样本个数

本文研究了批方式和序贯方式训练的外监督前馈网络的全局最小条件.对于序贯训练方式，我们证明在其对应误差表面上总是存在N（训练样本个数）个局部最小点；

This paper studies the global minimum conditions of the outer supervised feedforward neural networks ( FNN ) in the batch style and sequence style trainings . For the sequence style training , it is showed that there always exist N local minimum points in the error surface .

阶递归最小二乘（LS）算法适用于观察样本个数固定且模型阶数未知时FIR或AR参数的最佳估计和识别问题中。

The recursive in order least-squares ( LS ) algorithms can be applied to estimation and identification of FIR or AR system parameters where a fixed set of input signal samples is available and the order of the model is un - known .

主要包括两方面的内容，即样本个数约简和样本属性约简。

Two aspects are included , namely , the sample number reduction and attribute reduction .

在传统的模式分析中，许多分类方法的计算复杂度随着训练集样本个数的增加而快速增长。

In traditional pattern analysis , the computational complexity of many classifiers increases quickly as the number of training samples increases .

对于基于支持向量机的模糊辨识方法，辨识的计算复杂度与训练样本个数呈指数关系。

For the fuzzy identification approaches based on SVM , the computational cost of identification assumes the exponential growth along with the number of training samples .

一个特例是当隐层神经元的个数与样本个数相等时，就可以求得代价函数值为0的全局最小点。

In particular , when the number of neurons in hidden layer equals the number of training patterns , the global minima with zero cost functions are obtained .

和岭估计以及偏最小二乘估计一样，广义岭估计主要适用处理病态设计矩阵，并可以有效的解决普通多元线性回归无法解决的问题，如样本个数小于自变量的个数。

Same as ridge regression and partial Least-squares regression , GRR mainly applies to multicollinearity and is effective settlement for sample measure less than the number of independent variable .

网络拓扑结构简单，网络训练所需的神经元个数与样本个数无关，可以简单地表示成输入、输出样本向量维数之积。

The number of neurons needed for the training is simply expressed as the multiplication of the input patterns ' vector dimension by the output patterns ' vector dimension .

并指出代价函数全局最小值随隐层神经元个数的增加而单调减小.当隐层神经元个数不小于样本个数时，网络的代价函数全局最小值将等于零。

The GMCF decreases monotonically as the number of hidden neurons increases and it would be zero if the number of hidden neurons is not less than the number of preassigned patterns .

人脸识别实验表明，所提出的方法能够提高近邻分类方法的效果，并解决学习矢量量化分类方法典型样本个数难以合理确定的问题。

Experiments on a face database show that the new classifier design method can improve the classification effect of the neighbour method , and can solve the difficult problem of ascertaining justly the typical samples in the classification of the learning vector quantization .

研究了基于子空间的信道盲辨识与盲均衡算法，并分析了统计样本个数、信噪比、载波频偏和相偏等因素对该算法的影响。

The channel blind identification and blind equalization algorithm based on subspace decomposition is investigated , and the factors that may affect the algorithm such as the number of statistic samples 、 SNR 、 carrier frequency offset and phase offset have been analyzed . 6 .

分别研究了输入变量组合、隶属函数个数、训练次数、训练样本个数等对神经模糊系统的学习能力、泛化能力的影响以及神经模糊系统的学习能力和泛化能力之间的关系。

The effect that the combination of input , the number of membership function of input , training times and training sample affect the learning ability and generalization ability of NFS was researched respectively . The relation between the learning ability and generalization ability was studied .

形状描述子表示为一个N维向量，N是样本点个数。

The shape descriptor is an N-dimensional vector , N is the number of sample points .

采用三层BP神经网络作为ECG身份识别分类器，以最优特征子集作为分类器的输入，输出层由实验的样本的个数决定。

A BP neural network with three layers is used as the ECG identification classifier . Take the optimal feature subset as the input of the classifier , the output layer is up to the number of samples .

在基因表达数据中，基因的数量相对于数据样本的个数通常比较多；

In gene expression data , the number of genes is usually very high compared to the number of data samples ;

在保证支撑矢量机的分类能力情况下，该方法大大地减少了训练样本的个数，同时提高了支撑矢量机的训练速度。

This method greatly reduces the number of training examples and so improves the training speed of SVM , without the loss in generalization performance .

分析比较了上述几种方法的预报精度误差、训练算法速度、时间等性能指标。最后对训练样本数据个数、神经元个数的选择进行了探讨和经验总结。

Analyze and compare the prediction precision and the time of training between those algorithms . Finally , discuss and research how to select the number of the neural units and the training data .

研究或论述的类型是证据等级分类的基本依据，研究的质量、设计、偏倚、样本、个数和结果是许多标准考虑的因素。

The types of study and discussion are the essential foundations of evidence grade classify , the quality , design , bias , sample , number , and result are the factors of many standard consideration .

此外FTLM可以通过设置样本点的个数，灵活地调节定位精度与计算复杂度以满足应用需求，该算法尤其适用于本文研究的路径规划问题。

Moreover , the precision and complexity can be adjusted by the set of swatch nodes to satisfy applications ' requirements . It is very suitable for the path planning research in this paper .

经过协整检验发现，15个省区样本协整关系个数存在很大不同，各个地区的协整系数更是存在较大差异。

With co integration tests , there is obvious difference in the relationship of the samples of 15 provinces , and the coefficients of co integration are quite different .

进一步，以样本容量、变量个数、滞后阶数为输入指标，检验临界值为输出指标建立了神经网络模型。

And then , we established the neural network model for cointegration test with small sample , where the sample size , number of variables , and lagged order were employed as inputs and critical values as outputs .

利用古典概型概率计算方法，证明概率空间（Ω，F，P）恰有n个相互独立的随机事件的充要条件是样本空间Ω的样本点个数可以分解成n个素数的乘积。

We have proved that probability space (Ω, F , P ) just has n independent random events only if the element number in sample space Ω may be resolved into the product of n prime numbers , by means of probability calculate method of classic probability model .

该方法还通过重复利用样本和减少后续迭代中新增样本个数，提高了响应面法的计算效率。

Cumulative use of experimental points and reducing the number of new experimental points in subsequent iteration also improve the efficiency of the weighted nonlinear RSM .

对给定的样本点，由样本点为变量的协方差函数构成的矩阵，当样本点个数趋于无穷大时，证明此矩阵谱逼近于积分方程正定核的谱逼近定理。

For the given sample points , and matrix formed by covariance function with sample points as parameters , when the number of sample points approaches infinite , it is proven that this matrix spectrum will approach the spectral approach theorem for positive-definite kernel of integral equation .

通过样本遗传策略，遗传落在下代信赖域空间上的样本，减少实验设计样本个数从而提高效率。

With the help of sample inheriting strategy , the method can inherit samples falling in the next trust region to reduce the number of experimental design samples , and thus the efficiency is increased .