Artificial intelligence, neural networks, machine learning – new and highly popular expressions, but what really do they mean? A neural network itself is a system of many neurons (processors). These processors on their own are quite simple (even the PC processor is more complicated), but connected into a big system neurons can implement a variety of different challenges. Depending on the implementation field, there are many of neural network essence interpretations. For example, ANN in a machine intelligence implies methods of Pattern Recognition. In mathematics, it is a multiparameter task. From the perspective of cybernetics, it is a robotics adaptive managing model. An artificial intelligence needs ANN as the main – fundamental – component for a natural intelligence simulation with the help of computational algorithms.
For the first time the concept of artificial neural networks (ANNs) arose in connection with an attempt of the human brain process simulation. The McCullock-Pitts neural networks model creation of 1943 became the first and the major breakthrough in the sphere. For the first time scientists developed an artificial neuron model. They also suggested a network construction of the elements for logical operations implementation. However, what is the most important; scientists have proved the ability of those networks to learn.
A neural network can deal with several challenges at the same time, using a single set of input signals but having several outputs; also, it can predict the value of a number of indicators. This often helps the neural network to create more appropriate or universal “internal” – intermediate concepts (because it is required that all the intermediate calculations must be suitable not only for one but a number of tasks at the same time). As a result, it is possible to increase the accuracy of meeting the challenges as compared to an individual meeting.
A neural network can learn itself and find the way to deal with a challenge when an expert person may do it not accurately enough (it may be there is not even an expert for such a problem). Development of control systems and automated decision-making in challenging and uncertain situations based on different technologies such as expert systems, frame structures and fuzzy systems, neural networks. Smart technologies provide the opportunity of meeting management challenges that have no any rigid mathematical solution
Fast neural network learning system algorithms: it can be trained almost instantaneously on a normal computer even in case of hundreds of input signals or tens to hundreds of thousands of reference situations. Thanks to this ability, a neural network is available for any enterprises or plants. Here, the neural networks implementation is available for meeting a wide range of complex projection challenges, classifications and complete diagnostics.
Currently, artificial neural networks (ANN) became an important expansion of a calculation concept. They have already enabled to meet a range of intractable challenges. In addition, they can create new programs and systems, ANNs are able to deal with the challenges, which are out of human brain capacity.
Neural networks can be used in a wide variety of application areas such as text and speech recognition, semantic search, expert and decision support systems, stock prices prediction, security systems, text analysis. Neuro computers are very effective, especially in situations when there is a need for human intuition semblance. In particular, among such challenges one could include decision-making.
It took ITC more than a year and cost only $ 2.5 million to create a specialized neuro network system. Testing of the new system showed that a neuro network is able to detect 38% of fraud, while the expert system used before it gave only 14%.
Specialized soft program – robotics – is a very important scope for neuro networks, as far as these robots are designed not for physical work but for information processing. Intellectual assistants should make it easier for users to work with information and the whole process of communication with a computer.
In 1996, the firm Accurate Automation Corp., Chattanooga, TN by order of NASA and the Air Force initiated an experimental auto manned hypersonic reconnaissance aircraft LoFLYTE. LoFLYTE used neural networks that gave the automatic pilot an opportunity to learn by copying steering tricks of the pilot..
Neuro networks are actively applied in financial markets. For example, the American Citibank has been using the neuro networks’ predictions since 1990 and in two years, after they have been implemented the automatic dealing showed a yield of 25% per annum, according to the Economist magazine.
The Neural Innovation Ltd used a direct mail strategy while collaborating with marketing companies. At the beginning, it sent out only 25% of the total offers amount and collected information about the consumers’ responses and reactions.
The definition of text messages mainstreaming is one more example of the successful use of artificial neural networks. For instance, PointCast, Inc. selected the Convectis news server for automatic categorization of messages in 1997 as far as it was the leader of the personalized delivery on the Internet.