| This dissertation integrates the use of both transformational and derivational analogy into a general problem solving system. As the system faces new problems, cases are retrieved, adapted and subsequently generalized in order to enhance its performance. Empirical results show that the system can successfully address problems in simple domains and scales up smoothly to create solutions to problems in more complex domains including the management and processing of remote sensing and geographic information systems data for natural resource applications.; Most case-based reasoning systems rely on sophisticated indexing schemes and adaptation rules to find solutions to new problems. As a result, they expend considerable effort in retrieving and adapting cases to new problems. The approach presented in this dissertation introduces the use of generalization to case-based reasoning. Once a case has been retrieved and adapted to a new problem, the system will generalize the old case with the new case by using an algorithm similar to least general generalization. As the system gains experience, the case-base is generalized and, as is shown by the experimental results, the number of cases required to solve problems is significantly reduced.; To show that the approach scales up to real world problems, the system, dubbed PALERMO (Planning and LEarning for Resource Management and Organization), has been implemented and integrated into the SEIDAM environment. SEIDAM (System of Experts for Intelligent DAta Management) is a complex system that uses several AI approaches to manage large quantities of remote sensing and geographic data. It draws on expert system technology, software agents and case-based reasoning to gather and process remote sensing and digital geographic data. One of the goals of SEIDAM is to use remote sensing data to update digital forest cover maps to assist in land use decision making. |
