection systems
and found this technique has a much higher detection rate than other availabe
methods. It also appears to be relatively inexpensive when viewed against
loss savings. For instance, Mellon Bank's system paid for itself within
six months through realized savings.
Within the health care community, experiments with neural networks are being carried out in an attempt to improve decisions in medical diagnosis. An ANN can be shown a series of patient case histories with patient characteristics, symtoms and test results, along with given diagnoses from physicians attending the cases. The network can then recieve new patient information and make a diagnosis based on its knowledge. Such a system allows the expertise of an unlimited number of physicians to be called upon to give an immediate real time diagnosis. |
For Further Research Visit:
| Pattern
Recognition Technology
Introduction Neural Networks (with working example) Intoduction to Artificial Intelligence |
Ed Banks
Brad Laesch
Mark Sakowski
Neil Sawyer