Chinese Researcher Leaps into Prominence
By Rob Knies
December 28, 2008 8:55 AM PT

Eight-year-old Dongmei Zhang clenches her jaw and peers intently at the long-jump pit in the distance. With a deep breath, she is off, running hard, accelerating in a sprint calibrated to launch her airborne inches before she reaches the onrushing foul line. Torso erect, feet projecting forward, she sails toward her sandy target. In seconds, it’s over. She brushes herself off, notes her mark, shakes her head: I can do better.

Even as a youth, Zhang had her eyes on the prize. Success was there for the taking, a product of desire and will, technique and training, diligence and focus. She ran the 100 meters and in various relays, starting in the third grade and continuing into her collegiate years.

But it was the long jump, that contest between quick-twitch agility and the immutable laws of physics, that captivated her. Even now, decades later, her attentions having been diverted to computer-science research, she credits her track-and-field career with having instilled the characteristics for achieving professional success.

“That has had a great impact on me, careerwise,” says Zhang, lead researcher and research manager of the User Interface Group for Microsoft Research Asia. “You get a sense of competitiveness. It’s sports, right? You try to run the fastest; you try to jump the longest. What sports gave back to me is a sense of staying competitive and doing your best.”

As an undergraduate at Tsinghua University in Beijing, Zhang was a regular participant in the university sports games, finishing second in the 100 one year and winning the long-jump event. She had learned her lessons well. Now, in an arena far removed from the oval track, they continue to bear fruit. Competitiveness, determination, the drive to achieve: These pay dividends in professional life, too.

Dongmei Zhang
Dongmei Zhang

“Dongmei is a very disciplined researcher who always gets things done,” says Hsiao-Wuen Hon, managing director of Microsoft Research Asia. “Her unique experience of research and product development make her jump out in turning ideas into Microsoft product reality.”

Zhang says the direction of her group is “to research and develop enabling technologies and tools to help users explore, understand, and utilize data from various sources in the information-explosion era.” And she has exhibited a pattern throughout her career of taking on daunting challenges and delivering results. What’s interesting is that, having established that reputation while working in product development, she is extending it as a researcher.

Harry Shum had a great influence on my decision to join Microsoft Research Asia,” she says. “At that time, Harry was the managing director of the lab. He talked with me about the exciting research, as well as potential career opportunities, and gave me the opportunity to see it myself. Here, people are not only publishing papers. They really want to contribute to the company’s products. Because I have both a research background and product-development experience, I can use those strengths to do more things to contribute to the company, which is why I decided to join the lab.”

That decision came in 2004, and Zhang hit the ground running. She took the torch from Jian Wang, who pioneered digital-ink research for Microsoft Research Asia, and started to lead the lab’s research into more areas of digital ink. Her work has been pivotal in a couple of projects that have produced technologies that are enhancing Microsoft products.

Her Handwritten Math Equation Recognizer makes it easy to convert a handwritten complicated mathematical equation into an accurate, digitally rendered representation that can be used in a report or a presentation. It was released to the public in 2005 as part of the Education Pack for Windows XP Tablet PC Edition.

Then there’s her team’s East Asian Language Recognition technology, which greatly improves the accuracy of digital recognition of handwritten text in traditional Chinese, simplified Chinese, Korean, and Japanese. That work was transferred to the Windows Experience team in July. The huge accuracy advancement of Microsoft East Asian Recognizers achieved by the joint effort of the Windows East Asian Recognition group and Zhang’s team caught the admiring attention of Microsoft Chairman Bill Gates.

Gates learned about the East Asian Recognizer on June 27, which happened to be his final day in his hands-on role at the company. He was gratified with the results.

“This is fantastic to see,” Gates said at the time. “Microsoft is committed to making the Tablet mainstream, and this excellent work will make a big difference, particularly in countries where the nature of the written language means pen-based computing can be much more efficient than using a keyboard.”

