Need for Neural Approach

A modern computer can perform millions of calculations in a fraction of a second. The computer is also capable of executing the programmed tasks much more efficiently than a human being. However, the general operating domain of a computer is greatly constrained by the so-called lack of intelligence. Also, the current computing paradigm has not lent itself to effective parallel utilization.

Because of this inherent inflexibility, many tasks that seem trivial to human beings have turned out to be overwhelmingly complex for computers. Instead, many seemingly complex tasks, such as mastering chess, have been computerized successfully. Today, a computer with the most modern chess program can beat the best human chess player.

Traditionally, computers have been programmed by humans. Even to this date, the trace of intelligence within the current computer software is the result of the programmers' efforts only. During the laborious design phase of a computer software, the future operation of the software is determined. A certain set of operations is defined for each input that the program is expected to receive. This heuristic approach can be very effective within the problem area to which it is appplied. However, it is usually a very difficult and time-consuming process to update the heuristic rules as the problem changes. In addition, a totally distinct set of heuristic rules is needed for every problem.

If, by mistake, the program gets an input it has not programmed to receive, the consequences are most likely disastrous or at least unexpected. That is why the software designers try to cover all the states and inputs the program can run into. One major purpose of a graphical user interface is to limit the number of inputs that the user can give to the program.

In contrast to heuristic approach, various algorithms have been found to be a more efficient means of solving computational problems. Although any one algorithm is capable of modelling only a small fraction in a functioning whole, sometimes a single algorithm can solve an entire range of problems. An example of this is the minimax algorithm which has found its way to drive practically all types of board games. The common denominator in most board games is that the player strives to maximize the score for itself and minimize the score for the opponent. This is exactly the problem the minimax algorithm can tackle.

It seems clear that nobody can build fixed, universally applicable heuristic rules. This would mean that the computer is programmed so that it takes into account all the possible situations that 'there is' in order to face problems of any kind. Also, there obviously exist no algorithms which could be presented any kind of problem and be expected to come up with an answer to that. This inevitably results to the concept of learning. Is it possible to make the computer, to some extent, learn the data it is presented with?

Biological Neural Network

The biological neural network serves as a natural engineering example of a working, intelligent information processor. As a result of millions of years of evolution, the brain has evolved into a compact, optimum package of computing power capable of dealing with the myriad situations that it can run into. The power of the biological brain lies both in its unsurmountable flexibility and massive parallelism.

A neuron is the fundamental unit of a biological neural network. The structure of a neuron is represented in Figure 5. A neuron is made up of a nucleus, a cell body (soma), dendrites and an axon. The cell body and the enclosed nucleus do not play a significant role in the processing of incoming and outgoing data. Rather, the cell body is a place for the mechanisms that provide the cell its energy and cause the activation of the cell. Dendrites receive the signals from other neurons. The axon transmits the signal to other neurons. At the junction of the signal-sending axon and the signal-receiving dendrite lies a small gap called a synapse.

As the neurons learn to react to certain signals, the synaptic connections between neurons either get stronger or weaker. The strength of the synaptic connection determines how strong the receiving neuron finds the signal. The signals from different neurons are thus weighted differently based on the strength of the synaptic connections. If the total effect of all the received signals is adequate, the neuron is activated and it will begin to send a signal to the other neurons via its axon.

Neurons that are connected with each other with synaptic connections constitute a biological neural network. The human brain consists of about 100 billion (10^11) interconnected neurons. A single neuron has 1000 to 10 000 connections with other neurons.

Figure 5: Biological neuron.