By Suzanne Ross
March 19, 2004 12:00 AM PT
Women's History month celebrates the many contributions women have made to society. Microsoft Research spotlights women researchers who are working today to advance the state-of-the-art in computer science.
Advice: Take more math. "The usefulness of math excited me," said Dwork. "I wish I had taken more math classes."
Cynthia Dwork, a researcher at Microsoft's Silicon Valley lab, knew in college that she wanted to help bridge the gap between society and machines. Her early influences included her father, who was a mathematics professor at Princeton, Noam Chomsky, and Star Trek. "I wanted to build a transporter," she admits.
Instead, she developed technology to preserve people's privacy. Her most well-known project, code-named Penny Black, could help protect people's privacy by protecting them from spam. "When you think about privacy, one aspect is controlling access to our information and to our attention," said Dwork.
Now she's turned her attention to database privacy, though defining privacy isn't as easy as it may seem.
"Privacy is a great warm fuzzy that's hard to get specific about," said Dwork. "And if we can't get a good English language definition, then we can't do anything in math, because math requires even more precision." Because privacy is such a complex issue, with contention about even the definition of privacy, Dwork and her colleagues decided to take on one small piece at a time.
"You can think about the census as an example of a case where we'd like to ensure some kind of privacy. Individuals provide information, and the census bureau, which is legally mandated to provide privacy, publishes a sanitized version of that information. Given legally mandated privacy, what kind of utility can we achieve? If we change the data to ensure privacy, is there any use left in the data? There are two extremes - at one end we can publish random noise and there's total privacy but no information. And if we publish everything as received by the census bureau then there's information but no privacy. We're trying to find ways of sanitizing data that will preserve stereotypes, but individuals will not be identified."
Most people would agree that we sometimes need data about groups to make good decisions. City planners might need to know how many people are planning to move in or out and how many have school-age children or teenagers. This would help them decide how many homes, schools, and malls to build. But they don't need to know who we are and where to find us - just that we blend into a pool of people with the same characteristics.
Though their work is still preliminary, Dwork and her colleagues hope to create a framework for meaningful comparison of the different database techniques connected to privacy.
"We're not the first to think about this by any means. There's a large body of work about this. What we bring to this is our background in cryptography, which offers a way of defining things clearly and correctly," concludes Dwork.
Advice: Work hard, learn how to hold your own when you have a good idea, and let others know when you have those good ideas. Have the confidence and belief in yourself and you can accomplish just about anything you attempt.
"The sky is the limit. That's what I love about my job," said Mary Czerwinksi, research manager of the Visualization and Interaction for Business and Entertainment (VIBE) group. Czerwinski is not only pushing conceptual limits, she's pushing physical limits as well, making computer displays larger and more useful. Her group conducts studies and designs software to help users get their work done more productively.
Czerwinksi, who holds a PhD in Cognitive Psychology, has always been interested in how technology and human cognition merged. "The software user interface is where that was happening when I graduated," said Czerwinski. "I'd developed some computer skills and wanted to keep working in the computing domain, but with a psychological bent."
One of her most ground-breaking studies involved large displays. Her group found that by increasing the screen real estate, workers could gain an immediate benefit - 10 to 30 percent more productivity depending on the task. Their work encompassed changes needed in the next version of Windows, as well as looking forward to when wide screens would no longer mean side-by-side single monitors, but would instead take many different forms. Working with the Hardware Research group's Gary Starkweather, the group used a prototype wide-aspect monitor to test their ideas.
One of the findings from this original study showed that women benefit from a wider view even more than men do. "We had users go through a maze on the computer to find out which cues in a large display will help users build better spatial memory for objects in the environment that they would have to find later. We stumbled on these gender differences," said Czerwinski.
"We thought that smooth optical flow cues were what helped women perform better while navigating through the mazes. And sure enough, when we ran the study comparing how people performed during a smooth animation versus one that faded the users in and out of the scene, we found that women did a lot better. The men preferred the smooth animation, but the choppy animation didn't bother them or slow them down as much as it did the women."