For Zhang, it’s been an interesting journey from Beijing to Pittsburgh to Redmond and back home again. Hailing from a family of engineers—both her parents are aerospace engineers, and her brother is a mechanical and electrical engineer—she learned BASIC in junior high school and found herself fascinated with computer science while at Tsinghua University.

At the same time, she began thinking about extending her studies abroad. Zhang was one of six young women rooming together in a dormitory; they all had their sights set on the United States. Two of the roommates attended the California Institute of Technology, one went to the University of Texas at Austin, and one chose Rice University. Zhang and another picked Carnegie Mellon University.

“It was interesting,” she recalls of her introduction to a new country. “I still remember when the plane landed in L.A. I was by myself, trying to make the connection. I did not feel that strange, although that was the first time I went abroad. I just felt the excitement. Everything was new.”

The biggest transition she had to make was to gain increased familiarity with English.

“I could talk,” she recalls, “but I could not talk as freely as I would in Chinese, so that’s something I had to pick up very quickly.”

Zhang, though, is gregarious by nature, and that trait served her well.

“In general, I’m just a very open person,” she says. “I like to talk to people. I’m not afraid of talking or asking. As long as you ask nicely and you are friendly and open, you can learn quickly.”

THE DOCTORAL PATH

She was enrolled in Carnegie Mellon’s’ Robotics Institute and spent a couple of years enmeshed in her studies. The experience was eye-opening.

“When I was studying at Tsinghua, I was proud of my programming skills,” she smiles. “But I found that I was not that good yet after I joined CMU. You still have to learn. You have to practice. You have to do everything by yourself, and you have to do it quickly. That’s quite challenging.”

Eventually, she began to pursue research projects, and 5½ years after arriving in Pittsburgh, she earned her Ph.D. and took a job with an Atlanta startup called Alventive to work on 3-D modeling—or so she thought.

“That was an interesting decision,” Zhang says. ”Usually, after you get your Ph.D., you think about going into academia or you think about going to industry labs doing research. I was thinking: ‘Well, I’ve done some research, and I’ve published some papers. What I really want is to get something real, not only on paper.’ I had a strong desire to do something real.”

What she got was instruction in how the real world can work. Just before starting her new job, Alventive changed direction. While part of the company would continue to pursue 3-D computer-aided design, the bulk of the workforce would have a business-to-business focus. As a Carnegie Mellon Ph.D., Zhang, a prime hire, was allowed to choose the area in which she wanted to work.

“It was kind of surprising,” she says, “but I thought, ‘Maybe it’s a good thing.’ I am very confident in picking up new knowledge and skills, so instead of doing what I was familiar with, I just said, ‘Well, let me go with the majority of the company, working on the new stuff.’ Everybody was kind of new to this area, and so what you could do was really up to you.”

It was a productive 18 months.

“That year and a half really gave me good industry experience,” Zhang states. “When I later joined Microsoft, the culture and environment was not that unfamiliar to me. I already knew how things worked, how you should work with people, be respectful and professional. I also learned the importance of teamwork.”

But there were challenges, too, not the least of them the fact that, while she was working in Atlanta, her husband, Gary Zhou, was in Redmond, working as a software engineer for Microsoft. In 2001, Zhang took a job with Microsoft’s Digital Media Division. Once again, she found herself having to make a quick adjustment.

“I was working on an audio-related project,” she recalls, “and I had finished more than 60 percent of the work.” But a month before the product was due to be code-complete, the feature was cut from the product.

“All of a sudden, I didn’t have a primary feature to work on,” Zhang says. “Our team lost an important feature, too. We were thinking, ‘Well, what shall we do now?’ “

Zhang’s leader took initiative and proposed working on Photo Story, a product designed to enable consumers to create videolike tales from collections of still photos, augmented with music. There was one catch, though: They had one month to do it.

”My lead said, ‘If you think you can take care of the video engine, I will take care of the user-interface part,’ Zhang smiles. “To me, it was challenging, but also exciting. I really like to take these kinds of challenges, to get things done.”

