Communications of the ACM
Artificial Intelligence Joins the Fossil Hunt
Credit: Courtesy of Richard Austin/Rex Features
Researchers at Western Michigan University and Washington University in St. Louis have developed a predictive model that uses computer learning systems to identify potential fossil sites from satellite data.
The researchers programmed the software by inputting a list of known fossil sites in the Great Divide Basin of southwestern Wyoming and labeling them with one of five categories--fossil-rich, barren, forest, scrub, or wetland. The software then sorted unknown areas of the basin into the five categories.
In the first test, the model found "a huge portion of the basin was similar to what we had always found to be productive locations," says Western Michigan's Jay Emerson. The model also was able to identify that most of the area's fossil sites were in sandstone, but not all sandstone had fossils at the surface. The researchers modified the system to identify fossil-rich sandstone by adding two more geological requirements to the software. After the changes, the system correctly identified 79 percent of the known fossil sites as likely to contain fossils, and correctly classified 84 percent of all the other locations, according to Emerson.
From New Scientist
View Full Article - May Require Free Registration
Abstracts Copyright © 2011 Information Inc. 
, Bethesda, Maryland, USA
No entries found