Biomedical Engineering Reference
In-Depth Information
vector is necessary when the network is trained. An ANN of the supervised
learning kind, such as the multilayer perception, uses the objective result to
conduct the formation of the neural parameters. Hence to learn the behavior of the
process under study, it is possible to build the neural network.
In unsupervised learning, the working out of the network is totally data driven
and no target results for the input data vectors are provided. An ANN of the
unsupervised learning type, such as the self-organizing map, can be utilized for
clustering the input data and find features inherent to the problem. The ANNs are
computer programs based on a simplified model of the brain; they reproduce its
logical operation using a collection of neuron-like entities forming networks to
perform processing. ANN programs are multipurpose and with suitable training, a
single program could solve a number of problems [ 9 ].
A neural network required two physical components, (1) the processing ele-
ments (neuron) (2) the connection between the elements (links).Every link has a
load parameter associated with it.
Each neuron processes the information and produces the output from the
stimulus which they get from the neighboring neurons connected to it. A number
of methods are available in which information can be processed by a neuron and
various types of ways to connect the neurons to each other. For the connection of
neurons (with the help of using specific method) different types of neural network
structures are constructed by using different type of processing elements. Pattern
recognition, Control and signal processing, and many other types of applications in
a variety of neural network structures were developed.
There are a significant number of Pattern-Recognition algorithms used for e-
nose data processing and continuous developments are being made in this regard.
The basic requirement for neural networks is their ability to follow the brain's
pattern-recognition methods. The wide success of neural networks can be attrib-
uted to some key factors.
Neural networks, with their significant capability to derive meaning from
complicated or indefinite data, can be used to remove patterns and detect devel-
opment that are too difficult to be noticed by either humans or computer tech-
niques. A trained neural network can work as an ''expert'' in the class of
information it has been given for analysis. This skill can be used to provide
projections given new situations of interest and to answer ''what if'' questions.
Self-Organizing Map (SOM)
Unsupervised training is the methods in which the networks learn to establish
their own classifications of the training data without any external help. A self-
organizing map is a special type of neural network for clustering purposes pro-
jected by Kohonen [ 12 ]. The main aim of the self-organizing map (SOM) is to
transform a received signal pattern of arbitrary dimension into a two-dimensional
discrete map, and to do this transformation adaptively in a topological ordered
manner. These remarkable benefits proved it to be exceptionally successful for
data visualization applications. The connection between the input data as measured
Search WWH ::




Custom Search