Microsoft Research Connections Blog
Next at Microsoft
Social Media Collective
Windows on Theory
Posted by Rob Knies
Eight sentences. That’s all it took for Rick Rashid, worldwide head of Microsoft Research, to electrify a crowd of 2,000 students and faculty members in Tianjin, China, on Oct. 25 during the 14th annual Computing in the 21st Century Conference.
Why did those in attendance respond so rapturously for the conclusion of Rashid’s keynote? The answer was simple: He was speaking in English, but the largely Chinese audience was hearing his voice in Chinese.
Behind the scenes, a combination of powerful technologies was at work to bring the moment to life. One, by researchers at Microsoft Research and the University of Toronto, uses a technique patterned after the way people’s brains work, called Deep Neural Networks, which allows for speech recognition significantly more accurate than previous techniques. Another, by Microsoft Research, efficiently maps a person’s voice to another language. When these were combined with the engine behind Bing translator, the conference audience witnessed a dramatic new breakthrough.
After a few minutes of explanation, Rashid’s English-to-Chinese speech translation began in earnest. The audience murmured at first, but after a couple of sentences, each subsequent Rashid utterance was greeted with increasing excitement. When he finished by saying, “Thank you,” and the crowd heard “谢谢,” the auditorium erupted.
The highlight of the event, the demonstration underscored the intense interest exhibited Oct. 25-27 during the Computing in the 21st Century Conference, Microsoft Research Asia’s flagship event, co-hosted by Nankai University and Tianjin University, and the subsequent Asia Faculty Summit, conducted at the Shiing-shen Building on the Nankai University campus.
The theme of the 21st-Century event, attended by more than 2,000 students and faculty members, was Computing, Naturally. Hsiao-Wuen Hon, managing director of Microsoft Research Asia, used examples of cutting-edge results from his lab to explain how research efforts can be leveraged to provide the seamless, pervasive, natural user experiences to come. The concept of Computing, Naturally extends beyond natural user interfaces to include big data, machine learning, computational thinking, devices, mobile and cloud computing, trustworthy computing, social media, sensors and more. Altogether, they bring a more natural computing experience to users.
“For example, skills like speech or writing require years of learning,” Hon said. “With our continuous efforts in making learning more intuitive by the aid of natural computing, we can help communities learn to shorten the learning curve, and, over time, impossible things can become reality.”
Rashid, Microsoft chief research officer, underscored that theme during his opening keynote on Oct. 25, reflecting upon a recent Microsoft Research product contribution that has had far-reaching consequences.
“This year’s conference has the notion of ‘computing naturally,’” he said, “and I know many of you have probably seen Xbox Kinect. The Kinect part of Xbox was really the first time that real-time 3-D computer vision was made available to ordinary people in their homes. It has transformed the way that people think of how they interact with their computer.”
Another speaker, Michael I. Jordan, a professor at the University of California, Berkeley, discussed in detail the value machine learning is delivering to harnessing the power of unprecedented amounts of data generated in an era of near-ubiquitous sensing and information capture. He noted that data has become a computing resource, like time and space—with a significant difference that offers opportunity but also poses challenges.
“The funny thing about data is that it’s an unusual resource,” Jordan said. “It should be the case that, with any resource, the more of it I have, the happier I am. With data, that’s not the case.”
Eric Horvitz, deputy managing director of Microsoft Research Redmond, addressed machine learning during his Asia Faculty Summit talk, pointing to examples in health care, complementary computing combining human and machine perception, and integrative intelligence using a tapestry of components to provide a whole greater than its parts.
“Our grandchildren will expect elevators to not have to have legs jammed into them to open doors. They’ll expect that sort of intelligence to be in our environment,” he said. “That won’t even stand out as interesting. It’ll just be the right thing to have done. Applications of sensing, learning, and reasoning are still in their infancy, and they’ll bring unprecedented value to people and society as we mature these methods.”
Jeannette Wing, head of the Computer Science Department at Carnegie Mellon University, spoke on both days, first on trustworthy computing, then on computational thinking. In her talks, she pointed to a future in which computing will have evolved to deliver experiences that enable more natural ways of interacting with varied devices.