"Any one female can be just as good as or better than a man at spatial navigation tasks, but on average females are a little worse. So we need to support females with big displays, with wider fields of views when they're doing intense navigation tasks. They've been at a disadvantage in any 3D system, but just give them a wider field of vision and smooth graphics, and they're good to go."
One of the challenges Czerwinski faces when she's trying out new ways to help people interact with computers is how to keep the learning curve low.
"There is a demand to come up with visuals that allow users to see emerging patterns in their data streams, yet keep the user interface easy to learn. Whenever you have to come up with new metaphors it is a difficult task, since there are many people out there who are very comfortable with the way Windows and Office works today. But, if we can make the visualizations exciting and approachable, users can be energized into using new techniques or ways of interacting with large or novel displays," she said.
"This job is not just about the technology--it's about solving the user's problem and improving the quality of work or even life for them. I like that I get to observe how users approach new ideas. I like redesigning our products to be more useful to our customers. It makes me feel proud of Microsoft Research when we advance a technological area and improve the user experience."
Advice: Have confidence. Believe in yourself and believe that you can excel.
The human brain is an interesting machine. Damage one part of the brain and the function in that part can move over to another part. The brain was the inspiration for a lot of early work in designing self-repairing networks.
Lili Qiu, a researcher in the Systems and Networking group, is focusing her attention on building a network that can self-manage - in other words, do what the human brain does.
"My current project is managing wireless networks," said Qiu. "I'm working on enabling networks to self-diagnose faults and performance problems, and automatically recover from them. It is a very exciting problem. It would be of great use if we can provide automatic ways to facilitate the management process, and potentially for a network to manage itself."
Her research is aimed towards finding ways for networks to perform their best for the end-user. One of her projects was studying multi-hop wireless networks. She looked at the performance of a wireless network with a specific physical placement of wireless nodes and a specific traffic load. Her research, conducted with co-workers, suggested that 'shortest path routes' aren't always the optimum. Instead, there is a potential to provide much better performance through interference-aware routing.
Qiu is also interested in exploring the impact of selfish users in both Internet and wireless networks. "A recent trend in routing research is to avoid network-level inefficiencies by allowing hosts to choose their own routes using either source routing or overlay routing," said Qiu.
"Such routing schemes are considered selfish because the decisions are no longer based on system-wide criteria," she explained.
Theoretical results showing that selfish routing can result in sub-optimal performance have cast doubts on this approach. However, using realistic topologies and traffic demands, Qiu's research shows that selfish routing achieves close to optimal latency in such environments. Her work also points out selfish routing poses new research challenges to network engineering.
Qiu believes that technology can make our lives easier and more entertaining. However, she'd like to think that it won't replace more personal forms of communication.
"Technology sometimes helps bring people closer, but not always. Nowadays when holidays come, we send e-cards to friends instead of phone calls or making a personal visit. It would be nice to use technology to assist but not replace personal communication."
Advice: Follow your dreams and passions, no matter which field you are in. It is very important to believe in oneself. Sometimes it's hard to be in a field dominated by men, because there aren't that many women to talk to and get mentoring from. I think it is very important to encourage girls to follow careers in science and technology. I think having a balance is fundamental for the success of any discipline.
Developing a relationship with a computer isn't easy. A computer is not very smart. It doesn't talk back, it doesn't listen, and it doesn't think. Nuria Oliver, a researcher in the Adaptive Systems and Interaction group, isn't happy about that.
"Computers today have no senses at all. We need to add these capabilities into them to make them more useful to people," she said.
To reach that goal, Nuria has extensively studied machine learning, artificial intelligence, and computer vision. She'd like to make a computer that does more than serve as a passive receptor of information, she'd like to make a system that can see and interpret what it sees in some useful fashion.
For instance, if your environment, whether it's your home or office, was smart and had audio and video sensors to perceive what you were doing, it could anticipate your needs. It would know not to pop up an Instant Message from a friend saying, "hey, what's happening dude?" at some inappropriate time, such as when you're having a meeting with your boss. A smart chair could know if you were sitting in the right position or if you were sitting in a position that would make you feel like you'd been run through a pretzel-making machine. A smart environment could connect parents to their children's day care center, allowing them to peek in on their little one as he's fingerpainting his first masterpiece. That's the kind of smarts that Nuria wants computers to have.
