Skip to Navigation Skip to Page Content

The idea for an application of the research proposal that earned Assistant Professor Alina Zare a National Science Foundation Faculty Early Career Development, or CAREER, Award first occurred to her in 2010 when she first came to teach at MU and was trying to find her way around Columbia.

Zare

Assistant Electrical and Computer Engineering Professor Alina Zare has been awarded a National Science Foundation Faculty Early Career Development, or CAREER, Award to investigate ways to fuse different types of information with varying levels of accuracy for global scene understanding.

“I use Google Maps for everything. But even with that and directions, I still had to circle a plaza several times before finding a restaurant,” Zare said. She explained that maps, satellite imagery and other data are simply overlaid on each without any real understanding.

“Google street view, satellite imagery and maps should be fused to highlight what you are looking for in all three. I got pretty excited about global scene understanding. You could map the world. It would be so cool.”

Zare’s CAREER project, titled “Supervised Learning for Incomplete and Uncertain Data,” is funded from NSF’s Division of Information & Intelligent Systems. The research will investigate how to fuse different types of information with varying levels of accuracy. She will receive $454,077 over a period of five years for her “machine learning” project. The NSF CAREER award program, through which this funding is provided, is the NSF’s most prestigious award for outstanding faculty early in their careers.

Zare explained that by developing methods to train the computer to identify specific things in data — pixel by pixel — she will demonstrate examples of what she does and does not want it to find. The focus of her research for this project is to adapt this training process to handle data with inaccuracies and uncertainty in the demonstration examples. Although one application is global scene understanding, the approach is general and could be applied to many problems with inaccurate data.

“Satellite, Google Earth street view, tagged pictures on Facebook — there’s information from all over the world, but the challenge is that you can’t trust any of it completely,” Zare said. “Some social media like Flicker images are tagged with GPS coordinates, but they may be tagged in such a way that you don’t know which way the viewer was facing. If we can deal with inaccuracy and uncertainty in labels, we’d be another step on our way to having the algorithms understand.”

Zare will add to existing data through an online game she is developing.

“We’ll have people look at satellite images and ask them to identify what they see,” she said. “It’s a way to collect data in a controlled way.”

The game also is a way to introduce the concept of machine learning to students, and Zare plans to use it with high schoolers in MU Engineering’s summer camps and to cover related topics in her engineering classes.

“We’ll start small, right here on the MU campus,” she said. “We can go out there and look. We’ll be able to say, ‘Yes, there really is a sidewalk here.’”