Innovation Based on Recognition Research Platform

Invitation for Proposals 2008

Background

Microsoft Research Asia is creating a Recognition Research Platform. The platform is an open and extendable to support both research and developing applications in the pattern recognition area. The platform provides a unified workflow that includes related tools, such as data collection, labeling, pre-processing, feature extraction and selection, classifier training, and evaluation. Researchers can easily build new recognizers by following the workflow. The platform provides a data-sharing mechanism for researchers to contribute data, algorithms and leverage the inputs of other researchers. The platform’s backend computer cluster maintained by Microsoft Research Asia provides large-scale storage and computing resources. This helps reduce time spent on training and testing results.

The 2008 Mobile Computing in Education theme will focus on the handwriting recognition domain. The first version of Recognition Research Platform, to be accessed and used from the web browser, will be shipped in April. Before launching the platform, we are providing this thematic guideline to introduce the platform’s functions and invite proposals to innovate in building efficient handwriting recognizers.

Topics of Interests

In your proposal, you can cover one or multiple topics in the following three areas: collecting and labeling data, contributing recognition algorithms, and developing educational applications.

• Data collection and labeling
  Data is one of the important pillars of the Recognition Research Platform because large amounts of quality data are the foundation for learning-based pattern recognition approaches. The platform is targeted to build a rich data repository for pattern recognition community via encouraging individual researchers and research groups to collect and share data. The types of data in handwriting recognition include languages, symbols, and sketches. Example languages are Simplified Chinese, Traditional Chinese, Japanese, Korean and Tibetan. Example symbols are mathematical symbols and chemical symbols. Example sketches are categories of shapes, such as those in Microsoft Visio.
  
  In this area, you can propose to collect and label data in a domain of interest. We will add character sets of language, symbols and sketches you propose into the Recognition Research Platform for submitting and labeling data.
  
• Pattern recognition algorithm research and development
  The development of Recognition Research Platform comes from the support and contributions of pattern recognition community. The platform’s extendable mechanism enables the incorporation of various algorithm components in handwriting recognition.
  
  In this area, you can propose to research and develop handwriting recognition algorithms ranging from data pre-processing, feature extraction, feature selection, classification, and evaluation. The Recognition Research Platform will provide a consistent interface for you to upload and share research results.
  
• Creating pen-based educational applications with the recognizers trained using Recognition Research Platform
  In addition to facilitating the pattern recognition research, the Recognition Research Platform is also targeted to help apply pattern recognition and pen-based interaction techniques in the area of education. For example, a researcher can easily build an innovative digital ink application for teaching chemic if another researcher shares chemical symbol recognizer on the platform.
  
  In this area, you can propose a handwriting-based application to help improve education. You can leverage the recognizers to be shared on Recognition Research Platform or develop your own recognizer based on the platform to implement your proposed application.
  

Program Timeline

Back to Introduction page