Fri, Feb 29, 2008 - Page 11 News List

US firm sniffs a new opportunity

ARTIFICIAL INTELLIGENCE The challenge is to identify odors while filtering out other smells so the computer is designed to mimic the way the brain learns through experience

BLOOMBERG

A Silicon Valley startup is betting a computer can have a better sense of smell than a dog.

Paul Rhodes, the founder of Evolved Machines Inc in Palo Alto, California, is using graphics chips from Nvidia Corp to create a machine that mimics the way human and animal brains learn to tell one odor from another.

The goal is to create a device that helps firefighters recognize hazardous chemicals or that monitors shipping containers for drugs and bombs.

"Containers that come into every port in the country can't be inspected by dogs for drugs, bombs and toxins," Rhodes said in an interview. "There aren't enough dogs in the world."

The device is one example of new applications being created for graphics processors, the chips used in computers and game consoles to display images. Those semiconductors are designed to chop up program instructions and process them simultaneously, making them better for applications that require the machines to "learn."

Nvidia, the world's second largest maker of computer graphics chips after Intel Corp, is promoting more uses for its chips. Last year, the company released a product called Tesla, which lets companies write programs that are not solely designed to display graphics.

Computer maker Acceleware Corp is also using Tesla to help Boston Scientific Corp to design pacemakers that are not affected by medical scanners. The University of Illinois at Urbana-Champaign is using the same product for molecular research.

Rhodes, 48, says Nvidia's chips run his software as much as 100 times faster than general-purpose processors from Intel. His five-person company aims to have its first commercial prototype ready in about nine months.

The challenge for the device is to identify odors while filtering out wind and other smells. To do this, his software is designed to mimic the way the human brain learns through experience.

"The process of synthesizing these systems, of running them over and over again, of training them, requires a tremendous amount of computation," Rhodes said.

Rhodes said the program can learn in 16 hours what a baby picks up in three months.

While Nvidia's chips are helping Rhodes with his research, so called high-performance applications for graphics processors have not created a mass market yet.

"Nvidia's got to find new markets and new applications," said Jim McGregor, an analyst for In-Stat, a researcher in Scottsdale, Arizona.

The challenge of giving a computers artificial intelligence has occupied software and hardware makers for 25 years, without much progress, he said.

This story has been viewed 2608 times.

Comments will be moderated. Keep comments relevant to the article. Remarks containing abusive and obscene language, personal attacks of any kind or promotion will be removed and the user banned. Final decision will be at the discretion of the Taipei Times.

TOP top