85級畢業碩士　林晉洲　發表論文

This paper presents a multiclass neural network classifier to learn disjunctive fuzzy information in the feature space. This neural network consists of two types of nodes in the hidden layer. The prototype nodes and the exemplar nodes represent prototypes and exemplars in the feature space, respectively. This classifier contains two separate training algorithms. The pass 1 training algorithm DYNPRO (DYNamic numbers of PROtotypes) automatically generates and refines prototypes for distinct clusters in the feature space. The number and growth parameter of these prototypes are not restricted, so the prototypes will form near-optimal decision regions to meet th e distribution of input patterns and classify as many input patterns as possible. Next, the pass 2 training algorithm GFENCE (Generalized Fuzzy Exemplars Nested Creation and Expansion) is used to place and adjust exemplars to learn the patterns that cannot be classified by the prototypes. Such a training strategy can reduce the memory requirement and speed up the process of nonlinear classification. In addition, on-line ability is supplied in this mode l and the computational load is lightened. The experimental results manifest that the appropriate number of prototype nodes used to represent patterns clusters can be determined by this model, and the recognition rate for disjunctive fuzzy information evaluated by this model is encouraging.