By Rob Knies
March 5, 2013 8:45 AM PT
What is computing? These days, we tend to think in terms of devices and services, social computing and mobile capabilities.
But if there is anything we know for certain about computing, it is that our conceptions will change, our definitions will morph. Even as we speak, new, groundbreaking technology is being built. Revolutionary concepts are being explored and tested. Tomorrow’s sensation is today’s incubation.
That’s what TechFest is all about. Each year, Microsoft Research scientists from across the globe gather to display and discuss their latest projects, which encompass a broad spectrum of computer-science investigations. Some of these projects take a long-term, theoretical approach. Others will find their way into products soon to be in the hands of businesses and consumers the world over.
This year’s event, occurring March 5-7 at the Microsoft Conference Center, represents a glimpse into the future of computing technology, a time when those devices, services, social tools, and mobile ubiquity become even smarter and gain the ability to work on our behalf. A world of information technology will be transformed into a world of intelligent technology.
Such an assessment is based on a couple of fundamental technological trends. One is the explosion of data from machines, sensors, and people, combined with the availability of affordable cloud services, to enable powerful new tools to explore large data sets and detect patterns that can lead to insights previously unimaginable.
Another trend worth noting: more natural interfaces that deliver better—and sometimes completely new—computing experiences. With the ability to sense and understand the surrounding environment. such new interfaces can offer friendlier methods to communicate with computers, such as gestures, speech, and touch. The computing device then becomes less like a box with blinking lights and more like a trusted sidekick, anticipating your needs and providing effective guidance.
These directions are sure to play a key role in defining how we all soon will benefit from the research currently being conducted. Let’s explore the possibilities.
Addressing the ascent of big data is the goal of a project called Productivity Tools to Discover and Analyze Data, featuring technology called Inferno, created by a team based at Microsoft Research Cambridge led by Thore Graepel, principal researcher in the Machine Learning and Perception group.
Inferno is designed to put the power of machine learning—computers’ ability to examine vast amounts of information to detect patterns or learn how to react to certain situations—into the hands of people working with relatively small amounts of data using everyday programs such as Excel. The project builds on the foundations of Infer.NET, a powerful, flexible framework, developed at Microsoft Research Cambridge, for model-based machine learning,
“If you want to do machine learning, there’s often a lot of plumbing you have to do to get your data in, get the answers out, and validate everything,” Graepel says. “That involves a lot of coding or scripting.
“We thought, ‘What if we could take the machine-learning algorithms, put them where the data already is, and make it a seamless experience?’ That would make machine learning available to a much larger group of people.
“We’re talking about democratizing machine learning.”
Because many small or medium-sized data sets are in Excel, Graepel says, it made sense to try to bring the sophistication of machine learning to these small collections of information. But can a program such as Excel really perform the complex task of machine learning? Graepel says that Inferno—the name comes from its use of “inference” to perform its tasks—can do just that by building on Excel’s basic framework for organizing data.
For instance, Inferno can treat a set of data rows with all information present as a training set, which could represent past choices a group of consumers have made or sales data for different products, times, and locations. Inferno can “learn” what an Excel user wants to do with a set of data, and then, when presented with a row of data that has incomplete information, the technology can “infer” the missing information from experience obtained from the training set.
“We formulated the machine-learning problem here,” Graepel says, “as a missing-data-filling problem—an auto-fill program. While there exists an excellent Excel plug-in—SQL Server Tools for Data Mining—that provides machine-learning technology to data-mining experts, the goal of Inferno is to make machine learning a seamless experience fully integrated into Excel.”
Inferno also can conduct “unsupervised” machine learning, such as detecting anomalies or outliers, as with a data set that corresponds to measurements of a piece of human tissue. A change in the measurement could indicate a problem—a tumor, for instance. Inferno can learn to detect such anomalies.
Graepel sees a number of possible applications for Inferno. In business, it could help develop sales forecasts or help in lead generation for small businesses. In log data, it could help learn what caused a server fault. In stock analysis, it could examine trends and develop forecasts.
This project is just one of several, though, that examine the potential that harnessing big data could reveal. For example, a project called Adaptive Machine Learning for Real-Time Streaming, from Microsoft Research’s Advanced Technology Labs Europe, addresses how the processing of data in real time, combined with machine learning, can deliver on-the-fly updates to enhance customer scenarios in manufacturing and IT services.
Another intriguing project, ViralSearch: Identifying and Visualizing Viral Content, from Microsoft Research New York City, examines the concept of “going viral,” what that really means, and how such a phenomenon can be measured.
Another TechFest project, Telling Stories with Data via Freeform Sketching, shows how Microsoft Research scientists continue to make computing interfaces even more natural, in this case by making notes scrawled on a whiteboard more powerful.
Lee points out that sketching notes on a whiteboard has pros and cons. On one hand, the act of jotting down notes and talking about a problem fosters creativity and collaboration. On the other, the handwritten or sketched notes on the whiteboard rarely contain much real data. The information is implied or jotted down as “dummy” data.
SketchInsight takes advantage of advances in input technologies such as text recognition, gesture recognition, and pen-and-touch displays to create a “smart” whiteboard. When a team member starts to sketch something such as a chart or a diagram, SketchInsight pulls information from an existing data set, such as an Excel spreadsheet, determines what the goal is, and retrieves real data to complete the drawing. The power of cloud computing could be harnessed to provide an even more robust performance.
“People could start drawing parts of a visualization they are thinking of as a hint,” Lee says. “Then, the system automatically recognizes the elements of the drawing, consults relevant data sources, and infers and completes the target visualization.”
With that, a simple sketch that might have had little real data becomes a powerful discussion or presentation tool with concrete, relevant data.
The project, Lee says, could be useful in nearly any scenario in which people are brainstorming or problem-solving together—or a presenter is telling stories to others.
“SketchInsight’s presentation aspect would be really useful for education,” she says. “It could be used in lectures, where a lecturer could begin to sketch a complex diagram and the remaining part of the diagram could be completed with additional details pulled in from an outside source.”
The research behind Telling Stories with Data via Freeform Sketching is described in a paper: Understanding Pen and Touch Interaction for Data Exploration on Interactive Whiteboards, by Jagoda Walny and Sheelagh Carpendale of the University of Calgary, and Lee, Paul Johns, and Nathalie Henry Riche of Microsoft Research Redmond. The presentation aspect was designed in the summer of 2012 by intern Rubaiat Habib Kazi from the National University of Singapore, Lee, and Microsoft Research software-design engineer Greg Smith. Lee, Riche, Smith, and Carpendale continue to drive the project.
Natural user interfaces are being explored during TechFest 2013 in other contexts, as well, such as the 3-D Reconstruction by Portable Devices project, a Microsoft Research Asia effort to enable mobile phones and devices to generate augmented-reality scenarios.
Toward Large-Display Experiences takes a different direction, envisioning ways for people to navigate easily through wall-sized displays that seem likely to become a fixture in business scenarios before long.
There’s much more on tap, of course, and Rick Rashid, Microsoft chief research officer and head of Microsoft Research, will put it all in perspective during a March 5 keynote address to an audience of customers, academics, and partners, who will get a chance to explore the research projects firsthand. The following two days are reserved for Microsoft employees, who get crucial exposure to new ideas and technologies that could enhance and extend Microsoft products used by millions.
And you, too, get to participate. As usual, the Microsoft Research website will be offering a full array of TechFest-related information, and the Inside Microsoft Research blog will be delivering a series of real-time updates that shed light on the research accomplishments on display.
The future of computing is taking shape right now, and TechFest 2013 is certain to provide tantalizing glimpses into what tomorrow could bring.