傅立叶网络分析及降低其运算量的一种方法

本文对傅立叶神经网络进行了系统的分析，提出了一种降低其运算量的方法，这种方法对于程序设计和硬件实现都具有一定的参考价值．     

Fuzzy Entropy Based Combined Learning Algorithm for Neural Networks

ＦｕｚｚｙＥｎｔｒｏｐｙＢａｓｅｄＣｏｍｂｉｎｅｄＬｅａｒｎｉｎｇＡｌｇｏｒｉｔｈｍｆｏｒＮｅｕｒａｌＮｅｔｗｏｒｋｓ￥ＭｉｎＹａｏ（Ｄｅｐｔ．ｏｆＣｏｍｐｕｔｅｒＳｃｉｅｎｃｅ，ＨａｎｇｚｈｏｕＵｎｉｖｅｒｓｉｔｙ，Ｈａｎｇｚｈｏｕ３１００２８，Ｐ...     

电流模式MOS A／D转换神经网络的研究

从Ａ／Ｄ转换基本规则出发，提出了一种电流模式ＭＯＳＡ／Ｄ转换神经网络，并对该网络进行了分析研究，导出确定权值的公式。网络权值易存贮且调节简单。用ＰＳＰＩＣＥ对４－ｂｉｔＡ／Ｄ转换网络进行仿真，并对采用分立元件的构成的３－ｂｉｔＡ／’Ｄ转换网络进行了实验。     

基于神经网络的矿井井型确定方法

研究了基于神经网络的矿井井型确定方法，为矿井设计提供了一种崭新的工具，应用表明，该方法是切实可行的。图２，表２，参４。     

基于神经元网络的不确定性规则的获取

讨论了不确定性规则的知识系统同神经元网络相结合的问题;提出了适合处理不确定性规则的负梯度学习算法,使不确定性规则能方便地转换成神经元网络的权矩阵;研究了如何利用神经元网络的生成能力来获取新的不确定性规则,使知识系统具有知识自动获取能力,并使原有规则系统的脆弱性得以改善.     

模糊建模与控制的神经网络方法及其仿真研究

本文提出了一种复杂系统模糊建模及控制的神经网络方法。利用改进的ｐｉ－ｓｉｇｍａ 网络，对模糊规则的结论参数和隶属函数进行在线修正，实现模糊规则的自组织。这种方法被用于降水量预报和机器人解耦控制，取得了满意的仿真结果。为增强神经网络仿真算法的快速性，本文采用了一种基于向量的数据结构，并用标识阵指示神经元的连接状态，以实现有效的内存运算。     

反向误差传播的学习策略

本文探讨多层神经网络、反向误差传播学习过程收敛缓慢的原因,提出了修改后的激励函数(x),变步长策略,输出反馈抑制策略及模式矢量预处理,从而大大加速了学习收敛过程。还给出了计算机模拟的结果。     

应用神经网络理论的计算机通信网迟延新算法

计算机网络最小平均迟延的求解实质上是一个联合优化问题。进一步的分析表明，它可以近似为一个二次型优化问题。我们利用一种具有全局收敛性质的神经网络模型解决了这一问题。计算结果表明，这种方法是可行而有效的。     

A multichannel neural signal detecting module:Its design and test in animal experiments

A four-channel neural signal detecting module with an implantable 12-contact cuff electrode was designed for real-time neural signal recording on peripheral and central nerves. The mathematic coupling model between nerve and electronic system was analyzed. Electrode connection configurations were considered. The detecting circuit included an input coupling network, a pre-amplifier, and some filtering and notching stages. Shield guarding and the right-leg-driven circuit were developed for further elimination of common mode interference. By electrode switches, the module could cooperate with a nerve functional electrical stimulation circuit, building a neural channel bridge-connection system. It was tested by recording experiments on rat’s sciatic and spine nerves. The signals in spontaneous and evoked conditions have been captured successfully. In addition, an implantable neural signal detecting CMOS IC has been introduced.     

Analytical determination of flow resistance characteristic for combination channel inside hydraulic manifold block

Computational fluid dynamics （CFD） was used in conjunction with BP neural network to study the flow resistance characteristic of the combination-channel inside hydraulic manifold block （HMB）. The input parameters of the combination-channel model were confirmed to have effect on the pressure-drop by the numerical method, and a BP neural network model was accordingly constructed to predict the channel pressure-drops. The flow resistance characteristic curves of various channels were achieved, and a performance parameter was given to evaluate the through-flow characteristic of the channel according to the curves. The predictions are in agreement with the numerical computation, indicating that the method can be utilized to accurately determine the flow characteristic of the combination channel with high efficiency.