|
|||||||||||
|
|
||||||||
| Technology > Neural Networks |
|
Neural Networks 
Neural networks are adaptive systems for data processing
and analysis. They are mathematical structures that emulate some aspects
in the functioning of the human brain, reproducing such abilities
of the latter as the self-learning ability, the ability to formalize
and cluster unclassified information, the ability to independently
make forecasts based on historical data. The main difference between
neural networks and other methods such as, for example, expert systems,
is that neural networks do not need any already known model and build
this model themselves based on the information they have at their
disposal. That is why neural networks and genetic algorithms have
found application everywhere where forecasting, classification, and
management tasks need to be solved. In other words, they are used
in those areas of human activity where tasks are difficult to algorithmize
and require the permanent assignment of a group of qualified experts
or use of adaptive systems. A neural network receives input information and processes it the way it is done by the human brain. During this analysis, the network trains itself, acquiring experience and knowledge, and provides output information based on the previously acquired experience. The main task of an analyst who is using a neural network to solve a problem is achieving the most efficient architecture of the neural network, i.e. correctly selecting the type of the network, its training algorithm, the number of the neurons, and the types of the relations between them. This work cannot be done using any formalized procedures,it requires a profound understanding of the different types of neural networks, involves a great deal of research and analysis, and can be quite time-consuming. In solving tasks that are not or cannot be formalized, neural network models excel any of the traditional methods. However, use of neural networks is expedient if:
Advantages Neural networks are indispensable for data analysis, and, in particular, for the initial data analysis and selection, search for any missing facts or gross mistakes made by the decision-maker. Neural network analysis methods can be used independently or in an effective combination with the traditional methods most of which involve building a model based on an assumption or theoretical inference (for example, that the dependence being searched for is linear or that one of the variables has the normal distribution). The neural network approach does not involve any assumptions of the kind; it is suitable for both linear and non-linear complex dependences and especially effective in prospecting data analysis when the goal is to find out whether there are any dependences between the variables. The data can be incomplete, controversial, or even intentionally garbled. If there is any kind of connection between the input and output data, a neural network will detect it with the set degree of precision even if the traditional methods fail to discern this connection. Modern neural networks also provide some additional possibilities. They allow evaluating the relative importance of the different kinds of input information, reduce its volume without losing any significant data, discern the signs of an impending crisis situation, and so on. Application Neural networks have increasingly been used in business applications. In some areas, such as fraud detection or risk assessment, they are the undisputable leaders among all the methods used. Neural networks are increasingly being used in forecasting and marketing research solutions. The following is a list of the fields in which neural networks have proven their efficiency on practice: Financial Operations
|
    |