Robot misfits? No, artificial intelligence is much more than that.
Machine learning, as artificial intelligence is also known, can improve many real-world tasks, some that are surprising, said Rollie Goodman, a data systems analyst in Bozeman, Montana.
Goodman, 24, works at TechLink, the Department of Defense’s national partnership intermediary and a center within Montana State University’s Office of Research and Economic Development.
TechLink helps defense laboratories across the U.S. to license their inventions to businesses nationwide. Those companies use DoD technologies to grow by building new products and services.
To help businesses shop for new product ideas, TechLink maintains a publicly accessible database of the Pentagon’s patents, and that is where Goodman comes in.
To enhance database accessibility, Goodman built intelligent software that can read patents, a chain of three software modules actually, that decide what industry category the DoD’s invention belongs in, e.g., communications, energy, medical, or photonics.
Labelling helps businesses quickly sort thousands of available technologies and find those that fit their business areas, but because it is not always entirely clear where an invention might be useful, even to humans, Goodman’s research problem centered on accuracy.
The first classifier she created examines the patent’s Cooperative Patent Classification (CPC) code, a string of letters and numbers that are intended to harmonize U.S. and European patent documents. Humans cannot easily understand it without training, but computers have no problem understanding the CPC language.
We’re making it as easy as possible for businesses to find dual-use technologies because industry partnerships are good for the warfighter and the economy.
This first examination was categorizing the technologies with 87 percent accuracy, which is surprisingly high, Goodman said, but not good enough.
To raise that number, she built a second classifier, this one using natural-language processing, a form of artificial intelligence that understands human languages like English, to read through the patent’s abstract (a 50-to-150-word summary of the technology).
A third software module “chains” the first two together, looking at inputs from both and making a definitive industry area classification.
Using 9,000 test patents, Goodman and Joan Wu-Singel, TechLink’s technology scout, taught the system to achieve 99 percent accuracy.
“We’re making it as easy as possible for businesses to find dual-use technologies because industry partnerships are good for the warfighter and the economy,” Wu-Singel said.
Contact TechLink’s Troy Carter at firstname.lastname@example.org or 406-994-7798.