Nuria was not originally planning to come to Microsoft. She was born in Spain, and did her undergraduate work there. After she finished her graduate studies (PhD) at the Massachusetts Institute of Technology (MIT), she interviewed at all the major labs on the west coast, but at first overlooked Microsoft. "I knew a lot of people in the Bay area, and I didn't want to move to a place where I didn't know anyone," she said. "I visited or interviewed at all the important research labs: IBM Almaden, Ricoh Silicon Valley, Xerox PARC, and Fuji Xerox. I also visited Stanford University." Anoop Gupta, the manager of the Collaboration and Multimedia group at MSR, had taught at Stanford. He heard about Nuria's talk and invited her to come to MSR to give the same talk. The invitation changed her plans and her life.
"I had never come here so I couldn't imagine how it was, but I came and gave a job talk. I realized that it was the lab in the best position today for the research that I wanted to do.
"In terms of the quality of the research, there are a lot of well-known researchers here, and we have a very strong presence in the scientific community. At this year's SIGGRAPH conference there were many papers by MSR, and it's very difficult to get even one paper accepted at SIGGRAPH. It makes me feel proud to be part of this community. We have a lot of freedom in terms of publishing, traveling to conferences, giving talks and collaborating with other researchers and with universities. It's a pretty open environment, which I truly believe is necessary to have real research.
"At MSR you can feel a lot of energy and it's just getting better and better over time."
Advice: Work on what you enjoy and what you are good at.
Most people run in the opposite direction when someone mentions abstract math, but Kristin Lauter, a researcher in the Cryptography group, was hooked the first time she took a course in it at the University of Chicago.
"I love learning new mathematics and finding relationships at all levels between different objects and between different parts of a theory. I love putting theory into practice and seeing how it works, and then trying to make it work better," she said.
Lauter, who is also the mother of twin daughters, works on optimizing the security of public key cryptosystems. She'd like to work more in the area of factoring. "The RSA cryptosystem is based on the difficulty of factoring large integers, and I would like to find a better algorithm for factoring. Primarily I hope to find new and better algorithms to solve existing problems, but I also like to explore and try to discover new patterns, and then understand and prove why they occur."
Public key encryption uses a set of two keys, one private and the other public. Though it's very easy for anyone to encrypt data using your public key, once encrypted, only your private key can unlock the secret. The public key is available to anyone, while the private key is securely stored. One of the most popular public-key algorithms is RSA, which depends on the difficulty of factoring large numbers. The public key in RSA is based on the product of two large prime numbers, and is used to compute the private key.
"The security of public key cryptosystems is based on the difficulty of solving hard math problems. I work on optimizing the efficiency of these systems while at the same time trying to "break" them by finding better solutions to the hard problems they are based on. I work primarily on elliptic curve cryptosystems and related mathematical problems," explains Lauter.
As you might guess, cryptanalysis isn't an overnight fix. There are still several famous unsolved cryptography problems. "I am currently finishing up a proof with a co-author of a conjecture that I made based on experimental evidence to generalize a well-known piece of theory in the field of number theory," said Lauter. "It is exciting to get the proof, but it was a struggle to find the confidence to keep working on it for over 2 years."
Advice: Be patient and demonstrate your ability step-by-step until your colleagues see your achievements. Once you establish your reputation, you will be as respected as any one else.
No matter where you're from, if you watch movies you had to have seen at least one early sci-fi film with a robot that looked like a tin can and sounded like one too.
Tinny, mechanical voices coming from your computer speaker won't endear you to this way of getting information. But it would be nice if you could have your computer 'read' your email or say please and thank you, and do it all in charming, dulcet tones.
Min Chu, a researcher in the Speech group at Microsoft Research Asia, is working to make this a possibility. Chu's group has developed a text-to-speech engine that sounds natural. It's difficult to tell their synthesized voice from the voice of a real person.
Chu is the project lead for the text-to-speech (TTS) project. She provided the main idea to construct a natural TTS engine. She also helped design it so that it would be easy to maintain and scale well. The group started with a Chinese engine, and then worked out an English version. They're working on developing the engine in new languages.
