Connectionist learning of weights for k-NN retrieval

Case-Based Reasoning (CBR) systems support decision making in ill-structured domains. Retrieval of appropriate previous cases is critical to the success of a CBR system. The problem of assigning static attribute weights remains unsolved because of the very nature of the domain. To deal with such problems connectionist approaches have been proposed. This paper describes such a connectionist approach for learning weights for a CBR system. The teaming system takes as input the discrepancy between CBR ranking of cases and the choice of a domain expert. It then propagates this difference back into a perceptron to adjust weights for the k-NN match function. This paper describes test results on a publicly available soybean disease database, where the category of each case is known a priori.