Product Contributions 2011
- Academic Map
Researchers utilized various visualization techniques for this addition to the original Academic search platform. These enhancements allow a richer search and browsing experience in Microsoft Academic Search.
- Prediction technology
Researchers developed prediction technology incorporated into a Microsoft Amalga solution to predict hospital patient readmission risk at the point of care. Technology enables decision makers to see a quantitative risk probability for readmission within 30 days of discharge based on multiple data features at the point of care and at the time of discharge. This information, along with clinical judgment, can help prioritize resources, the level of post-discharge support, and necessary follow-up for the patient. This innovative approach provides a new data point, not previously available, that is critical in managing hospital readmissions.
- Semantic image tagging
The technology of semantic image tagging enables automatic detection and localization of anatomical structures within 3D CT scans. Researchers created a library adopted by Microsoft Amalga to recognize the various organs within CT images. First known as InnerEye, this work employs visual recognition that focuses on the analysis of patient scans using machine learning techniques for automatic detection and segmentation of healthy anatomy, as well as anomalies.
A project out of the eXtreme Computing Group, Orleans is a software framework for building client+cloud applications. Orleans encourages use of simple concurrency patterns that are easy to understand and implement correctly. It builds on an actor-like model with declarative specification of persistence, replication, and consistency and uses lightweight transactions to support the development of reliable and scalable client+cloud software.
Researchers created PingMesh, a full-mesh latency monitoring system that provides a key capability for deep understanding on end-to-end network latency characteristics crucial for Bing’s applications and services.
For those seeking to learn a foreign language, researchers created an approach to flash card–based learning in which news items are drilled into short-term memory for immediate prospective use.
Microsoft Research Asia contributed a new feature for Engkoo to make it easier for ESL Chinese users to complete their English tasks in Bing.
- Map improvements
Researchers transferred code to deblur the aerial images that the Bing Maps team acquires and publishes, making the acquisition process less costly and the resulting Bing images higher quality for an overall better user experience.
- Search results improvements
o Researchers delivered PNav, an algorithm that predicts with high accuracy which Web search results a user will click for repeat queries. Teams also made improvements in delivering significant gains in personalized relevance for Bing and developed a new engine for driving directions on continental road networks with arbitrary metrics.
o Large-Scale Face Image Search was incorporated into Bing multimedia search to index and show images from a search query.
o Researchers created a new corrective spelling system deployed in Bing that uses statistical machine translation techniques.
Researchers collaborated on static code analysis that allows SCOPE developers to discover certain types of issues and defects earlier in the development cycle before actually submitting a job to a Cosmos cluster.
o Researchers contributed the Semantic Suggest tool to refine users’ search and now allow users to quickly find the information that is most relevant to them. The team removed a step and added the refinements directly to the search box without having to filter search results by category, brand and price point.
o Researchers created algorithms to enable users to select a product and compare images of visually similar clothing and shoe products on Bing Shopping.
- Video Search
Researchers created a URL normalization technology that helps remove duplicate video pages in crawling and repository. They also contributed to site modeling technologies that automatically detect metadata from video websites to enable better video search functionality in Bing.
GFS (data centers)
- Microsoft researchers worked with several teams on power models for large disk enclosures to help GFS remove expensive power monitoring hardware from data centers.
- Touch Mouse
After two years of research and collaboration among Microsoft Research, the Microsoft Hardware team, and the Microsoft Applied Sciences Group, the Touch Mouse concept made its way successfully to consumer shelves. The award-winning product, unveiled at the 2011 International Consumer Electronics Show, combines the virtues of a traditional mouse with the rich natural language of gesture through novel hardware and software design.
- Microsoft researchers adopted and modified the PageRank algorithm that helps to effectively detect tens of millions of malicious Hotmail accounts.
- Debug Advisor
Microsoft researchers updated the Lync Debug Advisor, which is a recommendation system that uses enhanced full-text search to find bugs that are similar to a given “fat” query. The newest version was built on FAST.
Microsoft researchers contributed to a real-time social Q&A system that connects askers to answerers in real time via Lync instant messaging. This publicly available system allows employees within an organization to have their questions answered quickly in the workplace.
