Department of Computer Engg.,
Pantnagar,
Computer expert systems aim to go from the crisp binary
conventional control towards the wooly way in which humans think. Neural
networks go further by aiming to build models of the way human being thinks and
reaches conclusions. One of the very intriguing consequences of this approach
is that you end up with a control system, which works, but the designer doesn’t
know how precisely it does. The human brain operates on a totally different
principle as those of PCs. It is based on small individual analog processing
units called ‘neurons’. All of these neurons operate relatively slowly (at
about 100Hz) but there is vast Computing power because they are all operating
in parallel. Artificial Neural Network is a system loosely modeled on the human
brain. The field goes by many names, such as connectionism, parallel
distributed processing, neuro-computing, natural intelligent systems, machine
learning algorithms, and artificial neural networks. It is an attempt to
simulate within specialized hardware or sophisticated software, the multiple
layers of simple processing elements called neurons. Each neuron is linked to
certain of its neighbors with varying coefficients of connectivity that
represent the strengths of these connections. Learning is accomplished by
adjusting these strengths to cause the overall network to output appropriate
results. The first attempt to build an operational model of the neuron used the
simple binary comparator (known as binary decision neuron). Human intelligence
is acquired through education and learning. ANNs have also to be trained. This
is achieved by adjusting the weights and thresholds of the individual neurons
in the network. Neural networks are useful at solving problems where various
algorithms are used to achieve a solution, or the solution itself, is difficult
to define. Neural networks are more robust than conventional systems. Like the
brain they can continue to work with failures and tolerate bad, incomplete or
noisy data. This ability to plod on regardless can be advantageous in some
applications. One major disadvantage is that training is required and the
amount of training data can be large. Nevertheless, the ability of ANNs to deal
with poor, noisy and incomplete data has led to them being used in seismic
analysis, military radar recognition and medical diagnosis. Neural networks are
not replacements for conventional control systems. They are additional tools
that can be added to, and used alongside, more traditional methods. As aim is
to mimic the operation of the human brain to some extent, an obvious question
is are we building artificial intelligence. This is a long way to answer this
simple question. The brain has, it will be remembered, some 1011neurons
each with 104 connections. ANNs build today have to the order of 102perceptions
each with at most 10 connections. At present we are adrift by a factor of about
108, but we do not know where we will be in 100 year’s time.