Tech Transfer Central
University-Industry Engagement Advisor
University-Industry Engagement Advisor

GA Tech, Facebook partner to engage Black, Latino students in AI education

This article appeared in the January 2021 issue of University-Industry Engagement Advisor. Click here to subscribe.

Don't forget to sign up for University-Industry Engagement Week, the free online companion to University-Industry Engagement Advisor monthly newsletter.

You'll also receive info on upcoming webinars and other related products.

In the initial steps of a program of collaboration with U.S. universities “that serve significant populations of Black and Latino students,” Facebook has partnered with Georgia Tech to develop, co-teach, and fund graduate-level online deep learning courses. The program will be expanded in 2021 to include additional institutions.

The collaboration at Georgia Tech came about through discussions between Facebook and the university’s Machine Learning Center and the School of Interactive Computing, says Zsolt Kira, PhD, associate director of the Machine Learning Center at Georgia Tech and an assistant professor in the School of Interactive Computing.

This is not the first collaboration between the two partners, says Paco Guzmán, research scientist manager at Facebook and a lecturer for Facebook’s Co-teaching AI Program. “We’ve engaged with Georgia Tech for a number of years now — most recently with the Align Program and Co-teaching AI Program,” he notes.

The partners, Kira explains, had common goals and objectives. “We wanted to have an industry collaboration in order to better expose our students to industry relevance, and to problems in deep learning,” he says. “We teach the fundamentals in terms of theory, but we also wanted to expose them to how these methods are actually used in the real world, and problems they might encounter.”

Another objective, he says, was to increase inclusion and diversity — “democratizing” AI, if you will. “One of the key aspects of that is obviously incorporating materials into classes where you do have diversity,” he adds.

“Our vision is to engage with more minority-serving institutions and universities that reach the diverse population we are working to serve across the United States, so that we can provide scalable AI educational content to drive a bigger and broader impact within the community,” says Guzmán. “With this co-teaching program, we aim to make AI education more inclusive, and close the gap between what’s being taught in graduate-level CS courses and the deep learning techniques that are applied today by scientists and researchers. Georgia Tech is one of the most diverse institutions in the country and has established the OMSCS (online Master of Science in Computer Science) program, which particularly attracts people from diverse pathways.”

Bridging application, theory

In seeking to meet the goal of bridging theory and application, the inaugural course taught students techniques and theory along with practical applications on a step-by-step basis, so they could see how the technologies are applied at scale. Facebook’s AI experts provided mentorship and guest lectures, and they described a pathway to secure an interview at Facebook, with information on internships and residencies, as well as full-time employment. “There was a recruiting event with Facebook, where they would go over some ins and outs, and typical jobs that exist,” adds Kira.

“The collaboration started with the objective to complement Georgia Tech’s existing course in deep learning,” notes Guzmán. “We aimed to provide insights from Facebook AI’s researchers and engineers that demonstrated how to solve large-scale problems in an end-to-end approach (from data to deployment), [using] state-of-the-art models and emphasizing the importance of using/developing AI responsibly. We then worked with Georgia Tech to turn the on-campus course into an online class for bigger and broader impact within the community. The co-taught online course includes theoretical and real-world applications, so we can move in a direction of closing the gap between what industry is asking for and what universities are teaching in the classroom.”

The semester-long class on deep learning, he continues, covers fundamentals of neural networks and applications such as computer vision and language understanding. It also enabled students to innovate alongside Facebook AI teams on active projects, including AI Habitat, a simulation platform for embodied AI research, and fastMRI, a collaborative research project to investigate the use of AI to make MRI scans up to 10 times faster. “This fall students learned about PyTorch, and confident machine translation while also engaging in new capstone projects, like Facebook’s Hateful Memes Challenge, an open initiative to advance systems to detect multimodal hate speech,” Guzmán shares. PyTorch is an open source machine learning library used for developing and training neural network-based deep learning models.

