190 Chapter 7 ConneCtionist Models
An artificial neural network is connectionist model of programming
using computers. An ANN attempts to give computers human-like abilities
by mimicking the human brain’s functionality. The human brain consists of
a network of more than a hundred billion interconnected neurons. Neu-
rons are individual cells that can process small amounts of information and
then activate other neurons to continue the process. If an ANN is success-
fully implemented, machines like personal computers can be used more
effectively in various areas of problem solving and decision making. Typical
artificial intelligence (AI) methodology deals with the symbolic represen-
tation of knowledge, whereas ANN models document knowledge in the
connection of the network. That is why it is called a “network. ANNs are
exceptionally good at performing pattern recognition and other tasks that
are difficult to program using conventional techniques. The main advantage
such models offer is the ability to learn on their own and adapt to changing
conditions.
Although neural networks are modeled on the human brain, their pres-
ent state is far from the realization of actual intelligence. In fact, we know
less about the human brain—the model for neural networks that actual-
ly generates intelligence. There are noteworthy differences in the physi-
cal characteristics of the human brain and neural networks. The number of
neurons in the brain is on the order of 10
11
. The total number of neurons in
a high-performance system is, at most, on the order of 10
5
when we allocate
a processor to each artificial neuron. The typical neural network has fewer
neurons—for example, order of 10s or 100s.
7.1.1 Advantages and Disadvantages of Neural
Networks
The major advantage of neural networks is that they are well suited for par-
allel implementation because every neuron can work independently. Neural
networks can deal with new patterns that are similar to learning patterns, so
they are suitable for generalization. Nonlinear problems are difficult to solve
theoretically, whereas neural networks can tackle any problem that can be
represented as a pattern. Moreover, neural networks can deal with a certain
amount of “noise” in the input. They can perform even if part of the neural
network is damaged to a certain extent. Neural networks are utterly comple-
ment symbolic artificial intelligence.
76473_CH07_Akerkar.indd 190 8/4/09 4:59:39 PM

Get Knowledge-Based Systems now with O’Reilly online learning.

O’Reilly members experience live online training, plus books, videos, and digital content from 200+ publishers.