Chu has been working in speech processing since 1991. She wrote her PhD thesis on a Mandarin TTS engine.
"One of the best parts about this job is that everyone can tell when we've made improvements," said Chu. "Our demos always get a lot of applause from visitors."
TTS has many applications. It can be used to help blind users; they can browse the Web and get a much better experience. With TTS, you won't always have to use your glasses to read online - you can tell the computer to read to you instead. Speech can be a better way to 'read' your email, an article, or the latest news.
There are challenges in developing a natural text-to-speech engine. Chu's group studied and recorded about 20 hours of a professional announcer's speech, and extracted the natural intonation people use when they speak. They had to allow for multiple pronunciations of words such as read. Should the computer say 'reed' or 'red'?
People also know how to change their tone when speaking about different subjects. For instance, your voice tones would be completely different if you're talking about a sad subject vs. a sporting event. Chu and her team would like their engine to be able to analyze the content so that the voice tone can change appropriately.
"I hope to provide the best TTS technology to improve Microsoft products and help our customers," said Chu.
Advice: Get stuck in — computer science research can be very rewarding if you just have a go.
Ever hear a horrible noise in your car, but when you take it to the shop, it's gone? The mechanic looks at you as if to say, "what's your problem?"
Computer users sometimes have the same issues. A problem that was there one minute, can be gone the next, making it almost impossible to pin down. But you know it's sure to show up again.
Rebecca Isaacs, a researcher at Microsoft's Cambridge lab, enjoys digging under the hood to find elusive problems. She focuses her research on performance analysis of distributed systems. "I'm looking at how we can figure out why some requests occasionally go very slowly or fail. For example, why are some Web pages fetched really slowly and not others? Or why did something that worked fine 5 minutes ago, suddenly stop working?"
She was drawn to this type of work because there's always an opportunity to learn something new and do something different. "I enjoy the intellectual challenge and the job variety, the collaborations, conferences, seminars, reading, writing, thinking etc.," she said.
Her most recent project, code-named Magpie, is ambitious. Understanding the performance of distributed systems requires correlating thousands of interactions between numerous components. Their goal is a system that collects fine-grained traces from all software components, combines the traces across multiple machines, attributes trace events and resource usage to the initiating request; uses machine learning to build a probabilistic model of request behavior; and compares individual requests against the model to detect anomalies.
Magpie is an online modeling infrastructure that will look beyond aggregate statistics to model the path of each request through the system. This should reduce the aggravating 'it works for me,' reply to your pleas for help.
Advice: Be yourself, and build on your strengths. Think about what contributions you can make given specific things about your personality, interests, or experiences and use these.
The first week of college is usually crazy, hectic, exhilarating. For Lili Cheng, the research manager of the Social Computing group, it was portentous. She went to a lecture that made her question, "Why am I studying to be an architect? What am I doing here?"
"Don Greenberg, the computer graphics guru at Cornell, talked about his work. I decided to quit architecture and change to computer science. Then I changed my mind again, over and over."
Years later, she met Red Burns, chair of the Interactive Telecommunications Program at NYU, who told her, "sometimes you need to make a leap and follow what you believe in even if you're unsure." She made the leap and after studying at NYU, went to work for the Human Interface Group at Apple Computer.
But she didn't give up on architecture. The first project she worked on at Apple was QuickTimeVR, which focus on representing physical places on the PC. She is also a registered architect who has designed urban and public spaces in both Tokyo and Los Angeles. Today at Microsoft, she also gives guest lectures at the Harvard Design School and works on projects with the MIT Architecture School.
In 1995, Cheng arrived at Microsoft Research as a member of the Virtual Worlds group (now Social Computing), again mixing a physical place metaphor with people interacting, lessons from architecture and computer science.
Another of the projects she got involved in was a blend between her first and second careers. The professors at MIT were researching the link between technology and students' needs at the same time Cheng and Victor Bahl, another researcher, were looking for ways to help people connect.
Victor Bahl had come up with an idea to connect laptop users using location-aware technology. He tested his idea at a local shopping mall.
Unfortunately, though the system worked fine, he didn't find users lining up to try it. The problem, according to Cheng, was that "most people don't carry laptops in the mall, except maybe a few Microsoft employees. Most shoppers like talking to friends with common interests, but not strangers."
