| 前馈神经网络隐性单调与泛化能力 |
| |
| 引用本文： | 何继善,朱德兵.前馈神经网络隐性单调与泛化能力[J].物探化探计算技术,2001,23(4):289-294,298. |
| |
| 作者姓名： | 何继善 朱德兵 |
| |
| 作者单位： | 何继善（中南大学, 长沙 410083）
朱德兵（中南大学, 长沙 410083） |
| |
| 摘 要： | 应用前馈人工神经网络对广域单调的两组样本进行了模拟反演，引入单调前馈网络的概念对其权值和阈值定解问题和泛化能力进行了较诉研究。表明前馈人工神经网络是一个表达形式简单的复杂系统，其单调特征是隐性的，而且训练网络的成熟性对样本数量和样本内在规律性有一定依赖。强调了前馈人工神经网络的应用效果，指出单调与复合问题还需进一步深入研究。
|
| 关 键 词： | 前馈神经网络 泛化问题 单调性 隐性特征 人工神经网络 网络结构 响应函数 |
| 文章编号： | 1001-1749(2001)04-0289-06 |
THE IMPLIED MONOTONICITY AND EVOLVEMENT CAPABILITY OF THE FEED FORWARD NEURAL NETWORKS |
| |
| HE Ji-shan,ZHU De bing.THE IMPLIED MONOTONICITY AND EVOLVEMENT CAPABILITY OF THE FEED FORWARD NEURAL NETWORKS[J].Computing Techniques For Geophysical and Geochemical Exploration,2001,23(4):289-294,298. |
| |
| Authors: | HE Ji-shan ZHU De bing |
| |
| Abstract: | Analogue inversions of two groups of models with generalized monotonicity were made by the feed forward artificial neural network. Leading with the concept of monotone feed forward neural network, some researches were done for the problems such as real values of the weights and thresholds, relationships between the model values and the corresponding weights, and the evolvement of the specified neural networks as well. It shows that the feed forward neural network is a somewhat complex system with simple construction, and in common fields its monotonicity is concealed in the network. Moreover, the characters of the mature network after trained relies, to some extend, on the number and inner rule of the training model data. The efficiency of the feed forward neural network is emphasized and that the importance of deep going research in the field is also pointed out in this paper. |
| |
| Keywords: | feed forward neural networks evolvement problem monotonicity hidden character |
| 本文献已被 CNKI 维普 万方数据 等数据库收录！ |
