As the winner of a 2017 Fannie and John Hertz Foundation Fellowship, Carnegie Mellon University alumnus Linus Hamilton is confident he can make a global impact in machine learning.

The Hertz Fellowship is one of the most selective graduate programs in the country, and this year only 12 recipients were chosen from more than 700 applicants.

The foundation is the only U.S. organization that supports its fellows for an entire five years of “total research freedom…with the goal of supporting the early stage research endeavors of applied physical and biological sciences, mathematics and engineering students who possess the potential to change our world for the better by solving difficult, real-world problems.”

The Hertz Foundation describes Hamilton as a highly creative and brilliant innovator — one they expect will be a game-changer in the field of machine learning.

Prior recipients of the fellowship include two Nobel laureates, a Fields Medal recipient and a National Science Medal winner. Hertz fellows have founded more than 200 companies, registered 3,000 patents and played key roles at major universities, national laboratories and the U.S. military.

Hamilton, who earned his undergraduate degree in mathematical sciences at CMU in 2016, is working toward his Ph.D. in computer science at the Massachusetts Institute of Technology.

“CMU introduced me to how awesome computer science is,” he said. “I also got to know several professors who offered to work on research projects with me, and helped me apply to graduate school. CMU has top-notch teachers, both in and outside the classroom.”

One of those top-notch mentors is Po-Shen Loh, associate professor of mathematics and a 2004-2009 Hertz Fellow. “Linus is simply brilliant,” Loh said. “Often, when ‘teaching’ Linus, I would learn something new myself! He has the potential to become a leading researcher who combines deep and insightful creativity together with a most engaging way of communicating his discoveries to the world.”

Hamilton’s research entails developing a better understanding of artificial neural networks, or complex software systems that identify trends in massive data sets. These networks employ multiple processing layers to train themselves, changing in response to the information they encounter.

Current applications range from image and speech recognition to targeted online advertisements. Their potential could mean anything from computerized tumor identification to human genome analysis.

“Right now neural nets are this magic tool,” Hamilton said. “But it’s totally black magic. There are maybe a few hundred people in the world who can set up a neural network to solve a really hard problem.

“Who knows how much time is wasted trying to train gigantic nets on problems they have no hope of solving?” he continued. “I want to try to change that.”

During the next few years, Hamilton aims to help develop guidelines that identify which problems can be best addressed, and develop tools that can give researchers a clearer look inside the networks they train.

# # #

Lead photo caption: Hamilton (center) with his classmates during his CMU days.