Amazingly, they not only hit the code-complete milestone in time for the basic set of features of the engine and the UI, but also were able to squeeze in a few other features in the stabilization phase, and when Photo Story was released, theirs was the top feature in the package.

A couple of years later, Zhang’s husband moved back to China to join Microsoft’s Advanced Technology Center, and a few months after that, she found herself talking with researchers at Microsoft Research Asia.

“I realized that Microsoft Research Asia was really a good fit for me,” she says, “because I had both a research background and product-development experience. I could learn and achieve a lot from working with so many talented people with various backgrounds.”

HOMECOMING

In September 2004, she found herself back in Beijing. Not long thereafter, she began working on Handwritten Mathematical Equation Recognition. The project had started almost two years earlier, and a lot of the infrastructure work was already done. But there was still a gap between the prototype and a quality product that the Tablet PC team could ship. What was even more challenging was that there was not much time left for the team to fill the gap. But that was something Zhang had experienced before.

“I made two decisions to drive this project,” she says. “I invited a researcher from the Speech group, Jian-Lai Zhou. He was smart, a very good researcher, and I knew he would be able to contribute. The other thing was to work hard to improve the training infrastructure and clean up the training data. We train the models based on data. If the data is noisy or incorrect, then you just cannot get a good model from it. The training is time-consuming and requires a lot of computing power. It’s important to have an efficient training infrastructure in place in order to shorten the training cycle.”

And they only had a few months.

“We were under pressure for release,” Zhang says. “We had research to do to improve the accuracy. We had engineering work to do to create a product, not just a prototype. I had to plan everything very, very well, the product side and the research side, to make sure they aligned.”

Align they did, with significant contributions from Jian-Lai Zhou, Yu Zou, and a few other team members. The product shipped, with accuracy much improved. There was, though, a sad postscript. Jian-Lai Zhou, whose work had been instrumental in the success of the project, passed away from cancer in October 2006.

“That was so unfortunate,” Zhang says. “Jian-Lai was a great partner and a talented researcher.”

The Tablet PC team, suitably impressed with the equation recognizer, asked for help in improving the accuracy of its East Asian-language recognizer. First, Zhang’s team—including associate researchers Zou, Shi Han and Ming Chang—developed technology to recognize individual handwritten characters, and then it tackled the challenge of recognizing entire lines of handwritten characters. The advances are due to be released with the next version of Windows.

Now, Zhang and colleagues are working on the Recognition Research Platform (RRP), an effort to provide a community research platform for handwriting-recognition research across different domains.

“We thought, ‘Can we do something bigger, something that has more impact?’ ” Zhang says. “When you are working on this kind of project, you need to do quite a bit of infrastructure work. You need to have tools to collect and label handwriting samples, and you need tools to do model training, which is computationally expensive. After that, you need to evaluate how good your model is. The whole thing is a workflow and every step is important.

“The platform we want to provide is targeted to provide the infrastructure of this workflow. For example, as a member of the RRP research community, you could use our tools to collect and share data with other members of the community. Through sharing, researchers will be able to access much more data, which is important for learning-based recognition research.”

In facing the challenges she has surmounted in recent years, Zhang has learned a few lessons on how to succeed amid the challenging demands of software development and research.

“First of all,” she says, “if you want to be successful, you need to equip yourself with the knowledge and the skills you must have. The second thing is that you really need to communicate well with people, to respect people, to understand the importance of teamwork.

“And another thing: It’s true that engineering or science is hard. Compared with other areas, you need to devote more energy or time to really master the essence of it. But on the other hand, it’s also cool and fun. When you see the product or the research you have done, when you see that go into the hands of millions of people, you see people benefit from it, and you see it changing people’s lives, the kind of satisfaction you get is just unbelievable.”

Staying competitive and doing your best: traits that extend to the long-jump pit—and beyond.