Agents in ARM: Applying Artificial Intelligence to ARM Data Mining
Kuchar, O.A. and Reyes-Spindola, J., Pacific Northwest National Laboratory
Fourteenth Atmospheric Radiation Measurement (ARM) Science Team Meeting
We present a vision of a prototype environment that utilizes a co-operative community of intelligent software agents (a computer program that behaves in a manner analogous to a human agent) for the creation of an integrative, computer-based data analysis architecture to mine massive and complex atmospheric data sets in the ARM archive to aid in atmospheric research and discovery. In the discovery process, scientists ask questions that result in requiring multiple data streams and certain data requirements. ARM has been collecting atmospheric data from different sources and locations for over a decade, resulting in terabytes of data. To sift through this data, a scientist must undergo a laborious, time-consuming, and iterative process accessing the massive and heterogeneous data (data sets vary in time and space resolution) in the ARM archive to find cases that meet their requirements. Researchers use this data to create an information landscape and gain insight into their problem. To address this data retrieval, we envision an environment that aids researchers by integrating an intelligent software multi-agent system with a cognitive model framework. This framework will use domain-expert knowledge to aid in the high-level planning of how to decompose a large problem into sub-tasks. The cognitive model then solves each of these sub-tasks based on high-level user-defined objectives, expert knowledge, and resource/agent availability. The multi-agent system then functions as the end effector of the cognitive model framework by working as a retriever, filter, and transporter of data for its incorporation into a final solution. We are working closely with atmospheric scientists at Pacific Northwest National Laboratory to: elicit knowledge for populating our cognitive model; understand the scientists reasoning process when analyzing potential data sets; integrate their current tools into our system; and determine how our environment should present its results to a scientist upon completion. We will present our process and potential solution to the ARM community.
Note: This is the poster abstract presented at the meeting; an extended version was not provided by the author(s).
