Share on Facebook Tweet on Twitter Share on LinkedIn Share by email
Demos and posters at TechVista 2011

Multilingual Name Search 

Although People Search has emerged as a very important search service, it can be significantly more effective and universal if it were multilingual. We have developed a novel multilingual name search technology that allows users to search people directories across languages while typing queries in language of their choice. Our technology is based on powerful new hash-based search techniques and can cross-language search a directory consisting of millions of names in a fraction of a second. The proposed demo is intended to show our multilingual name search technology in operation in the following application scenario: Multi-Lingual Address Book Search in Microsoft Outlook. Along with cross-language search, we will demonstrate fuzzy name search in English. 
Wikipedia on TV-DVD for Low-Income Communities

While many technologies remain out-of-reach for households in the developing world, one exception to this rule is that of entertainment technologies. Even in poor communities, there is a strong drive to own devices such as TVs and, increasingly, DVD players. Though they are typically used for video content, ordinary DVD players also support rich interactivity and programmability, including the capability to browse over 100,000 menus using the remote control.

In this work, we have stretched the limits of the capabilities of DVD players by mapping a large subset of Wikipedia to a TV-DVD. Our disc contains the entirety of, encompassing over 5,500 articles and 250,000 screens on TV. Using the interactive capabilities of DVD players, we have preserved all of the major functionality of Wikipedia, including indexing, search, hyperlinks, and forward/backward navigation. We expect that such DVDs will have a large audience in poor communities that often lack access to rich educational materials.


WikiBhasha is a multilingual content creation tool for Wikipedia. Developed by Microsoft Research, WikiBhasha beta enables Wikipedia users and contributors to explore and source content from English Wikipedia articles, to translate the content into a set of target languages, and to use the content with user additions and corrections for contribution to the target language Wikipedia. The content creation workflow is flexible enough to accommodate new content creation, at the same time preserving reusable information, such as references and templates.


SpecNet: Spectrum Sensing Sans Frontières 

While the under-utilization of licensed spectrum based on measurement studies conducted in a few developed countries has spurred lots of interest in opportunistic spectrum access, there exists no infrastructure today for measuring real-time spectrum occupancy across vast geographical regions. In this demo, we present SpecNet, a first-of-its-kind platform that allows spectrum analyzers around the world to be networked and efficiently used in a coordinated manner for spectrum measurement as well as implementation and evaluation of distributed sensing applications. The demo will showcase applications built using SpecNet APIs. 

You are the controller. No gadgets, no gizmos, just you!

Kinect brings games and entertainment to life in extraordinary new ways without using a controller. Imagine controlling movies and music with the wave of a hand or the sound of your voice. With Kinect, technology evaporates, letting the natural magic in all of us shine 

Enriching Education Through Data Mining  

Textbooks play an important role in any educational system and serve as the primary vehicles for delivering content knowledge to students. Unfortunately, many textbooks produced in developing countries are not written well and they often lack adequate coverage of important concepts. We propose a technological solution to address this problem based on enriching textbooks with authoritative web content. We augment textbooks at the section level for key concepts discussed in the sections. We have developed a system that uses ideas from data mining for identifying the concepts that need augmentation as well as to determine the links to the authoritative content that should be used for augmentation. Our demo will demonstrate the effectiveness of our system in enriching textbooks on different subjects and across different grades with high quality augmentations using automated techniques, over a corpus of Indian textbooks. 
Approximability of subspace approximation 

Given a data set of points in large dimensions, the subspace approximation problem asks for a linear subspace that "fits" these points well. We focus on particular subspace-fits defined using l_p norms, which generalize the ordinary least-squares fits. We give approximation algorithms for these problems based on semidefinite/convex relaxation and randomized rounding. Moreover, we show that it's hard to do any better (unless the Unique Games Conjecture is false).

(Based on joint work with Madhur Tulsiani and Nisheeth Vishnoi.) 

Rich Interactive Narratives  Information coming soon. 
SensorMap Based on Windows Azure   Sensor Map is a web-based platform developed by Microsoft Research Asia to collect and share data with localization information from different kinds of sensors. This platform provides rich research opportunities in the fields of data visualization and data mining. The infrastructure built on Windows Azure makes it can provide fast and scalable services for users and enable the development of wide spectrum of applications with large-scale data and computing ability.
GigaPixel camera  The Visual Computing Group of Microsoft Research Asia designed and built a digital tile-scan large format camera capable of acquiring high-quality and high-resolution image of static scenes. This camera can be widely used in virtual museums, culture heritage, and digit art preservation. We will show a sample to capture high-resolution pictures of wall paintings at Mogao Cave at Dunhuang, China.
Face Recognition in Video   Information coming soon. 
OneAlbum   Information coming soon. 
Separations among Predicate Encryption Schemes 

Predicate Encryption (PE) enables the embedding of certain attributes into a ciphertext and a Boolean function on those attributes into a key such that the key can decrypt the ciphertext if and only if the attributes of the ciphertext satisfy the Boolean function of the key. This powerful notion generalizes many types of public key encryption such as identity based encryption and attribute based encryption. It is useful, for example, to enforce fine grained access control based on properties of encrypted data and roles of entities/users.

One fundamental question about PE is whether it needs stronger cryptographic primitives than traditional public key cryptography. In particular, one can define classes of PE based on the Boolean function complexity class that the underlying predicates come from. In this context, one can specifically ask if PE based on a “small” complexity class A are strictly weaker than PE based on a “larger” complexity class B. Current results in the literature and our ongoing work gives evidence that this is likely the case, suggesting that PE is indeed a new generation of encryption methodology.

In this poster, we will introduce PE, give some examples, and explain our current work on separations among PE schemes based on Boolean function complexity.