此學位論文提出了一個保持距離關係的流形來維持資料間的鄰居關係。因為所有的輸入資料在流形的空間都有他們的對應單元,所以與輸入資料鄰近的單元會跟輸出資料的很像。而且此流形在資料座標系被位移、旋轉和縮放下都能保持不變。這些單元之間的鄰近關係會根據降低保持距離的能量函式的演算法,不斷的被調整和改善。 此學位論文還延伸前述演算法提出一個多層神經元核心,此核心將那些在同一類的所有資料對映到輸出層的同一點上且將異類的點對映到不同點上。這些大幅分離的類別點可以進一步地被用來做分類。此核心是一個階層式的前饋網路。每一層皆使用類別差異來做訓練,且一層接一層、從下而上獨立的訓練,類別的值並不被直接用在訓練過程中,故此核心可以掌握多類別區分問題。
This dissertation presents a distance invariant manifold that preserves neighboring relationships among data patterns. Since all input patterns have their corresponding cells in the manifold space, the neighboring cells of the input pattern resembles that of the output patterns. The manifold is invariant under the translation, rotation and scale of the pattern coordinates. And the neighboring relationships among cells are adjusted and improved in each iteration according to the algorithm of reduction of the distance preservation energy. This dissertation also extends the algorithm to presents a MLP kernel. It maps all patterns in a one class into a single point in the output layer space and maps different classes into different points. These widely separated class points can be used for further classifications. The kernel is a layered feed-forward network. Each layer is trained using class differences and is trained independently layer after layer using a bottom-up construction. The value of class labels are not used in the training process. Therefore, this kernel can be used in separating multiple classes.