CamGraph Appliance for Scalable Machine Learning: Microsoft Research Product

Touch- and gesture-focused projects get most of the natural-user-interface (NUI) promotional love at Microsoft. One of these is CamGraph, an appliance for scalable machine learning, that is being codeveloped by Microsoft Research's FUSE Labs and Microsoft Research Cambridge (the Online Services and Advertising and Systems and Networking teams).CamGraph is one of the projects that Microsoft […]

Touch- and gesture-focused projects get most of the natural-user-interface (NUI) promotional love at Microsoft. One of these is CamGraph, an appliance for scalable machine learning, that is being codeveloped by Microsoft Research's FUSE Labs and Microsoft Research Cambridge (the Online Services and Advertising and Systems and Networking teams).

CamGraph is one of the projects that Microsoft showed off privately at its TechFest 2011 research fair in February.

In addition to being a NUI project, CamGraph also is an example of Microsoft dabbling more in "big data." Other Microsoft big-data-related projects and products include the "Dallas" DataMarket section of Windows Azure Marketplace and the Microsoft Dryad parallelization technology, which Microsoft ultimately plans to make part of Windows Azure.

CamGraph is aimed at making tools for inference in probabilistic graphical modeling available across distributed clusters of machines. CamGraph supports CamCube (originally BorgCube), a clustering architecture and software stack that can scale from 3 servers to 8,000 servers inside a shipping-container-type datacenter.

Here's more on CamGraph from Microsoft's Web site:

CamGraph leverages existing Microsoft Research work on infer.net to provide core libraries, and complex problems can be implemented in a few hundred lines of code. The platform also provides persistent data storage and management, in the form of a key value store. Semi-structured data can be loaded onto the CamGraph platform and then processed to build the graphs over which inference is to be performed.

The CamGraph researchers have been experimenting with three applications, Bing AdPredictor, Xbox Live TrueSkill and a "probabilstic triple store" to compare performance. The researchers have yet to publish their results.

[Via: All About Microsoft]