REDMOND, Wash. May 3, 2006 — Until new professors can build a reputation, they typically struggle to secure adequate funding for their research work. It’s a problem faced across the academic world as public funding agencies find their budgets stretched thin.
Recognizing this broad challenge for academia, Microsoft’s External Research & Programs group (ER&P) established the Microsoft Research New Faculty Fellowship program in 2005 to identify and support the best and brightest minds working to solve complex challenges in computer science, engineering and applied sciences at universities throughout North America. The objective is to stimulate and support creative research by promising researchers who have the potential to make a profound impact in the “state of the art” in their respective disciplines.
The program accepts just one nominee per university and includes a rigorous multi-round selection process that culminates in live interviews before a distinguished panel of reviewers from Microsoft Research and the academic community.
In just the second year of the program, more than 100 faculty members applied for five coveted positions. Winners were announced on April 26.
“Across the board, applicants for these fellowships represent some of the sharpest minds in science today,” says Rick Rashid, senior vice president of Microsoft Research. “Some of these researchers are working on ideas and concepts that sound like the stuff of science fiction movies, and it is very exciting for us to be able to support them in their work.”
The winners will receive a US$200,000 gift from Microsoft, paid in two annual installments of $100,000, as well as the ability to collaborate and share ideas and practices with Microsoft’s own team of researchers in facilities around the world.
The Microsoft Research New Faculty Fellowship awards can be applied in a variety of ways at the discretion of the winners, giving them the flexibility to take risks and explore ideas and realms that are not always possible with traditional funding.
Natural Language Processing
Winners of the 2006 Microsoft Research Faculty Fellowship (from left): Regina Barzilay, Aaron Hertzmann, Scott Klemmer, Eddie Kohler, and FeiFei Li.
“This money can support my students to work on the very edge of things,” says 2006 fellowship recipient Regina Barzilay, assistant professor of electrical engineering and computer science at the Massachusetts Institute of Technology.
Barzilay is focusing her research on natural language processing, in particular, statistical text generation, summarization and discourse analysis. The goal, she says, is to enhance technology with the ability to understand text on a structural level, such as a human would in producing an abstract of a document, rather than merely associating words like today’s search engines.
“Many tools we are using today understand the text as simply a sequence of words,” Barzilay says. “And you can already see that even those very linguistically simple tools make a lot of impact in how we access information. If you can provide more intelligent tools that can really understand the semantic structure, you can do much better things with information on the web today.”
Object and Scene Recognition
The ability to advance the state of the art and create more intelligent systems that help people do more with information is a theme echoed across this year’s fellowship winners.
Fei-Fei Li, an assistant professor of electrical and computing engineering at the University of Illinois at Urbana-Champaign, conducts research in computer vision focusing on high-level object and scene recognition. Fei-Fei’s proposed project confronts one of the most important questions in automatic visual recognition: how do we build algorithms capable of recognizing, and more importantly, comprehending cluttered real-world scenes?
“Vision is one of the most fundamental problems in science and engineering,” she says. “For humans, vision is the primary sensory source for survival and social interactions, and we take it for granted that we can recognize the objects in the picture, recognize the relationships between the objects, and recognize the theme, the activity. In computer vision, the overarching goal is to make computers do smart things like that.”
The University of Toronto’s Aaron Hertzmann is also working to develop more intelligence in computing systems. His work focuses on building realistic models of the world from data that can be used for computer animation as well as computer vision.
“A lot of people have are trying to make computers perfectly intelligent without telling them about the world,” Hertzmann says. “My approach has been to develop more realistic models of the world and then use data to build those models.”
Student, Faculty Collaboration
According to Scott Klemmer, an assistant professor of computer science at Stanford University, while the Microsoft awards will help advance the state of the art by funding research projects, they also provide an important assist for the university students working in tandem with faculty to develop those solutions.
“In computer science and engineering, and in much of the hard sciences, students and faculty work together in tight collaboration,” Klemmer says. “The work that my students and I have been doing on tools for designers has shown some really promising early results.”
According to fellowship recipient Eddie Kohler, an assistant professor of computer science from the University of California, Los Angeles, that element of the fellowship award is one of the most satisfying.
“I think that’s what we all hope as teachers, that we can watch our students change the world in whatever way they choose,” he says. “The goal as a teacher is to enable good things to happen, not necessarily to force them to happen yourself.”
And the Winners Are…
Regina Barzilay: The main thrust of Barzilay’s research is the development of probabilistic methods for modeling text structure. The focus of her work will be on two fundamental dimensions of text – content and cohesion. She is working on technology with the ability to describe topics present in a text and their organization, including the ability to represent text as a set of topics that are searchable, rather than just a document on one topic. The second area, cohesion, is concerned with how the information is realized in a given text. It captures the flow of the text and the linguistic devices that create it.