However, Cheng knew that Bahl's research would be good for students who needed to find each other to work on projects together - say, students who were studying architecture at MIT.
"The design studio course is the center of the architectural education. Groups of about 10-12 people work together in an open studio," explains Cheng. "Classes are typically held three days a week for four hours in the studio. However, students study in and around the studio all night, almost every night, designing their projects and working together. This intense pace creates a very close social group. We thought it would be very interesting to see how this would impact the learning process."
A more recent project that Cheng's group has designed is Wallop. The project is an experimental blog and photo sharing application. It helps people to connect with those close to them - families and friends, and friends of friends.
"Most of my current projects use computers to connect us with the people we care about and want to know about, rather than to process or retrieve information. It is challenging and exciting because computers weren't designed to do this, so we need to rethink many of our current assumptions about the way technology expects humans to think like machines rather than vice versa," said Cheng.
"In today's lifestyle, especially in the American culture, we often aren't around to communicate with the people and things we care most about. Technology can help us coordinate and keep in touch. It can show us different viewpoints. There are many problems technology can't solve. I hope it can help us be more aware of other people and other cultures and keep a perspective on the problems we are trying to solve."
Advice: I believe that a very exciting future awaits for scientists and technologists. For those who are now preparing to join these areas, having strong computing and analytical skills is crucial. However we need experts who can connect knowledge from various disciplines to push the frontiers of science. I believe, given the opportunity, that women's team-building talents and intuitive thinking skills will contribute to this.
Natasa Milic-Frayling is a mathematician who uses her analytical skills to improve computer systems, but her dream is to travel in the first space ship to Mars. Until that opportunity arises, Milic-Frayling spends her time developing innovative technologies by exploring new methods and continuously crossing the boundaries of scientific disciplines.
"I like to create innovative solutions to problems," she said. "I love the freedom I have here to choose to work on new research areas without being tied to any compartmentalized notion of a scientist or a researcher."
Milic-Frayling grew up in the former Yugoslavia where education was greatly valued and science and art were flourishing. From an early age her days were filled with rich learning experiences. She spent the morning at a regular school and in the afternoon enjoyed classes in the local music school. "I was in an all female class at my university because that particular year the men were joining the military service at age 18 instead of later in life. Thus I had four years of math education surrounded by female colleagues, learning in a very stimulating and competitive environment," said Milic-Frayling.
She later continued her graduate education at Carnegie Mellon University where she received a Ph.D. in Applied Mathematics from Carnegie Mellon University and afterwards taught at an all-female college near Pittsburgh. Before joining Microsoft Research Cambridge, she worked at Claritech, now Clairvoyance, a spin-off company from Carnegie Mellon. She showed strong initiative in both research and management and assumed the role of Director of Research.
She currently works as a member of the Integrated Systems team, designing algorithms and prototype systems in information management and communication technologies.
The group has recently focused on mobile information management. They worked on two projects, called SmartView and SearchMobil, that facilitate reading and searching Web pages on small devices such as PDAs and mobile phones. The underlying technology analyzes Web pages to understand its layout. It then devises the best strategy for displaying it on a mobile device. It also annotates elements of the page so that users can quickly see if it contains information they need.
One of her more specialized projects is text mining. She applies mathematical models to perform operations on text, such as classification or summarization. She wants to help people make better sense of the enormous amount of data that exists, to find relationships and learn from data. A simple search doesn't always work because the user doesn't know what question to ask the computer.
"If I wanted to find out about the projects related to machine learning on my website, I would have to think about how to word the query. The word machine learning is probably not even on my page, because machine learning is just a general term referring to different methods and models. But instead, if my documents could be classified in groups that have similar notions behind them, the system can help me find what I want. I wouldn't have to know the exact word. The algorithms automatically discover the related concepts using linguistic analysis and statistics."
Milic-Frayling likes the challenges that doing research for real-world problems provides. "In research," she said, "you can hypothesize that for every interest you have 100 example documents. That's a good sample size to be able to learn from and characterize the topic. But in the real world, I probably have only two or three. There's a dramatic change in performance if I have only two or three documents to learn from. So the question becomes, how do I modify the model to work under such constraints?"