Powered by machine learning and data mining techniques, researchers developed SAS, an integrated system that helps SharePoint engineers efficiently analyze performance issues in SharePoint Online based on a huge volume of performance monitoring data, such as performance counters, system events and transaction logs generated by various SPO components.
- Researchers contributed to the design and development of enhanced workload provisioning and patching support in DCS Hydration Pack 2.0 and System Center 2012.
- .NET Bio
Formerly Microsoft Biology Foundation, researchers created .NET Bio, a language-neutral bioinformatics toolkit, as an extension to the Microsoft .NET framework. Currently, the toolkit implements a range of parsers for common bioinformatics file formats and a range of algorithms for manipulating DNA, RNA and protein sequences. .NET Bio is freely available under an open-source license.
Researchers integrated and leveraged the Visual Studio database project system into CoreXT to treat database schema and artifacts as first-class source code.
- Debugger Canvas
Debugger Canvas is a new way for developers to debug C# and Visual Basic code. Created as a collaboration between Brown University, Microsoft Research, and Visual Studio, Debugger Canvas is a pan-and-zoom display containing the parts of the code through which the user has stepped (by using the debugger) or visited (through navigation commands, such as “go to definition”). Debugger Canvas presents the code the user explores as a call-graph diagram in which each node contains the method’s body in a full-featured editor.
- Try F#
Researchers implemented Try F#, which enables the .NET language F# to be used in a browser-based development environment, thus making it easy for everyone to try out F# across operating systems.
Microsoft researchers implemented an ARM keyboard, mouse, and Ethernet windows driver that works with the simulated hardware of the ARM Fastmodel Simulator and allows users to interact with WOA using mouse and keyboard.
- Windows Embedded
Microsoft researchers devised a performance enhancing way to stream high-definition video over wireless link in Windows Embedded.
Microsoft researchers provided a power monitoring and modeling tool to the Microsoft OEM team that they licensed to Samsung which helps them run Windows more efficiently on their hardware.
Microsoft researchers used machine learning and data mining techniques to implement StackMine, a scalable performance analysis system for analyzing performance bottlenecks from large-scale PerfTrack trace sets.
Microsoft researchers developed sensor fusion algorithms to derive absolute 3D orientation from a phone’s accelerometer, magnetometer and gyroscope.
- Project Hawaii
In partnership with universities around the world, Project Hawaii enables students to develop inventive cloud-enhanced mobile applications. Researchers contributed to early adoption, prototyping and provided feedback to the Windows Phone team to release the Optical Character Recognition (OCR) service which delivers a more robust service for the Windows Phone Mango update. Project Hawaii provides the tools, services, and mobile-plus-cloud platforms that students need to create their applications; students bring their creativity and imagination.
- Speech recognition
Microsoft Research developed a codec that improves the accuracy of Speech@Microsoft’s cloud-based speech recognition systems, such as Bing Mobile Voice Search. They also deployed a toolkit for large-scale discriminative training of GMM-HMM based acoustic models for large-vocabulary continuous speech recognition.
- Touch keyboard
Microsoft researchers collaborated to reduce error in latency-based hit-target resizing on the Windows Phone keyboard.
- Kinect for Windows Software Development Kit
Microsoft researchers collaborated with the Interactive Entertainment Business team to implement the machine learning algorithm needed for the non-commercial SDK, which provides tools, compilers, headers, libraries, code samples, and a new help system that developers can use to create Kinect applications that run on Microsoft Windows.
o Researchers collaborated to create a new recommender system for games, videos and music. The team used a Bayesian Matchbox algorithm for large-scale recommendations as a starting point for inferring user preferences. The initial version is now serving over 35 million users.
o Researchers developed techniques to combine accelerometer, magnetometer and gyroscope sensors to derive steering wheel heading.
o Researchers applied gesture recognition technology to the problem of recognizing and distinguishing different baseball throws. By collecting data and making changes to the recognition system, the team delivered improvements to Kinect Sports game.
- TV (Leibniz)
o Researchers leveraged machine learning algorithms to create the ability to sync technology, product, and service for matching movies and TV shows/episodes. Leibniz was first deployed in Bing as the basis for Entity Actions, making Bing a task engine, starting with automatic tasks such as renting and streaming movies and TV shows.