Raviv Raich and Cora Borradaile (in front) are the most recent faculty in the School of Electrical Engineering and Computer Science to receive Early Career Development awards from the National Science Foundation. A few of the 19 other award winners are pictured in the background.
Another year, another two award winners for the School of Electrical Engineering and Computer Science (EECS). This year Glencora Borradaile and Raviv Raich join 19 other faculty who hold similar awards. Both received a Faculty Early Career Development award from the National Science Foundation. These prestigious five-year grants recognize promising faculty at the beginning of their career for excellence and innovation in both research and teaching.
“Nearly half of our faculty have received these awards, but it’s actually very unusual for a department to have so many Early Career award winners. It demonstrates the high caliber of our faculty,” said Terri Fiez, head of EECS.
Understanding complex data
Raviv Raich explains his research by describing a simple example from the beginnings of data analysis. In the 1920s, statistician and nature lover, Ronald Fisher, set out to classify examples of iris into three different species based on four measurements of each flower.
These days we have access to data that is much more complex. Digital photos and audio recordings, for example, contain thousands of data points. To characterize this data in a way a computer can analyze requires multi-instance representation.
“So, it’s no longer just three or four numbers, instead we look at a collection of elements and each element is characterized by a set of numbers. It is a much more extensive representation, but it's needed for the complex analyses we want to do, such as identify bird species by a collection of chirps,” Raich says.
In a collaborative study with ornithologists, multi-instance learning is used to train computers to identify bird species from recordings made in the wild. This helps scientists track which species are present in a certain area or identify when an unrecognized species moves in.
Raich also uses multi-instance learning to help cancer researchers improve automated systems to test blood samples for the presence of certain types of blood cancer.
“The automated systems are very accurate and characterize 1,000 cells per second which is much faster than a human could ever do it,” Raich says.
Raich will use the $477,066 award to support undergraduate and graduate students to help further research in multi-instance learning which he expects will be a topic of considerable interest for years to come.
“Receiving this award has been fantastic because it allows us to continue the work that we were already doing and is validation of the work we have already done,” Raich says.
The world is not flat
Criss-crossing power lines inspired Borradaile to seek a better way to solve theoretical math problems that would have more impact on practical problems.
photo by Dorothy Delina Porter©
Glencora Borradaile advances mathematical techniques to solve problems such as how to connect wind generators to a power grid. Because these types of questions are so complex the standard way of solving them is to simplify the data. For instance, the 3D world becomes a planar surface (like a road map) that is very structured and easier to design algorithms for.
Simplification has its costs though, which became clear to Borradaile as she was biking on the back roads near Albany, Ore. and looked up to see two huge power lines crisscrossing each other, one above the other — an impossibility in a planar world.
It was an “ah-ha” moment that propelled her to seek a better way to solve theoretical math problems that would have more impact on practical problems.
Borradaile also solves problems in the Euclidean plane where paths can cross, but does not take other aspects of reality into account, such as restrictions on where a power line could be installed. In some cases this might be along a road, but there could be other situations where a cable could be buried underneath a farmer’s field.
To take into account a broader range of data, Borradaile is developing new paradigms for designing algorithms that can handle graph restrictions within Euclidean spaces, as well as other inputs.
“I'm hoping that the inputs we will be able to handle will be more useful in practice,” Borradaile says.
The $500,000 grant will support research by undergraduates and graduate students working in her lab, and Borradaile will also involve high school students in learning the fundamentals of discrete math which is the foundation of her research.