SBIR-STTR Award

We propose to develop a system of mobile, autonomous, evolving agents for collaborative decision making, called EMMA. EMMA will be a neuro-genetic hybrid multi-agent collaborative decision making system based on an evolutionary control strategy for coordinating decision making agents. The system architecture will be two-layers, consisting of a central control agent and a distributed set of mobile collaborating decision making agents. The control strategy will use a neuro-genetic hybrid (NGH) approach to training and evolving autonomous decision making agents. The system will find information, evaluate it, and combine multiple results to offer recommendations and alternatives for action. Phase I will deliver a feasibility study and working prototype with one agent. It will prove the concept and establish the necessary architecture. Phase II will add multiple and mobile agents, and create a working collaborative multi-agent system. Design issues for EMMA include: mechanisms for inter-agent communication and coordination of problem solving; optimum implementations for a polling mechanism and for the genetic algorithm including selection mechanism and reproduction process and development of agent fitness measures; and incorporation of feedback from the user into the learning process.Anticipated Benefits/

Potential Commercial Applications:
A robust decision-making agent technology, which is not domain specific, will have wide application in the military, government, and corporate arenas. Any organization which must make decisions is plagued with too much information and not enough time to digest and understand it. EMMA will provide the needed decision support to organize and digest this information; generate, evaluate, and weigh alternatives; and make recommendations for action.

The goal of Phase II is the development of "EMMA" (Evolving and Messaging Decision-Making Agents), an agent-based decision support system incorporating natural language processing (NLP) techniques for user interaction and information extraction. Our Phase I research work successfully addressed the critical risks and technical challenges of the proposed system. Specifically, we successfully demonstrated the use of: - JKQML as an agent communication language - Hybrid agent adaptation scheme, employing both neural networks and genetic algorithms - A learning data fusion agent for merging results from multiple domain agents - A natural language interface for user-system interaction In Phase II we will produce a substantially enhanced system by: - Developing more complex agent behaviors through the introduction of additional local learning techniques and less constrained user interaction-system interaction - Integrating TextWise information retrieval and extraction technology to provide for dynamic knowledge base construction - Integrating knowledge base and expert system shells to support effective storage and retrieval of both extracted facts and derived rules. At the conclusion of Phase II we will deliver a robust demonstration system that will "forage" for information based on the users' needs and interest, presenting results in the form(s) most useful for user decision-making

Keywords:
adaptive agents collaborative agents decision support systems information extraction intelligent