网络函数

In addition, a training algorithm based on a composite object function is proposed. This algorithm simplified the form of the composite object function and use adaptive method to adjust parameter of the composite object function so that the training speed is increased in the later period.

此外，还提出了一种基于综合目标函数的神经网络训练算法，在该算法中对综合目标函数进行了简化，并采用自适应的方法来调整目标函数的参数，使训练后期的收敛速度明显提高。

The experimental results show that temperature error and non-linearity of silicon piezoresistive pressure sensor can be reduced markedly.

本文针对其缺点，提出了以正交的局部函数——小波函数作为激励函数的小波神经网络对硅压阻式压力传感器的温度误差进行补偿。

In this paper,we propose a novel model named distribution-free data density estimation,which is based on distribution-free(i.e.,independent of data distributions) sampling on global cumulative distribution to achieve high estimation accuracy with low estimation cost regardless of distribution models of the underlying data.

分布无关密度估计算法首先将底层数据的任意分布转换成一中间分布——累计概率分布函数。由于累计概率分布函数的输出在[0，1]之间均匀分布，因此接着对累计概率分布函数的输出随机采样，可以准确估计当前网络中数据的密度分布。

By iteratively combining these two procedures we achieve a way of training the configuration of RBFNN. In addition, the algorithm is very robust with respect to the noise level. Simulation experiments show that the proposed algorithms are sound. In chapter 3, which is the extension of chapter 2, the improvements of the architectures of RBFNNs and the solutions of their relating problems are discussed, Firstly, an new distance metric is advanced and a forward orthogonal least square selection procedure is applied to learning the parameters of classification function and selecting the important input nodes.

首先分析了目前存在的同时确定RBFNN结构和参数的方法的缺陷，在此基础上提出了用基于拉马克的进化学说的进化编程算法来改进算法，以克服某些缺陷；然后针对受到严重噪声污染的系统，如何提高RBFNN的泛化能力的问题，利用基于AIC的适应度函数的改进遗传算法学习结构和参数；最后介绍基于MDL原理的方法，将优化网络的结构和参数分为两个阶段：训练和进化，先自适应地改变RBFNN基函数的中心和宽度，同时训练输出线性权值，再用基于MDL原理的适应度函数的标准GA来优化隐层节点，通过交替使用这两过程达到训练RBFNN的结构和参数的目的，该算法具有较强的鲁棒性。

Suppose that the local source of traffic demand of the intersection is a stochastic variable of known parameter. An optimal control assignment model of the link of the dynamical transportation network flows of stochastic system is founded. Allowing for the construct of the model, based on the objective function of the link is transformed to based on the objective function of the intersection in order to solve it.

假设交叉点的局外入流是某已知参数的随机变量的情况下，建立了目标函数基于路段的随机系统最优的动态交通网络流分配的最优控制模型，充分分析该模型的结构，将该模型的目标函数转化为基于交叉点的目标函数，以便于求解该问题。

By analyzing some receivers' behaviors of internet advertising, the paper constructs a behavioral structure model based on the flow structure. Receivers' online flow experience is a function determined by receivers' skill, attentiveness and cyber environmental challenge, restricted to interactive speed and perception of media environment. It studies the relationship among motivation of using internet, online durational time, interactivity, attitudes and purchase intentions. The paper also examines the actual effects of internet advertising.

杂谈 摘要：构建一个网络广告受众行为结构模式，把受众在网上流的经历转化成由受众技能的高低、集中注意的程度和网络环境提出挑战的大小三个变量来直接决定的函数，并且受交互速度的快慢和对媒体环境的感知两个条件的约束，通过一个网络广告受众行为过程模型研究受众接触广告的动机、网上持续时间、交互性、态度和购买动机之间的关系，然后对网络广告效果测评作出简要评价。

By analyzing some receivers' behaviors of internet advertising, the paper constructs a behavioral structure model based on the flow structure. Receivers' online flow experience is a function determined by receivers' skill, attentiveness and cyber environmental challenge, restricted to interactive speed and perception of media environment. It studies the relationship among motivation of using internet, online durational time,论文代写英语新课标再认识感悟新理念, interactivity, attitudes and purchase intentions.

构建一个网络广告受众行为结构模式，把受众在网上流的经历转化成由受众技能的高低、集中注意的程度和网络环境提出挑战的大小三个变量来直接决定的函数，并且受交互速度的快慢和对媒体环境的感知两个条件的约束，职称论文发表，通过一个网络广告受众行为过程模型研究受众接触广告的动机、网上持续时间、交互性、态度和购买动机之间的关系，然后对网络广告效果测评作出简要评价。

The process of establishing neural network is as follows: two-layer neuronic network configuration is selected, S-type active function is used, input parameter of network are selected for thickness, preheat temperature, heat input and sampling point, output parameter of network is selected for fracture toughness property CTOD, learning algorithm of network is used for back-propagation algorithm.

本文选择了2层神经元的神经网络结构，采用S型激活函数，将板厚、预热温度、线能量、取样位置作为神经网络的输入，将环焊接头的裂纹尖端张开位移CTOD值作为网络的输出，通过BP算法的网络学习，采用VC语言建立了X70管线钢环焊接头CTOD与环焊工艺之间的神经网络。

Uncertain upper boundary function is learned on-line by using FNN, and proportion controller strengthens the completeness of FNN control strategy. FNN is trained by using disparity target learning error produced by FIE. This can avoid supersaturating problem made by feedback error directly, and restrain influence of measuring noise and improve control performance.

由于模糊技术与神经网络的结合能弥补彼此的不足，因此本文提出了一种模糊神经网络控制器与比例控制器相结合的控制方案，采用模糊神经网络在线学习不确定函数的上界，比例控制器增强模糊神经网络控制策略的完备性。

In Chapter 2, the learning algorithm of the single-layer perceptron and its limitation are considered firstly.

前馈结构的神经网络是最重要的神经网络模型之一，本文系统地研究了前馈结构神经网络的能力，包括多层感知机和径向基函数网络两部份。

The split between the two groups can hardly be papered over.

这两个团体间的分歧难以掩饰。

This approach not only encourages a greater number of responses, but minimizes the likelihood of stale groupthink.

这种做法不仅鼓励了更多的反应，而且减少跟风的可能性。