Each year, Facebook awards fellowships through the Facebook Fellowship Program to talented PhD students conducting research in areas related to computer science and engineering. This program is designed to encourage and support promising PhD students who are engaged in innovative and relevant research across computer vision, networking and connectivity, UX and well-being, and more.
In this blog, we highlight Facebook Fellows Pang Wei Koh (Stanford University), Sergio Romano (University of Buenos Aires), Samaneh Azadi (UC Berkeley), and Michael Abebe (University of Waterloo). These PhD students showcase the range of backgrounds of our 2018 Fellows, as well as the wide array of research interests that they represent.
Pang Wei Koh is a third-year PhD student at Stanford University. His research focuses on making machine learning models interpretable and reliable.
Pang Wei’s interest in machine learning started with a chance connection in high school. “I’m Singaporean, so I had to serve in the military for two years after high school. And one of my friends in my army section had just come back from MIT,” Pang Wei recalls. “He said that since I was going to Stanford, I should check out this professor called Andrew Ng, who was also from Singapore.” Ng, who would eventually cofound the edtech company Coursera, was teaching Stanford’s machine learning online class at the time. “Taking Andrew’s machine learning class online was what got me really excited about machine learning,” Pang Wei says. After starting at Stanford in 2009, Pang Wei worked as an undergrad research assistant with Ng and Professor Daphne Koller, studying deep learning and then computational cancer biology.
After following his bachelor’s with his master’s, Pang Wei joined Ng and Koller as an early employee at Coursera. There, he served in a nontechnical role as the director of partnerships and course operations and then eventually as a product manager.
After a few years working in edtech, Pang Wei decided to return to research. In 2015, he joined Professor Anshul Kundaje’s lab in the Stanford genetics department to work on computational biology, and a year later, he started a PhD at Stanford, where he currently works with Professor Percy Liang and his group on problems in machine learning. “We tackle [machine learning] problems around understanding and trusting machine learning models in real-world applications,” Pang Wei tells us. “My current focus is better understanding the interplay between the training data and the model that we learn from this training data.”
To learn more about Pang Wei and his published research, visit his website.
Sergio Romano is a fourth-year PhD student in computer science at the University of Buenos Aires. After graduating with his bachelor’s degree in computer science, Sergio spent a couple of years working as a developer before returning to the university for grad school. Sergio’s two main areas of interest are artificial intelligence and computational cognitive science, particularly probabilistic programming and its application to learning.
Sergio aspires to try to build a more humanlike AI, which means helping computers to have more structured representations and to build models like humans do, as opposed to statistical pattern matching. “From when you are a kid, you are able to understand the idea of a dog with very little data — even if you see a dog with three legs, you know it’s a dog,” Sergio explains. “For computers, that’s very hard.”
Sergio mentions a particularly engaging presentation by FAIR research manager Larry Zitnick at the 2018 Facebook Fellowship Summit that relates to the type of research Sergio is interested in. Larry’s presentation was about the work that Facebook did when they open sourced ELF OpenGo. “[Larry] said that at the end of the day, the machine is just doing some pattern matching with the game, but the computer doesn’t even know what ‘Go’ is. It doesn’t even know what a game is.”
According to Sergio, although the University of Buenos Aires is the largest university in Argentina, the computer science department has a smaller budget and size in comparison to the top universities in the world. This caused Sergio some initial apprehension about applying for a Facebook Fellowship: “At first I wasn’t going to apply because Facebook is considered a very big company, especially outside the U.S. And so I was thinking there would be millions of applicants from around the world. Plus, in these programs, you see people from all these bigger universities from the United States, like MIT and Harvard.” After applying, Sergio realized that there were fellows of all educational backgrounds in the program.
For those with similar reservations who are considering applying, Sergio provides a few words of encouragement. “If you have a good idea and you are working on something that’s interesting, just apply.”
Samaneh Azadi is a PhD candidate at UC Berkeley, advised by Professor Trevor Darrell. Her research focuses on deep learning and computer vision. Before moving to the Bay Area to pursue a PhD in computer science, Samaneh earned a bachelor’s degree in electrical engineering from Shiraz University in Iran. After she graduated, she spent a year as a visiting researcher at UC Berkeley, where she was exposed to the world of computer vision and found herself in the middle of the deep learning revolution.
For the first part of her PhD, Samaneh was particularly interested in improving object recognition systems in challenging scenarios such as noisy supervision by considering label- and instance-level relations in the scenes. Starting in 2017, Samaneh began research in generative adversarial networks (GANs) and is currently interested in understanding the structure of various visual domains. “GANs have made a significant improvement on image synthesis in recent years and have opened up research perspectives in areas like image editing, creativity, and design,” Samaneh explains. “So I’m excited about both the theoretical foundations of these models and the practical applications that they introduce to get a better understanding of our visual world by learning how to model complex visual distributions.”
After she graduates, Samaneh plans to continue doing research, either as a postdoc or as a research scientist. “Collaborative environments within big tech companies provide very helpful insight in identifying the open problems in the field and finding their potential solutions,” Samaneh says. She notes her excitement and passion for making an impact in the field, citing her recent industry collaboration with a major creativity software company in Silicon Valley. This collaboration led to a technique that was promoted as one of their top new technologies of the year.
To learn more about Samaneh and her published research, visit her website.
Michael Abebe is a third-year PhD student at the University of Waterloo in Canada, where he is supervised by Professor Khuzaima Daudjee. Michael’s research is focused on designing and building large-scale data systems that are elastic, scalable, and self-managing. Before grad school, Michael earned a bachelor’s degree in computer science at the University of Waterloo, with a specialization in bioinformatics. As part of his undergraduate career, Michael did six internships, the last two of which were at Facebook Seattle.
Michael shares how his industry experience through an internship at Facebook influenced his graduate school research. “The team that I was on at Facebook was working on a distributed storage system that uses Erasure coding. This inspired the first research project that I did as part of my PhD at the University of Waterloo: We developed a system for managing Erasure-coded data that dynamically places data to reduce access latencies.”
Michael’s overarching research goal is to build systems that can store and manage data efficiently, which interests him not only from a practical standpoint but also from a purely technical one. “The challenge of reducing the amount of money spent on servers where you can store data or network bandwidth, I think, is interesting from the perspective of how you can help businesses save money,” Michael explains. “But there’s also something really exciting about being able to say that we can do something the fastest. There’s that desire to build the state of the art.”
The Facebook Fellowship Award includes a research grant and covers tuition and fees for up to two academic years, but it also provides financial support for conference travel. This is meant to encourage fellows to become involved in the academic community by ensuring the opportunity to do so. “It’s been really nice to not have to worry about money in that regard,” Michael says. “Having that support relieves a lot of stress with supervisors having to find money in their budget, so that’s been very helpful.”
To learn more about how to become a Facebook Fellow, visit the Facebook Fellowship Program page.