What are Neural Networks ?
Artificial neural networks are distributed parallel systems
composed of a number of processing elements that compute
simple mathematical functions. Neural networks are ``inspired'' in some properties
and characteristics of the human brain and learn to solve problems through
a process of learning. The network units are linked to each other through connection
weights which store the network knowledge about the environment.
The steps that describe the operation of a neural network during learning can be
basically described as follows :
- A stimulus or input signal is presented to the network
- Each processing unit computes a function and produces an output which is a
result of the influence that the stimulus has on that specific unit
- The unit outputs are propagated forward until they reach the output layer
of the network, thus producing a network output to the input stimulus.
- The network weights are modified through a learning algorithm in order
to improve its response to the input patterns.
An Example
The figure below illustrates a 2-D view (left) and a 3-D view (right) of
a neural network in a character recognition application.
The units on the left (input units) are used to represent the characters (input
signal) that are presented to the network. The units on the rightmost side are
ouput units where each one of them represents a class of alphanumeric characters.
During network use (after training) when a pattern is fed into the network such as
the character ``I'' in the figure, the corresponding unit representing that class
provides the strongest response at the output layer.
Why Neural Networks ?
Neural networks are a powerfull technique to solve many real world problems. They
have the ability to learn from experience in order to improve their performance and to
adapt themselves to changes in the environment. In addition to that they are able
to deal with incomplete information or noisy data and can be very effective especially
in situations where it is not possible to define the rules or steps that lead
to the solution of a problem. The advantages and attractive characteristics
of neural networks can be summarized as follows:
- The ability to learn from examples and to generalize from the training
patterns providing appropriate outputs to stimuli not seen during learning.
- Natural parallelism.
- Distributed representation of information.
- The ability to cope with incomplete information or fuzzy data.
- Neural networks can be used in many problems for which there are no rules
or steps that leads to a solution.
- They do not require a priori information about the environment.
- The solutions developed for a problem can be easily extended to other applications.
- They can deal with a large amount of data and automatically discover complex relationships
between them.
Application Areas
Neural net have been used in such a diverse range of practical applications that
it is almost impossible to find an area where they have not ben applied nor have their
potential use being investigated.
Large companies such as UK High Street Banks, Visa, Mastercard, American Express, European
Bank, USA Bank, Asia Pacific Institution, UK Brokers, Britvic Soft Drinks are just a
few examples of the corporations that are investing in the neural network technology.
A recent article in The Economist points
out that world sales and business involving the technology of neural networks will
reach the amount of R$ 1 billion dollars in 1997. Below you find just a few
example areas where neural networks have been applied.
- Adaptive system monitoring (fault detection, plastic explosives detection)
- Finance and Business (credit scoring, consumer behaviour, stock selection, fraud detection)
- Medical diagnosis (heart problems, cancer detection)
- Quality control and robotics (visual inspection, robot control)
- Pattern recognition and classification (speech, face, fingerprints, handwriting)
(go back)
