In the video above, Penn State geoscientist Peter Wilf describes a computer program he and colleagues developed that learns, and can classify modern and fossil leaves over 70 percent of the time and place them in the appropriate biological family. By comparison, he says, it can take a carefully trained human two hours to classify just one leaf.
The software actually taught itself botany, learning from a large number of already classified leaves, but developed its own methods for classifying a leaf. It creates heat maps that place a red marked square onto the image grid to signify features of the leaf that are critical for identification. The problem is, often trained humans can’t figure out exactly why that particular feature is important.
Wilf and collaborators spent nine years refining the program. He hopes eventually to use it to create a more accurate picture of plant evolution.
Members of the news media interested in talking to Wilf should contact Patty Craig at 814-863-4663 or plc103@psu.edu.
