This week, we are hosting the Facebook Fellowship and Emerging Scholar Summit at our Menlo Park headquarters for researchers to share their work and network with the broader Facebook Research community.
At the Fellowship and Emerging Scholar Summit, the PhD students join Facebook researchers for a multi-disciplinary program designed to share ideas across research areas and spark new ones. The Fellows and Emerging Scholars will share their research and also learn more about the technology challenges Facebook is working on.
We had a chance to catch up with five Fellowship and Emerging Scholar winners and learn about their research interests and current projects. They showcased a diverse range of research areas, including bridging the gap between cryptography and the law; making telecommunication systems more reliable in emergencies; improving online security advice and examining ethics in virtual worlds; applying integrated photonics to data center networking; and adaptive machine learning in real-world settings. Here is a little more about them and their research.
Aloni Cohen is a sixth-year PhD student at MIT’s Computer Science and Artificial Intelligence Lab (CSAIL), advised by Shafi Goldwasser. While Aloni was researching cryptography at MIT he also took law classes at Harvard Law School. This inspired him to explore the interplay between theoretical cryptography, privacy and law. His current research focuses on bridging the conceptual divide between legal and technical thinking about privacy and cryptography.
He is also working on research that involves applying theoretical privacy attacks to the real world. This year, he and a collaborator earned the highest prize in the world’s first bug bounty program for anonymous data re-identification.
In his work, Aloni argues that the jargon of privacy regulation can be poorly defined and confusing, so he is researching ways for technologists and lawyers to work together more effectively. “Lawyers and computer scientists don’t always know how to talk to each other. We don’t understand what the technical analogs of legal requirements are. And once we understand the technical side, we don’t know how to translate it back to the language of law,” he says.
Formal modeling and careful definitions are essential components of theoretical cryptography. Using the methods and definitional frameworks of cryptography, he hopes to mathematically operationalize some of the various privacy laws and regulations. An ideal goal would be to develop a collection of mathematical notions that capture different facets of the patchwork of regulations. These notions could then be used to support or refute arguments that a given technique satisfies or violates a particular privacy regulation.
Aloni has also explored how the Fifth Amendment plays out in cases involving cryptography. He co-authored an article for the Harvard Journal of Law and Technology that explores the Fifth Amendment and compelled decryption in criminal investigations. They examine how the Fifth Amendment privilege against self-incrimination affects “compelled decryption” in criminal investigations. In these compelled decryption cases, law enforcement wants to force a suspect or defendant to decrypt a device (a phone or computer) but the suspect “pleads the Fifth.” The article explores the legal doctrine surrounding these cases from a technological perspective.
Dylan Foster is a PhD candidate in computer science at Cornell University, advised by Karthik Sridharan. His research focuses on theory for machine learning in real-world settings, where data is too large to fit in memory or where there are constraints on computation or information. He is currently working on problems in interactive learning, deep learning and generalization. Dylan received his BS and MS in electrical engineering from USC in 2014 and was a recipient of the NDSEG fellowship.
Dylan’s research focuses on developing mathematical tools and theory for adaptive machine learning. This includes identifying properties of real-world instances that lead to better (computational or statistical) performance and developing scalable, efficient and provable algorithms to exploit these properties.
A common challenge in modern machine learning and statistics is trying to learn from a dataset too large to fit in memory or even on a single machine. A frequent approach to tackling this problem is to apply an online learning algorithm, which processes examples in the dataset one-by-one or in mini-batches. Dylan’s approach has led to a new “plug-and-play” framework for adaptive online learning.
He has also recently been collaborating with other researchers on extending the adaptive learning tools to more challenging interactive settings that have limited (or “bandit”) feedback. He has been focusing on a setting called “contextual bandits,” which is a simplified reinforcement learning setting.
The limited feedback in contextual bandits presents a serious challenge, both for our understanding of the statistical complexity of the problem and for algorithm design. Dylan has done some initial work adapting the tools he developed for the online learning setup to this setting. He wants to keep pushing this direction, both toward a sharper understanding for contextual bandits and towards richer interactive learning settings.
Dylan will graduate from Cornell this year and become a postdoctoral fellow at the MIT Institute for Foundations of Data Science. MIFODS is focused on the intersection of mathematics, statistics and theoretical computer science. After the postdoc he plans to continue to pursue this line of academic research as a professor.
Originally from Ghana, Edith Ghunney is currently pursuing a PhD in electrical engineering at Georgia Institute of Technology, advised by Dr. Mary-Ann Weitnauer. Edith’s research interest lies in understanding, modeling and analyzing resilient wireless and mobile communications network for new connectivity technologies. Edith holds a BS with Honors in telecommunications engineering from the Kwame Nkrumah University of Science and Technology, Ghana.
Edith’s research focuses on linking the Internet of things (IoT) world and new connectivity technologies for prompt emergency response and recovery. Nearly every season, Ghana is hit with emergencies that threaten the lives of the public, first responders and properties. In most cases, the telecommunication infrastructure that is essential for communications and recovery during such critical times is unavailable and national disaster management funds are scarce.
