Implementation of MLP based Nonlinear PCA and Its Applications
| Autor: | Zheng-You Guo, 郭政佑 |
|---|---|
| Rok vydání: | 2018 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 106 Artificial neural network is mainly used in the classification and prediction and improve the accuracy of prediction and classification. In order to improve the learning, data must be learned from the large of the past statistical data. In general, the past statistical data has thousands to tens of thousands of data and it is very time-consuming learning on learning procedure. Therefore, we proposed a structure different from traditional multilayer perceptron (MLP) network methods and a way different from traditional learning methods to load data directly. Firstly, we extend the traditional five-layer MLP network to the seven-layer MLP network under tensorflow to implementation nonlinear dimensionality reduction based prediction with nonlinear principal components analysis (PCA) and circular PCA. During compression, we required that the data features can be completely preserved. During the reconstruction, we also required that there is no missing information of the original data. Using the features of nonlinear dimensionality reduction as input on the proposed structure is proposed. Consequently, the proposed structure has better performance than the traditional MLP neural networks. Secondly, we also proposed to convert the data into distributional-valued variable, not only can it is easy to represent large amounts of data, intuitively indicating the shape of the data being distributed. Besides, this transformation allowed us to understand the complex distribution of large amounts of data. The proposed concept can effectively to reduce the amount of data input and to shorten the learning time. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|