Research On The Congnitive Improvement Strategies And Learning Ability For Case-based Reasoning

Case-based reasoning (CBR) is originated from the exploration of human rea-soning and learning mechanism in cognitive science, and it is a new reasoning me-thod in artificial intelligence. A lot of researches and applications can be found inpattern recognition, regression prediction, product design, etc. During these re-searches, the focus is on how to improve the reasoning performance and learningability of CBR. The open questions which influence the reasoning performance ofCBR are case retrieve, case reuse, case revise and case retain, of which, making agood retrieval strategy is the key to improving the reasoning accuracy; selective casememory with forgetting function can significantly improve the reasoning efficiency.Therefore, to improve the reasoning performance and learning ability of CBR, theintrospective learning principle and memory and forgetting theory in cognitivescience, and the group decision-making thought are respectively introduced to rede-sign the case retrieve and case retain. The main contents are as followed:(1) In order to improve the learning ability of static weight distribution method,a convergent iterative method for adjusting the attribute weights based on introspec-tive learning principle from congnitve science is studied. Attribute weights can beiteratively adjusted according to the problem solving performance of CBR classifier.Based on the success-driven strategy, when a new problem is solved successfully,the weights of matching attributes should be increased, meanwhile the weights ofunmatching attributes should be decreased, therefore the weights can be distributedmore reasonably, which establishes the foundation for similarity measure and casematching.(2) In order to solve the problem that the single case retrieval cannot make fulluse of the information of case library, the thought of group decision-making thatutilize the group congnitve ability to solve the problem is introduced to improvecase retrieval strategy, and a convergent case group-retrieve is proposed. First mul-tiple sets of initial weights are obtained by genetic algorithm, second each set ofweights are iteratively adjusted by the introspective learning method, then the ad-justed weights are used as group weight to calculate the similarities, finally the group decision-making retrieval result which satisfies the plurality rule can be ob-tained, which provides a foundation for case matching and reuse.(3) Aiming at the utility problem caused by the incremental learning of CBR, adynamic maintenance method for case base by memory and forgetting principles incognitive science is proposed. In this method a review step is added after the mod-ified retain step, and thus the new congnitive model for CBR with functions of se-lective memory and intentional forgetting is constructed. It can selectively store thenew case by the memory strategy, update the forgotten values of the retrieved near-est-neighbor cases and delete the cases according to the forgetting strategy, whichprovides the guidance for the deletion of redundancy cases and the dynamic controlof the case base.(4) In order to test the validity of the improved CBR cognitive system, someexperiments are conducted. With the classical classification datasets from Universityof California Irvine (UCI), a series of5-folds cross validation experiments are de-signed to test the validity of iterative weight adjusting method, case group-retrieve,and dynamic case base maintenance. And the results show that these methods canrespectively improve the quality and the efficiency of problem-solving by CBR.Then the three methods are applied to the CBR as a whole, and an auxiliary diagno-sis system for cardiovascular disease is established. Comparative experiments arecarried out on the accuracy, sensitivity, and specificity. And the results show that thismethod can improve the reasoning ability and learning performance of traditionalCBR system effectively.