September 28, 2017

The 2017 Facebook PhD Fellows Workshop

By: Meta Research

This week, we are hosting 27 Facebook PhD Fellows and Emerging Scholars at our Menlo Park headquarters to share their work and network with the broader Facebook and fellowship community.

At the annual Facebook Fellowship research workshop, the PhD students join Facebook researchers for a multi-disciplinary program designed to share ideas across research areas and spark new ones. The Fellows will share their research and also learn more about the technology challenges Facebook is working on.

We had a chance to catch up with three Facebook Fellows whose research ranges from boosting Internet performance to social roles in online communities to addressing global inequality. Here is a little more about them and their research.

Improving Internet performance in an open, collaborative environment

2016 Facebook Fellow Brandon Schlinker

Brandon Schlinker is a fifth year Computer Science PhD student at the University of Southern California (USC), advised by Ethan Katz-Bassett. He received his B.S. in Computer Engineering from San Jose State University in 2013. His research focuses on improving the Internet’s performance and reliability.

Today’s Internet brings challenges and opportunities

Users expect Internet services like Facebook to be fast regardless of the time of day or their location: photos should load quickly, and videos should not be stuck buffering. However, the limitations of the Internet’s over 20-year-old Border Gateway Protocol (BGP) make it challenging to achieve reliable, fast connections.

But the Internet is changing, and today a substantial volume of the Internet’s traffic comes from a small set of content and cloud providers, including Facebook, that have built global content delivery networks. By coupling this rich connectivity with software-defined traffic controllers that conduct real-time measurements, Brandon believes these providers can bypass many of BGP’s limitations and improve people’s experience.

Brandon’s research at USC focused on realizing this vision and included developing PEERING, a community BGP testbed that enables researchers to conduct previously impossible experiments. “It’s difficult to perform Internet research in academia, largely because you lack the control and connectivity that companies like Facebook have,” he says.

Moving fast at Facebook

Brandon was awarded the Facebook Graduate Fellowship in 2016 and has been collaborating onsite with Facebook for over a year. The team at Facebook was very supportive of his research from the get-go. “When I talked to them about my ideas and potentially collaborating, their response was: When can you start?”

That spirit of collaboration exists throughout the company. “People at Facebook are very open about what they are working on and exchanging feedback on ideas,” he says. “This environment facilitates organic collaboration between teams and ultimately results in better infrastructure.”

The collaborative environment extends to Facebook’s interactions with academics, and Brandon notes that his experience at Facebook has been key to his graduate research. “One of the largest challenges in a PhD is choosing important problems and ensuring that your solutions can have impact,” he says. “In academia, this is difficult because you have limited insight and are forced to make assumptions. By collaborating with Facebook we can use production data to help us identify key challenges and evaluate solutions. Instead of evaluating a system in simulation or on a small scale, I can build it, bring it into production, collect measurements, and get a concrete answer. We’ve learned a lot at USC because of our collaboration with Facebook.”

Brandon credits the lack of division between production and research, as well as the company’s ‘move fast’ philosophy, with enabling him to have impact and improve the experience of Facebook’s billions of users around the world. “With my team, I led a fundamental redesign of Edge Fabric, Facebook’s Internet routing controller. I was able to learn and achieve a lot very quickly and be rewarded when I got to see my ideas go into practice.” The redesigned Edge Fabric now helps Facebook make routing decisions that avoid congestion and improve performance.3

Brandon and his collaborators (from Facebook, Columbia University, University of Michigan, and UFMG) shared details about Edge Fabric in a paper presented at SIGCOMM 2017, the top networking research conference. “My hope is that publishing details about systems like Edge Fabric and building testbeds such as PEERING will help others in the academic community identify and solve important research problems.”

Understanding social roles in the online world

2017 Facebook Fellow Diyi Yang

Diyi Yang is a PhD student at Carnegie Mellon University’s Language Technologies Institute (CMU), advised by Prof. Robert Kraut and Prof. Eduard Hovy. Her research interests are in Social Computing and Natural Language Processing. She has done internships with Facebook, Microsoft Research and the Wikimedia Foundation. She received her bachelor degree from Shanghai Jiao Tong University.

Her research spans three areas:

  • Analyzing semantics for language understanding, such as identifying semantic structures behind humor and recognizing different types of social support
  • Extracting social roles that group members occupy and modeling how role collaboration affects teamwork success
  • Facilitating social interaction by developing interventions, such as building recommender systems to match users with appropriate tasks and connect them with similar others in online communities.

“The online world reflects the offline world, both the beauty and dark side of human life,” says Diyi. “The online environment adds another dimension to the offline world. I want to understand how our society is magnified online, and how to use algorithms to enable better human interactions.”

Diyi’s work studying social roles in online communities spans projects in healthcare, education and publishing.

