Intelligent Agents with Learning Capabilities

Standard expert systems do not have learning capabilities. Their knowledge bases are built entirely by hand-encoding of an expert's knowledge. Such a process is time-consuming and prone to error. This project is concerned with the development of a PC-based expert system shell with learning capabilities. The system incorporates a knowledge base for storing rules and a data base for storing facts and examples. It has a learning program for rule acquisition, and a powerful inference mechanism.

The project is based on our earlier experience with ADVISE and AURORA systems. ADVISE is a large-scale inference system with rule learning capabilities and multiple control schemes. The system served as a laboratory for experimenting with methods for knowledge acquisition, multiple knowledge representation and machine learning. Aurora is a PC-based inference system, and an expert system shell that incorporates a program for incremental rule learning and improvement.

A related project concerns a method for discovering qualitative and quantitative models from data characterizing the behavior of a system. This method builds upon our experience with the ABACUS system for quantitative discovery. The current system is capable of determining a set of equations that fit a given set of datapoints, and a set of symbolic descriptions characterizing preconditions for the application of these equations. ABACUS integrates methods for data-driven quantitative discovery, concept learning from examples and conceptual clustering. This research has applications in building advanced expert systems and discovering quantitative and qualitative regularities in data. This project is being done in collaboration with AGH.

Bloedorn, E. and Wnek, J., "Constructive Induction-based Learning Agents: An Architecture and Preliminary Experiments," Proceedings of the First International Workshop on Intelligent Adaptive Systems (IAS-95), pp. 38-49, Melbourne Beach, FL, April 26, 1995.

Imam, I., "Intelligent Agents for Management of Learning: An Introduction and a Case Study," Proceedings of the First International Workshop on Intelligent Adaptive Systems (IAS-95), pp. 95-106, Melbourne Beach, FL, April 26, 1995.

For more references, see Publication section.