Enhance the Internal Dynamics of Artificial NeuroMolecular System And Its Application to Category Type and Time Series Type Data

• Main Authors: Guo-Xun Liao, 廖國勛
• Other Authors: John-Chen Chan
• Format: Others
• Language: zh-TW
• Published: 2007
• Online Access: http://ndltd.ncl.edu.tw/handle/96214211886352359044

博士 === 國立雲林科技大學 === 管理研究所博士班 === 95 === The common artificial neural networks that using connection model emphasize the characteristic that neurons connects each other. But they overly neglect the intraneuronal information processing. Relatively, Artificial NeuroMolecular System (ANM system)(Chen, 1993) emphasizes the intraneuronal information processing. And it adopts the characteristic that neurons connects each other of traditional connection models, with the method of self-organizing learning, come to assemble neurons of unique intraneuronal dynamics into a collection capable of performing a required task. The objective of this study is to enhance the intraneuronal dynamics of ANM system. Use space characteristic of intraneuronal dynamics and time characteristic of transmission of signals apply to category type and time series type data. And the system was applied to three different problem domains, a study on the weaning results of ventilator-dependent patients in respiratory care center, a study to investigate the relationship between weight changes of premature babies and the total partenteral nutrition filld by physicians and a study of forecasting of the stock-market price fluctuation in Taiwan. The experimental results of the system were compared to those of the statistical tool, backpropagation neural networks, support vector machine and decision tree algorithm C4.5. Lastly, we investigated the degrees of influence of each parameter, and compare noise tolerance capability of those methods. Experimental results showed that ANM system has substantial data differentiation capability and noise tolerance capability in problem domain of predication of data classification. It also can handle hybrid type (numeric type and category type appears at the same time) data availably, and apply to problem domain of time information processing.