Barzilay became interested in languages when at a young age she moved from country to country, and began to contemplate the ability that humans possess to absorb languages in a relatively short time.
“It was fascinating how a person can go from a situation where you don’t understand anything and then in three months you start speaking,” she says. “I realized when I was learning a new language that I was looking for patterns. Computer science is all about seeing patterns in the data, building computational models, and then using those to build tools that process information in a new and different way.”
Aaron Hertzmann is an assistant professor of computer science at the University of Toronto, and also an affiliate professor with the University of Washington. Hertzmann’s research focuses on the intersection of computer graphics, computer vision and machine learning.
The major focus of Hertzmann’s work is machine learning for physics-based character animation. This work has the potential to create more accurate models of human motion than before, leading to widespread adoption of physical models for animation and tracking. Similarly, his methods may have impact on biomechanics research.
Hertzmann’s combined focus on graphics and computer science is no accident – as an undergraduate he double-majored in computer science and art, and was an avid painter. While those may sound like wildly disparate disciplines, Hertzmann says the two have complemented one another and helped propel and inspire him in his career.
“I ended up doing computer graphics because it turns out there is a lot in common between painting and computer science, and in fact a lot of my work has been about artistic media and computational models of art,” he says. “The practice of painting and computer programming are also quite similar. It’s been kind of a surprise to see that these two things, which I thought would be completely separate when I was in college, both turned out to have a big impact on my work since then.”
Scott Klemmer: Klemmer’s research in human computer interaction focuses on the design, implementation, and evaluation of software tools and interaction techniques that more seamlessly integrate physical and electronic media — bridging the gulf between the physical and digital worlds.
Klemmer’s goal for the next two years is to seamlessly integrate various media to capture data for designs, while immediately deploying prototypes to computer users, and to create a fluid way for passing that feedback to the development team.
The interaction techniques and design tools developed by Klemmer and his students speak to a future of human-computer interaction that is beyond the experience of today, and accordingly, much of his interest lies in helping designers create and realize a world where computing is embedded into everyday life, as opposed to isolated on desktops.
“When I think of computing beyond the desktop, it’s, how are we going to build that?” he says. “It’s not for computer scientists like me to create that, but for designers who are interested in the experience of technology. What we’re looking for is ways to enable a prototyping culture.”
Eddie Kohler is focusing on the cohesiveness and effectiveness of operating systems, Kohler hopes to develop a novel, elegant computer system that will cross domains – including secure operating systems, networking systems, and sensor networks.
One of Kohler’s overarching goals is to make fast systems more flexible and robust while keeping them highly effective. In particular, Kohler hopes to aid that process by focusing programmers on fine-grained component development, where errors are easier to detect.
In all his work, Kohler’s goal is to develop the simplest and most explicit abstractions that create understanding. “I hope eventually that my group will have analyzed and rethought enough components of an operating system, a web server and similar things, that we have a whole operating system that’s made up of these small components, so that the system itself is easy to understand.”
Fei-Fei Li is an assistant professor of electrical and computing engineering at the University of Illinois at Urbana-Champaign. Her areas of interest are computational vision, machine learning, visual psychophysics and perception. Specifically, Fei-Fei’s research in computer vision focuses on high-level object and scene recognition. Her proposed project confronts one of the most important questions in automatic visual recognition: how do we build algorithms capable of recognizing, and more importantly, comprehending cluttered real-world scenes?
More specifically, Fei-Fei hopes to address questions of multiple object categorization and scene-level classification. She aims to advance the state-of-the-art of visual recognition from both the theoretical and practical sides. Fei-Fei’s research lies at the intersection between the natural world and computer science, trying to build computational models that can accomplish the same functions as our natural “computers” do, but not necessarily trying to re-create the neural networks that nature provided for us.
“Brains can inspire computers and vice versa, but we don’t have to replicate them,” she says. “The plane doesn’t fly like a bird, but throughout the process whereby humans designed and invented airplanes, there was a lot of inspiration from nature. So that’s similar to the kinds of things we’re trying to do.”
“The visual world is so complex and every picture is different. There’s just so much in our visual world that to make a computer smart and able to handle all this is a major challenge. One of the unique elements of our work is its interdisciplinary nature. We are interested in both the scientific and the engineering questions. So this kind of interdisciplinary work is inspiring to both sides.”
The Microsoft Research New Faculty Fellows will be recognized at the 2006 Microsoft Research Faculty Summit, July 17-18 at Microsoft’s Redmond, Wash. campus.
