REDMOND, Wash., Feb. 23, 2004 — Microsoft recently honored some of the keenest young minds in computer science, mathematics and electrical engineering by awarding Microsoft Research Fellowships to 13 promising graduate students in the United States and Canada. The scholarships provide students who are considered "the best and the brightest" with an opportunity to pursue doctorate-level research for two academic years.
To learn more about the nature of genius in higher education today, PressPass hosted a roundtable discussion with Jim Kajiya , director, at Microsoft Research and four of the students who were named MSR Fellows for 2004-05 (see sidebar for complete list):
Vladimir Jojic , a Ph.D. student in computer science within the Probabilistic and Statistical Inference Group at the University of Toronto, Canada, whose field is machine learning and computational molecular biology
Jinyang Li , a Ph.D. student in the Parallel and Distributed Operating Systems Group at Massachusetts Institute of Technology's Laboratory for Computer Science, whose research is in the areas of computer networks and distributed systems
Chandra Nair , a Ph.D. candidate in electrical engineering at Stanford University in California, whose research focuses on crossover areas in probability, combinatorics and statistical physics
Alan Nash , a Ph.D. candidate in mathematics at the University of California, San Diego, whose work focuses on applications of logic to computer science
PressPass: Why does Microsoft Research award fellowships?
Jim Kajiya: We want to recognize academic excellence by providing material support to outstanding students in computer science and closely related fields. Fellowships have been recognized as one of the most desirable and effective mechanisms for doing so. These awards also demonstrate our support for outstanding university departments throughout North America.
PressPass: What does MSR look for in award recipients? In other words, what characteristics define a good Fellow?
Kajiya: We look for candidates with the highest potential for contributing to their chosen field. These people are highly creative and very bright, and they have interesting views on the world.
PressPass: Would each of you summarize your work and describe the research that you plan to focus on in the coming year?
Vladimir Jojic: My research focuses on building probabilistic models for different mechanisms and processes in biological systems. For example I am interested in the mechanics of immune system response, reconstructing the phylogenetic tree of the evolution of species, and the design of peptides for use in RNA interference. These areas are of interest because understanding these processes has direct application in how we deal with viruses and diseases. For example, an evolutionary model allows us to investigate how SARS jumped species and then infer how likely other viruses are to do the same. Similarly, by modeling specificity of mechanisms such as RNA interference we can ensure that we obstruct the building of faulty protein in a cancerous cell while not interrupting the production of normal proteins.
Jinyang Li: My work deals specifically with routing in dynamic peer-to-peer networks, such as music-sharing networks like Kazaa or Gnutella. Routing is a way to figure out the best path for traffic to travel in your network to get to its destination. But the way peer-to-peer networks are formed is often very ad hoc, so to find a piece of data, essentially you have to flood the request to many people on the network. If you want to limit the expense of that, you sacrifice accuracy; in other words, you might fail to contact someone who holds the data. The scientific community has devised an idea called distributed hash table, which is a more intelligent way of organizing this overlay network such that searching is much easier. With DHT, a routing algorithm tells you which path to take to find a piece of data. My current work focuses on analyzing these routing algorithms and design choices and understanding how they perform, especially in a dynamic network where a lot of nodes join and leave.
Chandra Nair: My main work has been in the area of combinatorial optimization. A typical problem in this area takes the following form: I have some objective that I want to optimize -- say, maximize profit or minimize cost -- but the number of possible choices of variables is so extremely large that an exhaustive search is impossible. This field deals with finding out optimal solutions in a more efficient manner. My main work has been to prove a conjecture in a particular combinatorial optimization problem. Optimization occurs in nature. For example, leaves in a tree arrange themselves so that each leaf gets a large amount of sunlight. Statistical physicists use existing physical laws to simulate the optimization that nature would do, even to problems that don't occur naturally. Their results are stunning, but the arguments are not completely rigorous. I will focus on using my rigorous solution to the particular problem I worked on to understand, explain and rigorize the work of the statistical physicists.
Alan Nash: I'm interested in roughly three areas in database theory and complexity theory. With regard to the more applied aspect of database theory, I'm interested in information integration. And in the more theoretical areas, I'm interested in two questions. One is the role of ordering in computation, and the other has to do with the open question of P vs. NP, which is one of the most important problems in theoretical computer science. This is difficult to explain, but it's easier to explain a related question: Roughly, what can we compute without using too much time, and what can we compute without using too much space? The intuition is that we should be able to compute more things if we're limited in space than if we're limited in time because you can reuse space but you can't reuse time. But people have been working on this for a long time and we have not been able to show yet that you can compute more things having limited space than you can having limited time. This is a big, open question, and I'm not by any means hoping to solve it, but I'm working on several questions related to it.
PressPass: What drives you to study in your particular area? What promise do you think it holds?