Edith’s goal is to design the air interface to improve access and connectivity during emergencies for the next generation of communication systems.
Her research seeks to answer these questions:
- What are the stringent communication requirements for an emergency-ready network system?
- How do we ensure that mobile users are discovered within the shortest possible time?
- How do we ensure that once users are discovered, there is an uninterrupted connection and fast and reliable transfer of data under resource-limited multi-user environment?
She hopes that her work may lead to the improvement of emergency response and recovery in general. Edith is particularly excited to be an Emerging Scholar at Facebook. “Facebook is a brand I have come to love. Coming from Ghana, I have firsthand experience of how Facebook has served as a powerful tool for citizens to connect with our political leaders, to provide constructive criticisms, and keep them on their toes in spearheading the development of the country,” she says. “Facebook stands for and connects people from all cultures and backgrounds and it is an honor to be associated with the amazing brand through this scholarship.”
Elissa Redmiles is a PhD candidate in computer science at the University of Maryland and has been a visiting researcher with the Max Planck Institute for Software Systems and the University of Zurich. In addition to the Facebook Fellowship, Elissa is the recipient of an NSF Graduate Research Fellowship and a National Science Defense and Engineering Graduate Fellowship, and the John Karat Usable Privacy and Security Student Research Award.
Elissa’s academic research interests are broadly in the areas of security and privacy. She uses computational, economic and social science methods to understand people’s security and privacy decision-making processes, and she is interested in research that reduces digital inequalities. She is currently working on several main projects, including:
- Evaluating the rationality and fairness of people’s behavior with respect to online security. People’s actions can help them avoid incidents, but can also increase costs, both to individuals (in time and mental effort) and to platforms (in user engagement and engineering resources). Elissa is working on designing a way to optimize the deployment of security nudges and requirements to maximize benefit for both platforms and people and minimize inequality between those with different skill levels.
- Measuring the quality of security advice. While a plethora of security advice is available, even those receiving advice still fall victim to online incidents. While some victimization is inevitable, difficulty comprehending security advice, inaccurate advice and vague recommendations that are not actionable may all contribute to increased victimization. Elissa is currently doing research to establish a set of methods for continuous evaluation/auditing of the security advice ecosystem. As part of this work, she plans to release a system to help organizations create new, high quality security advice that is accessible to all.
- Understanding behavior in virtual reality. Elissa also works with a group of underrepresented minority undergraduate students at the University of Pittsburgh Information Inclusion Institute. They study how people behave and make ethical choices in VR and draw from these findings to co-create best practices for VR content creation with developers.
While Elissa plans to pursue a career in academia, she is very excited that both companies and government have shown interest in implementing tools based on the findings of her work.
Greg Steinbrecher is a PhD candidate in the Quantum Photonics Group at the Massachusetts Institute of Technology, advised by Professor Dirk Englund. Greg received his SB in physics and electrical engineering from MIT in 2012 and his master’s in electrical engineering and computer science in 2013. His research interests lie at the intersection of quantum physics, communications and systems engineering.
With the exponentially increasing data requirements of networks, electrical links over copper wires are increasingly being replaced with photonic links over optical fibers. At first, this affected only the largest systems, like the long-haul links that form the backbone of the internet. In recent years, optical communications systems have made their way into both the datacenter and the home, driving rapid improvements in the miniaturization of optical technologies. Most recent improvements rely on integrated photonics systems, small computer chips that directly control light at a scale and efficiency previously impossible.
Greg’s research is focused on the design and application of integrated photonics—particularly silicon photonics—to both quantum computing and datacenter networking. In particular, he focuses on systems that contain hundreds of individual optical components, allowing for a scale of operation impossible without integrated technologies.
For quantum computing, he has explored the design and application of Programmable Nanophotonic Processors (PNPs): systems with hundreds of switches for light that can be individually controlled. These PNPs can be programmed to perform an endless range of optical transformations. Their reconfigurability allows for the realization of ultra-high-fidelity quantum gates, overcoming manufacturing defects.
With the support of the Facebook fellowship, he has been exploring the application of PNPs in the datacenter, particularly as a novel form of optical switch, capable of not just 1-to-1 connections, but reconfigurable 1-to-many configurations on millisecond timescales. He believes that integrated photonics are going to be an increasingly essential part of datacenters, allowing for optical networks that can be reconfigured on the fly to make more efficient use of the capacity available, at lower power.
Greg also did a summer internship at Facebook this year. “Facebook is pushing up against the edge of what’s possible,” he says. “I got to visit a Facebook datacenter. You can’t get a sense of the true scale until you’re in a place like that. It gave me an appreciation of how much there is to learn. I also saw how few people run things. It’s not a large company for what it manages to do.”
After graduation, Greg hopes to work in industry for a few years to get more familiar with the true challenges of hyperscale datacenter networks and where both the short-term and long-term opportunities for radical improvement are. He will explore whether he can have the most impact as part of a large organization or whether starting a company will more effectively enable him to drive disruptive change in the industry.