The National Science Foundation selected UT on Aug. 26 to host the AI Institute for Foundations of Machine Learning, which will help improve the basic technology behind artificial intelligence.

The machine learning institute is one of seven with different focuses on AI that the foundation created in universities across the country in partnership with the U.S. Department of Agriculture. The foundation provided a $20 million grant to establish the institute and conduct research for the next five years starting Tuesday. 

The University will also establish the Machine Learning Laboratory with a donation from Austin-based tech entrepreneurs Zaib and Amir Husain. The lab will foster cross-disciplinary work and house the new AI institute, said Adam Klivans, who will direct the new institute and the lab.

“The (institute) is exciting because it's trying to improve the core algorithmic and modeling techniques that are common to almost every AI system,” computer science professor Klivans said. “When you make improvements there, you have the ability to automatically impact all of the AI systems that are deployed in a variety of different fields.” 

Jennifer Lyon Gardner, deputy vice president for research, said the institute will also create a new, fully-online master’s degree program in artificial intelligence, similar to the existing online master’s in computer science program. She said the institute will also focus on diversity through the initiative “40 by 24” with the goal of increasing the percentage of women in AI majors to 40% by 2024.

Gardner said the institute will also help undeclared freshmen transfer into computer science by holding some seats in CS 311 and CS 313, the core beginning Computer Science courses. She said this system piloted last year and people who were in the courses successfully transferred into the major.

Gardner said the institute is unique because it allows researchers to study and improve the fundamentals of machine learning rather than looking for ways to apply the current technology. 

“Our institute … goes all the way into the math that underlies the algorithms to try to figure out how to make them way more efficient, learn things way faster (and) be more accurate,” Gardner said.

Alex Dimakis, a co-director of the institute, said while machine learning is already in practice in many areas, the systems are still unpredictable. One goal of the institute is to make machine learning algorithms more consistent, Dimakis said. 

“You have this (machine learning) system, you show it thousands of images and it learns to classify images,” Dimakis said. “You show it a fresh image of a cat, and it says it’s a cat … but then turns out if you modify five pixels … then it can classify the cat as something completely absurd, like a rifle.”

Dimakis said the team will make some adjustments to the original plans due to the pandemic, but the program is well suited to be fully-online since there are no physical experiments. 

“We'll be kicking off research activities and things right away,” Gardner said. “We had developed the whole thing and submitted it before (the COVID-19 pandemic), so we're having to go back and think about how to make some of those face to face things all remote, which is definitely doable.”