Nair: I'm interested in mathematics in general, and combinatorial optimization offers lots of exciting opportunities for solving difficult, open mathematical problems. Clearly, optimization problems occur in every walk of life in one way or another. We join the shortest queue in a grocery store, take the shortest path to our destination, and so forth. So I believe optimization problems are an important class of problems to study.
Jojic: Machine learning is very exciting because it allows you to get at the underlying truth using noisy and incomplete observations. In immunology, for example, we might ask, "Given the incomplete knowledge that we have today about immune systems, what are the factors that influence how an immune system would respond to a virus?" An observation of the rough data in different stages of the immune response, to my eye, would appear like a bunch of letters and numbers. But it's extremely interesting when you can postulate a probabilistic model of the data, find the important factors and discover an optimal solution for the problem at hand, for example, the best stimulus for the immune system.
PressPass: How do you think a fellowship like this affects your prospects? Do you see it opening up opportunities for you?
Li: I definitely see it opening up opportunities for me. I'm currently working on a specific research area, but with the Microsoft Research Fellowship, I have the freedom to pursue other areas if I have good ideas. I wouldn't have that same freedom in another position, for example, as a research assistant or teaching assistant at MIT.
Nair: It would have been difficult for me to go to any common funding agency and request money to work in an area that is extremely theoretical and from which there is probably no monetary gain for some time, in terms of product or practical applications. But a fellowship gives
you freedom to pursue fields that you feel really need to be pursued, both for furthering human understanding and for the impact this field will have several years down the line. Clearly, this fellowship also offers the opportunity for me to collaborate with the best theoretical group that exists outside of some top universities.
PressPass: What outlook do you see for IT as a career? Has computer science lost its luster in the post-dotcom boom? Or do college students today still see this as a good career choice?
Jojic: Sure, because innovation goes on and new technologies keep coming out. As far as research is concerned, there are popular areas, machine learning being one of them, as well as the merging of machine learning with other areas such as biology and speech/video/text processing. As long as people continue to be interested in solving problems, innovation in the computer science and IT area is going to continue. It might not be in terms of inventing new markets, which is what the dotcoms seemed to focus on, but there are definitely areas where you can work and solve interesting problems, and there are definitely business cases for solving such problems.
Nair: In the dotcom boom, non-genuine businesses got hyped up, causing a huge bubble. Now it's filtering out to the really good people who stayed in the field and are making real progress. What people expect today is slower but much steadier growth, not another huge boom. That's the general outlook I'm seeing among my peers. They're not hoping to become millionaires overnight.
Li: It's definitely healthier today. A couple years ago, people signed up for computer science because they thought it was going to make them quick money. Now they sign up more because they really like the subject. I see a lot of freshman signing up for computer science. It's down from a couple years ago, but computer science is still the most popular major at MIT. It's still a good career choice, plus it's definitely fun.
PressPass: What's your opinion of the state of the art of computer science in U.S. education these days?
Nash: I'm new to the academic system, but having come from Argentina, I'm pretty impressed by the professors you have access to. At the same time, coming from the mathematics department, I'm often underwhelmed about computer science students' ability to prove things or express themselves clearly. But in general, I think graduate education in the United States, especially in computer science, is certainly extremely competitive with respect to other countries.
Li: At MIT, my advisor is revising the computer science curriculum. It's still a very young field and fast evolving, so it's very exciting in that sense. A lot of new things are popping up in this field, therefore you have to constantly revise the curriculum and add new classes. But a few classes that are the foundations of computer science are not really in the core curriculum. Freshmen who sign up for a major in computer science are inexperienced and need guidance. So I feel MIT should revise the core curriculum, adding basic computer science theory, a compiler class and a programming class that everybody should take.
PressPass: Who are your heroes or role models?
Li: When I was doing an internship at UC Berkeley, I met Professor Richard Karp. He's a theoretician and a winner of the Turing Award, which is like the Nobel Prize for computer science. It really impressed me that someone so intelligent can be so humble and so curious. He's my role model. I want to be like him -- to keep an open and curious mind. He's around 70 and still doing active research and collaboration.
Nash: My two sets of heroes are scientists (including mathematicians) and the founding fathers of personal computing. In particular, since I'm interested in mathematics and physics, I would say Einstein and others who have contributed to the birth of quantum mechanics; Cantor, who was the father of set theory; and Gdel, who was the most important logician of the 20th century. But also in my teenage years, growing up and working in Argentina, two people I noted very early on were Steve Jobs and Bill Gates. I'm still very impressed with what they're doing.
PressPass: What about aspirations? Where would you each hope to be in five or 10 years?
Nash: It is hard to predict, but one likely possibility is that I'll be a professor in a university computer science department (in a database or theory group) or a researcher at an industry research lab.
Nair: In theoretical research or in academics as a professor.
Jojic: The setting doesn't matter -- wherever I can work on interesting problems.
Li: In five or 10 years, I hope to be doing very solid research in distributed systems and networks. And I hope people will respect me because they think I do very cool research.