网络函数

The model adopts cumulant generating function to analyze path-level packet transmission probability from end-to-end measurements and estimate link-level packet transmission probability.

该模型基于端到端的测量方式，采用累积母函数对网络路径级数据包通过率进行分析，估计出网络链路级数据包通过率的统计信息。

In allusion to the defect of grads descension of traditional back propagation network algorithms, some improving measures such as adaptive learning and additive momentum has been taken, and better application result is acquired.

针对基于梯度下降法的BP网络存在的学习效率低、收敛速度慢、易陷于局部极小状态、网络的泛化及适应能力较差等缺陷，采用了附加动量法及自适应学习率、合理选择激励函数等策略对算法中的缺陷给出了改进措施，取得了较好的应用效果。

We use wavelet neural network to construct the nonlinear cointegration function, and give the DFP algorithm for network training.

提出了利用小波神经网络逼近非线性协整函数的方法，并给出了训练小波神经网络的变尺度算法。

For a class of networked control systems with equality constraint ,this paper addresses the problem of designing controller,sampling period and the maximum allowable delay bound.

针对一类具有等式约束的网络控制系统控制器、采样周期以及最大允许延迟时间的设计问题，采用李亚普诺夫函数、线性矩阵不等式的方法，推导出了该网络控制系统渐近稳定的充分条件。

On the basis of energy function, the stability and convergence of the proposed neural network are analyzed, which guarantee the global stability of the equilibrium point of neural network.

此外，利用能量函数对神经网络的稳定性和收敛性进行了分析，进而保证所提出的神经网络具有全局稳定性。

In theory, any continuous function of n variables can be arbitrarily wellapproximated by the BP Neural Networks.

BP网络从理论上能够逼近任意多变量连续函数，BP网络结构简单，使用方便，是应用最为广泛的一种人工神经网络。

It defines the vector norms as satisfaction function to revise the weighting factors of metrics given by analytic hierarchy process. Meanwhile, the scheme exploits vector geometrical relationships in Hilbert sub-space basing on revised weights to measure similarities between vectors of network ability and user target.

该算法定义Hilbert空间中的向量范数表征满意函数，以修正由层次分析法给出的各指标权重因子；并利用基于修正权重因子的Hilbert子空间中向量间的几何关系，衡量网络参考向量与用户目标向量间的相似性，选择与用户需求最匹配的网络。

In this algorithm,the fuzzy hypersphere membership functions of the state estimation of local filters are combined with the weights of the fusion network,and the weights of the network can be adjusted online,so that the whole estimation is adaptive and fault-tolerant.

该算法将局部滤波器状态估计的模糊超球隶属函数引入网络，在线调整融合网络权值，实现全局估计的自适应性和容错性。

Firstly, two covering hypersphere algorithms were proposed, referred to as minimum volume covering hypersphere algorithm and moving medium centers covering hypersphere algorithm, to heuristically and rapidly select the promising hidden center vectors from all the training samples. Then the particle swarm optimization algorithm combined with recursive orthogonal least square algorithm was used to further prune the initial hidden centers and optimize the control parameters.

为在实际应用中加速优化速度，降低优化计算复杂度，首先提出了最小体积覆盖超球和移动平均中心覆盖超球两种算法，用于启发式的从训练样本中快速的初选出径向基概率神经网络的隐中心矢量，然后将粒子群优化算法和 ROLSA 结合进一步优选隐中心矢量，同时优化核函数控制参数，实现径向基概率神经网络的全结构优化。

By redefining multiplier associated with inequality constraint as a positive definite function of the originally-defined multiplier, it is no longer necessary to convert inequality constraints into equality constraints by slack variables in order to reuse the method dedicated to equality constraints for constructing Lagrange neural networks.

若重新定义与不等式约束相关的乘子为正定函数，则在构造Lagrange神经网络时，可直接使用处理等式约束的方法处理不等式约束，不需再用松驰变量将不等式约束转换为等式约束，减小了网络实现的复杂程度。

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。