Task 3.1: Data to Information

[Bilmes, Fox, Guestrin, Jafari, Jones, Lee]

The volume and rate of city-scale streaming sensing data will challenge most state-of-the-art data analysis technologies. The goal of this task is to develop techniques that extract actionable information from the data flood. To this end, this task will enable deployment of large-scale, parallel machine learning algorithms on streams of data collected from the swarm. There are three key components in the proposed system:

• Adaptation of existing graphical modeling and machine learning frameworks, including GraphLab and GMTK, for on-line use over networks and for handling large volumes of streaming data.
• Machine learning toolkits encapsulating sophisticated algorithms and making them usable by swarm programmers who are not experts in machine learning.
• A Swarm Analysis Language, a high-level scripting language enabling users to quickly program large-scale machine learning applications on TerraSwarm data.

Finally, in addition to supporting detection and prediction tasks on the sensor streams, this task will enable users and applications to query the swarm for concise summaries of relevant information.