“Today or in the future,” she stated, “we’re going to build new security infrastructures that provide new trust relations, possibly unmediated. These are akin to today’s recommendation services and recommendation systems. There’s an economic ecosystem in this network of computers and humans that we have to understand. Security can bring us new mechanisms that can bring new economic value to interacting with each other over the Internet.”
She described a grand vision of computational thinking as a fundamental skill used by all by the middle of this century, “just like reading, writing and arithmetic,” whether it be in scientific research or in what is taught to undergraduates and even K-12 students.
“Computational thinking is the thought processes involved in formulating problems and expressing a solution in a way that computers—human or machine—can effectively carry them out,” Wing said. “Computational thinking is what comes before a computing technology. Computing technology has to come from somewhere, an idea someone pitched, and that’s the computational-thinking part.”
Other speakers provided equally fascinating scenarios. Daniela Rus, professor at the Massachusetts Institute of Technology, focused on robotics and extrapolated to future, robot-like assistance possible from the likes of “smart paper” and “smart sand.” Hong Tan, senior researcher at Microsoft Research Asia, discussed advancements in haptics that will provide touch-based feedback to users of touchscreens.
Peter Lee, corporate vice president of Microsoft Research Redmond, outlined the tradition of basic research leading to unexpected, game-changing results, and he championed the need to embrace diversity in computing research as a means to building a pipeline from blue-sky research to technological innovation. And a panel on disruptive innovation, moderated by Henrique Malvar, chief scientist of Microsoft Research, produced a lively, opinionated conversation including a couple of admonitions:
The theme for the Asia Faculty Summit, a key component of Microsoft Research Asia’s academic outreach in the Asia Pacific region, held for the 10th time this year, was Advancing Research and Education. A multimonth project from the team headed by Lolan Song, a senior director for academic collaboration at Microsoft Research Asia, the event attracted 300 attendees from the most influential research universities in China, Japan, Korea, Hong Kong, Taiwan, and Singapore, and many have collaborative projects with Microsoft Research Asia. On Oct. 27, those in attendance got a chance to preview more than 20 cutting-edge research demos.
The educational portion of the theme was the focus of the afternoon session on Oct. 26. In the first of three panel discussions, Anoop Gupta, a Microsoft distinguished scientist based at Microsoft Research Redmond, led a conversation entitled The Edu-Tech Tsunami: Scaling from 100 to 100,000, which focused on online education and the opportunities it engenders, particularly for emerging markets such as China.
“There’s been a tremendous amount of recent interest around online education and its transformational potential,” Gupta said. “Is it something that is going to stick? Is it just about large-scale courses? Is it going to transform education? There are a lot of questions that all of us have in terms of what technology and online learning can do for us.”
The answers to the questions Gupta posed, the panel indicated, all are affirmative. The confluence of rising educational costs and the broad availability of technological tools is increasing an individual professor’s reach by orders of magnitude. As Jordan commented later in the day:
“I want to train hundreds of Ph.D. students at a time!”
The final panel of the day, moderated by Baining Guo, assistant managing director of Microsoft Research Asia, addressed the need to sustain and grow a pipeline of technological talent emerging from academia. The gist of the discussion was crystallized by John Hopcroft, a professor at Cornell University and the winner of the prestigious A.M. Turing Award in 1986.
“If you’re going to have Ph.D. students do world-class research,” Hopcroft told a junior faculty member during a Q&A session, “it’s got to be research that they want to do, not what you want them to do.”
With that, and a brief overview by Tony Hey, vice president of Microsoft Research Connections, two days jam-packed with top-flight speakers and generously shared insights came to a close, with a few breakout sessions and a DemoFest on Oct. 27 completing the agenda. At heart, as became obvious during the events, technological researchers remain optimists with their eyes focused just beyond the horizon. Lee indicated as much.
“There is a potential for magic to happen,” he said. “It happens when those with a visionary product concept work hand-in-hand with world-class researchers.”