The American Cancer Society wanted to enhance its online cancer survivor network to provide members with the support they needed. Diyi built a model of text analysis to figure out what type of support (informational or emotional support) cancer patients need, and found that their needs varied depending on the stage of their disease. “There were also needs beyond what we defined,” she says. “In Sheryl Sandberg’s book Option B, she explains that when something bad happens, it’s important to share your story with others. Self-disclosure was another very important need for many cancer survivors.”

Diyi used these insights to build a recommendation system for the American Cancer Society’s cancer survivor network. When patients participate in the forum, the system helps identify what type of support each person needs at that stage, whether it’s information or emotional support. The system can also make recommendations to match patients with other similar patients. Members of the network also take on different social roles. Some provide emotional support while others choose to share information. “I’m so glad that this work can help real people from all over the country connect and support each other,” she says.

Diyi has also studied social roles in Wikipedia’s online communities. “Contributors to Wikipedia work together to produce a masterpiece for the rest of us. Combining knowledge from regular people is a beautiful thing,” she says. She discovered that there are eight types of social roles in Wikipedia communities, ranging from copy-editors to content experts to editors to updaters. Different roles are needed at different stages. Her work can determine what role a person occupies based on the history of their participation. If we know a person’s role, we can make online communities more effective. We can make recommendations, match them with other users and tasks, and help maintain their engagement online.

In her internship at Facebook one of the areas she worked on was to use deep learning techniques to recognize hate speech. “My mission is to use my technical strength in machine learning for social good, to build something that helps people. Real world problems are very complicated,” she says. “I had the chance to collaborate with many different types of people including designers, engineering, policy, community operations and UI researchers. My mentors were very supportive. Every day I felt like my contributions were so useful for this group. It was an amazing experience.”

Combining algorithms, AI, and social sciences to address inequality

2017 Facebook Emerging Scholar Rediet Abebe

Rediet Abebe is a Computer Science Ph.D. student at Cornell University, advised by Jon Kleinberg. Her research lies at the interface of algorithms, computational social science, and applications for social good. She is interested in using ideas from algorithms, networks, and data science to better understand and implement interventions in socioeconomic inequality and opinion dynamics. Prior to Cornell, she completed an M.S. in Applied Mathematics from Harvard University, an M.A. in Mathematics from the University of Cambridge as a Harvard-Cambridge Fellow, and a B.A. in Mathematics from Harvard University. She has also completed two research internships at Microsoft Research.

Rediet grew up in a low-income family in Addis Ababa, Ethiopia, where she followed the national curriculum before getting a merit-based scholarship to attend a local international high school. “Going to an international high school opened up a series of possibilities to attend fantastic universities abroad,” she says. “But I haven’t lost my roots. I care deeply about where I am from and improving access to opportunity for others.”

She takes an interdisciplinary approach to address inequality, by complimenting her computer science research with graduate courses in sociology and economics. She is also an active member of the Center for the Study of Inequality and Social Dynamics Lab in the Department of Sociology at Cornell.

Rediet believes that ideas from algorithms, network science, and artificial intelligence can both deepen and inform our understanding of inequality. With the availability of large datasets and increasing collaboration across fields, there is a wealth of areas where nuanced questions and novel techniques can reveal powerful observations about social phenomena and propose innovative solutions with far-reaching impact.

“Inequality has many dimensions. It could be economic, social, and health-related,” she says. “No one of these conveys the whole picture.” Rediet wants to shed light on facets that are harder to measure such as access to information and resources, and social ties. She wants to know what techniques in algorithms and artificial intelligence could improve our understanding of inequality.

Rediet has also co-founded several initiatives to help identify and forge new research paths and create more collaborative environments at the intersection of computer science and social sciences. These include the first Workshop on Mechanism Design for Social Good at EC 2017 and a corresponding interdisciplinary, multi-institutional reading group on the same topic, which have hosted speakers from computer science, economics, global health, operations research, public policy, and sociology. The goal of this initiative is to build domain knowledge in topics including affordable housing, economic inequality, and social mobility, to identify algorithmic, optimization, and mechanism design problems aimed at improving access to opportunity.

She is also passionate about increasing diversity within computer science. She is co-organizing the first Black in AI Workshop at NIPS 2017 and the corresponding Facebook group aimed at fostering collaborations and increasing the presence of black researchers in AI. She has also been a dedicated mentor to many women and under-represented minorities through various programs.

The Facebook Fellowship has been helpful towards pursuing her work. “I am passionate about improving access to opportunity for all individuals,” She says. “Recently, I’ve been thinking a lot about the range of insights available from Facebook, especially due to the high level of penetration in regions where there are many hard-to-reach populations and frequently changing demographics, like Ethiopia. This is really valuable information, and Facebook is in a unique position to explore some impactful questions related to inequality.”

Facebook Fellowship Program

We are pleased to host so many of our Fellows to learn more about their work, and to continue to support them in their PhD studies.  If you want to learn more about the Facebook Fellowship and Emerging Scholar awards, visit our Website.  Fellowship and Emerging Scholar applications for our 2018 awards are open until October 31, 2017