本篇論文旨在探討與評估類神經網路分類器，且探討並分析。靜態類神經網路和動態類神經網路兩種不同形式分類器。此兩種分類器本質上是屬於不同的結構和演算形式。靜態類神經網路在結構上，為一固定結構，即其網路神經元數是靠經驗法則、人工給定的，而動態模糊類神經網路在結構上為一種動態調整，其網路神經元數是靠一系列的學習規則所衍生的。其中倒傳遞網路，主要對學習演算法採用Levenberg-Marquardt 方法改善演算的收斂速度。其次，動態類神經網路中探討部分，分為兩部分:架構的學習和參數的學習，在其架構上用刪減技巧使結構更精簡、更容易去實現。最後在實驗結果方面，利用UCI 樣本資料庫進行分類處理，以評估兩種分類器的準確率。 This thesis aims to investigate and evaluate neural network classifiers, especially on back propagation neural network and dynamic fuzzy neuralnetwork. And we further analyze and improve of both classifiers to ensure the high accuracy of internet. In back propagation neural network, we mainly focus on the learning algorithm and adopt the Levenberg-Marquart method to improve the performance. Moreover, the discussion of the dynamic fuzzy neural network could be divided into two parts: structure learning and parameter learning. The optimal parameter learning is the main work in this study. And it is used by the pruning techniques for dynamic fuzzy neural network structure and would lead to an easy operation for internet, structure simplification and facilitating the accomplishment. Finally, from the experimental results, the classification is made from the UCI database to evaluate the accuracy of both back propagation neural network and dynamic fuzzy neural network classifiers.
