Project Number: 5438-32000-025-01
Project Type: Specific C/A

Start Date: Sep 01, 2002
End Date: Jun 30, 2005

Objective:
The objective of this cooperative research project is an infrastructure of software agents to support data acquisition, mining, integration and analysis; and sequence annotation. The result will reconcile differences in syntax, semantics, pragmatics, and processes enabling disparate tools and databases to be composed dynamically by end users and interoperate seamlessly.

Approach:
The implementation will entail first preparing a detailed specification of the software agent-based infrastructure and of the tools and data sources. The specification will require the definition of an ontology that can be used to reconcile the schemas of the data sources and the semantics of the application programs that use the data from the sources. The investigators will also produce a characterization of the semantic translations that will be applied to the ontological descriptions of the sources and tools. The agents will apply the translations to the data and control messages that are exchanged by the agents. Overall, the system will behave as a distributed agent-based workflow execution system. The architecture of the agent-based system will be a mediated one, containing a registry of services, an ontology agent, user agents, analytical tool agents, and database wrapping agents. The system will implement and adhere to current Web service standards. Agents will access information from data sources, apply a Bayesian analysis to the information to produce or specialize scenarios about pathogens, and store the results in a database for later display and use. Pathogenic-threat situations will be modeled as Bayesian networks and influence diagrams, annotated with information about the who, what, when, and role of the nodes and links in the diagram. Received events will be used to find the models in the scenario base that best fit the observed events. The models will be customized and presented to a user, who can then modify and refine them. A Bayesian analysis will be performed on the result, which can then be used to explore what-if scenarios, and can determine which missing pieces of evidence would be most useful in refining the predicted outcomes of the model. The scenarios will incorporate models of the domains (such as the pathogens, their geographic locations, and their environment), and models of the information processing systems (such as limitations of various data-mining algorithms). An agent-encapsulated Bayesian network formalism provides a single uniform representation for all of these models.

Project Team

Keele, John Laegreid, William Kappes, Steven Michael Huhns - Director, Center For Information Technology, Usc 402-762-4251

Project Annual Reports

FY 2003

Related National Programs

Food Animal Production (101)