NEURAL SYSTEM MIMICS HUMAN LEARNING

A Computer than can think like the human brain is the ultimate goal of many scientists. A neural computer being developed in the US shows that the goal is not so elusive after all.

It is still light years away from HAL, the computer with the killer personality in the movie “2001″. But an advanced technology is now being developed that will allow computers to learn in a similar way to humans.

The experimental neural network computer developed by Bellcore, a research organization owned by US regional telephone companies, learns in a manner similar to the way school students learn to write by copying letters from a textbook or blackboard.

It catches on by following simple examples, coded in the form of electronic signals, which are provided by an electronic ‘teacher’.

The computer then processors these signals ‘much like the way a person learns to add two numbers together and then generalizes that knowledge to solve other addition problems,” said Josh Alspector, director of neutral network research at Bellcore.

The new neural network technology was inspired by structures in the human brain, he said.

“The prototype computer can process information at significantly high speed because it works in parallel, similar to the brain,” he said.

In comparison, most standalone computer and simulated neural networks process information serially, one step at a time.

This new computer can learn and process patterns in more than 100,000 individual signals per second, a speed about 10,000 times faster than what is possible with a sophisticated workstation.

The experimental computer may eventually provide a new generation of neural learning computers that will be able to solve some classes of complex problems faster and more efficiently than conventional machines.

And one day, these computers may also be to identify spoken words,  read handwriting,   identify

fingerprints and recognize a smell.

The Bellcore computer relies on an experimental computer chip which the research centre first brought out in 1988.

Like that chip, the more advanced version contains interconnected circuits which were inspired by the neural processes of the human brain.

Scientists believe that in the brain several billion cells called ‘neurons’ transmit motor, sensory and other impulses across gaps called ‘synapses.’

Smaller scale

On a much smaller scale, the advanced chip processes information by using electronic versions of 496 synapses and 32 neurons.

This computerized collection of interconnected circuits provides the neural computer with the processes to learn and make decisions.

In order to prevent the chip from getting ‘stuck’ during the decision making process, special electronic structures on the chip prod neurons with electronic ‘noise,’ encouraging the chip to make a good decision.

The experimental chip is ‘cas-cadable,’ allowing it to be linked with similar chips. Eventually, this could allow a computer to be linked into a larger system so that it can process information faster.

Bellcore researchers will add additional chips to their prototype computer to create an experimental neural learning computer that will help solve problem faced in telecommunications network management and operation systems.

These include assigning frequencies to wireless equipment, routing telephone calls, compressing telephone company business data for storage and transmission, and recognizing speech.

Commercial applications would be hard to predict and Bellcore must first demonstrate the technology’s usefulness in practical applications, Alspector said.

“We’re not making any plans for the commercial market,” he said, but it promises a lot of potential for the future.

“It’s a very generic technology,” he said, “It can be used for a lot of things.”

For example, it can be used for pattern recognition systems for lace, speech patterns and voice. It will be especially practical in optimisation type problems such as assessing frequencies in cellular telephone systems.

Another area where it can used in is signal processing for the equalisation of regular telephone and fiber optic lines.

It can also can eventually be incorporated into expert systems by giving training examples and sets of rules for handling load applications, he said.

A Californian company has already brought out the first product that uses neural network technology-Last monty, VeriFone unveiled its Gemstone Onyx cheque reader which it considers the first commercial application of neural network chip technology.

Onyx uses an integrated electronic ‘retina’ and electronic neurons to instantly recognise the magnetic ink character recognition characters printed on the bottom of checks, even if the MICR print quality is poor.

The neural network chip used in Onyx was developed by Synaptics.

This new computer can learn and process patterns in more than 100,000 individual signals per second, a speed about 10,000 times faster than what is possible with a sophisticated workstation.

MARTIN CHEEK

Subscribe / Share

Article by admin

Authors bio is coming up shortly.
admin tagged this post with: , , Read 238 articles by
It's very calm over here, why not leave a comment?

Leave a Reply