"Towards an intelligent agent for science data processing"

4/25/2003: 10:30am - 12:00pm
11th Floor Large Conference Room

Abstract: NASA is a data-collection agency. NASA-operated spacecraft, rovers, observatories and Earth-observing satellites return vast quantities of data, which are used to answer to questions ranging from the history and fate of the universe to health of farms and forests here on Earth. The quantity and diversity of data available presents huge opportunities and also significant challenges. Nowhere is this more apparent than in the Earth sciences, where terabytes of data are produced every day, and where the vision for the future is an Earth-spanning "Sensor Web," which detects and tracks changing events on Earth, such as severe storms and forest fires, while monitoring long-term trends in the evolution of the Earth system. Achieving this vision will require greater levels of automation in aggregating, processing and delivering data. Currently, almost all data processing is automated using inflexible scripts, which cannot be easily adapted to changing goals or circumstances. Our approach to this problem is to cast it as a planning problem. Data-processing operations, ranging from simple filters to sophisticated Earth system models, are represented as planner actions, desired data products are described using planner goals, and a planner is used to generate data-flow programs, composed of the available actions, whose outputs satisfy the goals. This sort of planning domain poses a number of challenges for planning, including very large universes, large plans, complex constraints, and the need to optimize for time, resource consumption (such and CPU, storage and bandwidth), data quality, and other factors. The talk will discuss our ongoing work in this domain and the approaches we have adopted to address these challenges.

About Keith Golden: Keith Golden is a research scientist in the Autonomy and Robotics Area at the NASA Ames Research Center, where he has worked for the past five years on problems ranging from rover diagnosis to software agents. His current interests include software agents, planning, constraint reasoning and knowledge representation. He is the PI of the IMAGEbot project, whose purpose is to develop intelligent agents for processing, tracking, archiving, and visualizing scientific data, and he is collaborating with Earth scientists to build such an agent for ecological forecasting. In 1997, he obtained a PhD in Computer Science from the University of Washington, where he was a principal architect of the first Internet Softbot, a planner-based software agent for performing tasks and finding information using Unix, the Internet and the Web. His dissertation was on "Planning and Knowledge Representation for Softbots." He received a BS, also in Computer Science from UCSB in 1992. Before that, he worked in game development.

Last updated: Mon Jun 19 17:44:06 2006