A human child can look at a cartoon picture of a chicken and recognize it is a chicken, but that is a show-stopper for machine learning. Unless it matches a cartoon chicken programmed in, it will not understand cartoon chicken-ness.

Devi Parikh of Virginia Tech has been given $92,000 of unrestricted funding by Google to work directly with Google researchers and engineers as they explore how to best teach machines from visual abstractions. Obviously if anything comes of it, that will be a real bargain.


Image: freepik.com

“People are the best vision systems we have,” said Parikh, assistant professor in the Bradley Department of Electrical and Computer Engineering at Virginia Tech. “If we can figure out a way for people to effectively teach machines, machines will be much more intelligent than they are today.”

Parikh is proposing to use visual abstractions or cartoons to teach machines. She works from the idea that concepts that are difficult to describe textually may be easier to illustrate. By having thousands of online crowd workers manipulate clipart images to mimic photographs, she seeks to teach a computer to understand the visual world like humans do. Parikh has expertise in computing areas such as computer vision and pattern recognition. Based on her earlier successful creative work on how to learn from visual abstractions, Google has selected Parikh to receive one of its Faculty Research Awards.

Parikh  is also a U.S. Army Research Office Young Investigator, working with the government on ways to reduce failures in computerized vision recognition programs.

“We need to build intelligent machines that can understand our visual world from images just as humans do. These machines must be capable of answering high-level semantic questions about an image such as what objects are present, where they are, and how they are interacting,” Parikh said.