Even if the trip to Mars never materializes, Milic-Frayling has her hands full in Cambridge, designing innovative algorithms and systems.
Advice: Know yourself, find your passion, work hard and take risks.
The Tablet PC is a wonderful invention that allows us to write in a natural way, but it has some trouble when it comes to recognizing handwriting that resembles chicken scratches. Wenli Zhu is a researcher in the Multimodal User Interface group at Microsoft Research Asia whose work will help users with less than perfect writing to correct recognizer errors in digital ink.
Zhu was inspired in graduate school to help users interact with technology. "I was in graduate school, working on an Industrial Engineering/Human Factors degree. The first time I used Microsoft Word, I was inspired by how the typewriter metaphor made it so much easier to learn the program. I later interned at two companies, designing and usability testing user interfaces. I found it fascinating that my work on understanding users, user tasks and designing a user interface could affect the way so many users work. That was the main reason I decided to come to Microsoft, because Microsoft products were used by millions and millions of people," she said.
Zhu has since become a member of the team that pioneered many new technologies for the innovative Tablet PC. Her team developed the feature that allows you to write in any direction on your tablet surface, and still have the computer recognize what you are writing.
"There are many ways that users interact with a computer today that are unnatural or hard. It is still difficult for users to access information anywhere, anytime. I'd like to work on projects that allow users to access information more freely and naturally, such as new mobile technology," said Zhu.
However, there is still more work to be done. "Technology can make people be more productive and better informed. Technology can improve the way people live and communicate," states Zhu.
Advice: Get a good foundation in statistics and in various machine-learning algorithms. To be truly successful in computational linguistics you also need to be a linguist at heart, with a solid grounding in linguistic analysis.
"Not enough time! Not enough data!" Those are the challenges that Lucy Vanderwende, a computational linguist, faces while trying to help computers process language.
Vanderwende, who is a linguist at heart, began her studies by focusing on the communication patterns we exhibit in conversation and narratives. After her first encounter with the combination of computers and language, she decided to switch from theoretical linguistics to computational linguistics.
"In theoretical linguistics, we might develop a theory about a syntactic phenomenon, but it is very difficult to test that theory, except by introspection or a limited amount of data inspection," said Vanderwende.
"Computational linguistics allows us to formulate hypotheses we can test on large amounts of data. Introspection is important for formulating the hypotheses, but with computational linguistics we can leave the testing to the machine." And computers don't inject their own culture or experiences into processing language, as humans do.
"Consider the phrase "chocolate bar" - we might think this means only one thing," explains Vanderwende. "But it is possible that "chocolate bar" is a place where you can go to drink chocolate milk -- like a wine bar --or a place where you can select various chocolates - like an olive bar. The computer doesn't yet have the commonsense that would prefer one reading over another."
Currently, Vanderwende is working on summarizing multiple documents that have common elements. "The idea is to make it easier to read small bits of interesting information on a wide variety of topics. If you want to read more details, you can go to the original documents. Some people might also think of this as an aid to browsing - by reading the summary, you know if you want to read the full document," said Vanderwende.
Another possible use would be to have the computer recognize the information you already have, and only give you new data. "For example, why would you want to read that Olympia is the capital of Washington State if you know that already? If we could provide multi-document summaries, and if we could personalize them, then I think most people would agree that the computer is doing something interesting and useful," said Vanderwende.
"Multi-document summarization is a project that will take a long time to do well. Almost all problems dealing with language are very challenging and are not amenable to a quick solution, or a tweak of an existing algorithm. There is a lot of engineering involved, even if one takes a machine-learned or statistical approach to solving the problem. And if you take a learned-approach to tackling problems of language, you need lots of tagged data to learn from, and tagging data with linguistic annotations is a very specialized skill and time-consuming," she said.
For the future, Vanderwende will continue her research to improve computer processing across multiple documents. One of her goals is to improve the collaboration between researchers with different specialties.
"There are serious challenges in making connections across different specializations. It would be nice if people involved in speech recognition, information retrieval, machine-learning, logic, reasoning, and linguistics contributed more to each others' research," she concludes.