“There’s been heavy collaboration,” adds Kiras. “Several faculty members at Georgia Tech are already part-time research scientists at Facebook. My students do internships and get jobs there all the time.” Although the deep learning course had been taught “for quite a while,” the material that existed prior to the collaboration was heavily modified to include the Facebook lectures, and to also add some industry aspects to the assignments, Guzmán notes.

“One of the big things students really love is a capstone sort of project — finding some research problem to solve,” Kira observes. Facebook provided several capstone project topics that were highly relevant to them and published some datasets as part of their involvement. During the projects, the students also received some mentorship from Facebook experts. At the end, they presented their work to Facebook and received feedback.

Collaboration has been key throughout the program, says Guzmán. “During the design of the first pilot, the Facebook team met with Dr. Kira to better understand his needs for the course and collaborated internally among Facebook contributors to develop lecture material and identify capstone projects,” he says. “Once the class was live, we ran surveys to understand the student sentiment and capture feedback, which was overwhelmingly positive, to improve the course structure and material. The Facebook team meets regularly with Dr. Kira and others at ‘GT’ to follow up on course development and discuss ways to evolve it to meet the needs of students and professors alike.”

‘Scaling up’

The collaboration, says Kira, is unfolding in phases. “This semester was part of an online masters’ program, which already has a better profile,” he states. “A lot of the students have jobs already and do courses on nights and weekends. The third phase, currently in planning, is to release all the education material online as part of a massive online course to allow other universities [to use it] — including top-tier institutions focused on research, but also teaching universities and HBCUs.” Next year, he says, he anticipates 150 “in-person” students (undergraduates and graduates) and 500 to 550 online students. “We’re really scaling this up,” he says.

Pre- and post- surveys were conducted for the pilot, says Kira, covering different aspects of the course and what students felt they got out of the Facebook interaction. “We’re now posting a similar survey with the online masters’ program,” he adds.

Kira sees clear benefits from the program. “One of the key aspects during the course is studying the real-world gap between the theory we teach and the set of problems you encounter when you deploy these algorithms,” says Kira. “Obviously there’s a whole set of challenges in terms of scaling, how to include them, the effect on users, and how to make the users’ lives better.”

That translation, he continues, was a big part of what Facebook teaches. “Through interaction, office hours, capstone projects, and recruiting events, they gave students a more holistic picture of theory and interested them in finding [work] with any industrial research position — Google, Amazon, other companies that encounter the same deployment challenges. We never removed the fundamentals, but they’re more prepared when they go for jobs.”

In addition, he sees a clear connection to ethics. “It’s really important that all students familiar with deep learning are at least cognizant of the fact that we have to be responsible for the algorithms we deploy, and what the larger implications are,” says Kira. “That’s one of the reasons we wanted to include diverse populations — so they could be involved in the process.”

Guzmán agrees. “We believe it’s important that people from all backgrounds get to be part of building AI systems, and investing in education is one way to bring more people from diverse backgrounds into the field of AI,” he says. “This benefits the entire industry; everyone wins.”

As for future partnerships, Kira has no doubt they will evolve from this experience. “Other courses are being discussed,” he reports, “and Facebook is also taking this as an example it can expand to other universities.”

“We’re excited to expand this program to additional institutions and universities with highly diverse populations,” adds Guzmán.

Contact Kira at

About University-Industry Engagement Advisor monthly newsletter...

Find more articles like this one when you subscribe to University-Industry Engagement Advisor monthly newsletter. Sign up today and get immediate access to our growing archive of UIEA back issues, as well as the distance learning program, Metrics and Benchmarking for University Corporate Engagement: Breaking Down Silos to Gauge Performance. This one-hour on-demand video will provide a roadmap to bringing your corporate engagement efforts into a more cohesive, holistic strategy, and gathering, organizing, and reporting the data you need to gauge the effectiveness of those efforts, report on your progress to stakeholders, and point you to areas needing improvement.