切比雪夫正交基神经网络的权值直接确定法

经典的BP神经网络学习算法是基于误差回传的思想.而对于特定的网络模型,采用伪逆思想可以直接确定权值进而避免以往的反复迭代修正的过程.根据多项式插值和逼近理论构造一个切比雪夫正交基神经网络,其模型采用三层结构并以一组切比雪夫正交多项式函数作为隐层神经元的激励函数.依据误差回传(BP)思想可以推导出该网络模型的权值修正迭代公式,利用该公式迭代训练可得到网络的最优权值.区别于这种经典的做法,针对切比雪夫正交基神经网络模型,提出了一种基于伪逆的权值直接确定法,从而避免了传统方法通过反复迭代才能得到网络权值的冗长训练过程.仿真结果表明该方法具有更快的计算速度和至少相同的工作精度,从而验证了其优越性.     

融合卷积神经网络与多层感知器的鞍部识别方法

针对传统鞍部识别方法中特征选择困难及未考虑鞍部与其它地形要素的共生关系等问题,利用深度卷积神经网络的特征自学习性能,提出了一种卷积神经网络与多层感知器相结合的混合模型实现DEM数据中的鞍部要素识别.首先设计改进的卷积神经网络模型自动提取鞍部的深度特征,经过Softmax分类器得到候选鞍部点,再运用多层感知器对候选鞍部点的位置进行精细回归,标识出最终的鞍部要素坐标.通过自建的鞍部样本集SADDLE-100训练网络模型,并在三种不同的山地样区进行实验,实验结果表明该方法比其它鞍部识别方法的漏提率减少约50%,正确识别率提高6.7%,在一定程度上避免了人工选择特征造成的鞍部语义信息缺失现象,为DEM中的点状要素识别提供了新的技术途径.     

基于优化卷积神经网络结构的人体行为识别

为了提高卷积神经网络对非线性特征以及复杂图像隐含的抽象特征提取能力,提出优化卷积神经网络结构的人体行为识别方法.通过优化卷积神经网络模型,构建嵌套Maxout多层感知器层的网络结构,增强卷积神经网络的卷积层对前景目标特征提取能力.通过嵌套Maxout多层感知器层网络结构可以线性地组合特征图并选择最有效特征信息,获取的特...     

多传感器系统估计的稳健切比雪夫中心估计融合

针对多传感器线性回归模型的参数估计融合问题,在观测噪声是范数有界的情况下,提出了稳健切比雪夫中心估计融合方法。描述参数的可行集合,为线性系统的所有可行解。可行集合的切比雪夫中心是最坏情况下使得估计误差最小的点,可用它作为多传感器系统参数估计的稳健融合。该问题在复数域上的某些情况可以精确求解,但目前的研究在实数域上只能得到近似解,即松弛的切比雪夫中心。严格证明了在实平面上可行集合的切比雪夫中心可以通过有限个约束的凸优化问题求解,因此,切比雪夫中心可以通过高效优化算法得到。在高维情况下,通过将可行集合投影到各坐标平面,设计了近似的切比雪夫中心融合方法。数值实验结果表明:该方法优于松弛的切比雪夫中心融合。     

通信系统中的信道均衡问题研究

针对通信系统中的信道均衡问题进行了研究,设计了一种基于MLP多层感知器的信道均衡系统。首先,对传统线性均衡器存在的问题进行了描述;其次,针对传统线性均衡器存在的问题构建了基于MLP多层感知器的信道均衡系统模型,即以MLP均衡器为核心,增加训练模块与判决模块;然后引入一种快速学习算法对模型进行训练,即通过学习率和动量项优化后的BP算法,减少了模型训练的时间;最后,对快速学习算法进行测试,通过星座图。均方误差折线图以及误码率折线图三种方式对传统线性均衡器与基于MLP多层感知器的信道均衡系统进行测试与对比。结果表明：在复杂环境下,传统线性均衡器对星座图的校正会出现模糊和偏移的现象,而基于MLP多层感知器的信道均衡器对星座图的校正更为精准有效;均方误差折线图显示,在整体上,MLP结构的均衡器比传统LSM均衡器降低了5 dB;误码率折线图显示,在信噪比达到16 dB之后,MLP结构的均衡器误码率更低。综上可知,本研究提出的MLP结构的均衡器性能更佳且稳定性更好。     

感知器在语言模型训练中的应用

感知器（perceptron）是神经网络模型中的一种,它可以通过监督学习（supervised learning）的方法建立模式识别的能力．将感知器应用到语言模型的训练中,实现了感知器的两种不同训练规则以及多种特征权值计算方法,讨论了不同的训练参数对训练效果的影响．在训练之前,使用了一种基于经验风险最小化（empirical risk minimization,ERM）的特征选择算法确定特征集合．感知器训练之后的语言模型在日文假名到汉字（kana-kanji）的转换中进行评估．通过实验对比了感知器的两种训练规则以及变形算法的性能,同时发现通过感知器训练的模型比传统模型（N-gram）在性能上有了很大的提高,使相对错误率下降了15％～20％．     

基于前馈多层感知器的网络入侵检测的多数据包分析

提出了一种新型网络入侵检测模型,在该模型中,首先将截获的数据包结合历史数据包数据库进行协议分析,找出可能存在的入侵行为的相关数据包,然后采用前馈多层感知器神经网络对这些相关的数据包进行回归分析,最终获得检测结果。该模型与传统采用单数据包检测方式的网络入侵检测系统(NIDS)模型相比,具有更低的漏检率。     

神经网络在信号除噪技术中的应用

介绍了在非线性滤波和自适应滤波技术中神经网络的应用,并以四层感知器为例对其噪声抑制的机理进行了定性分析。文中给出的一个多层感知器实现噪声相消的实例,充分地显示了神经网络技术在信号处理领域中的魅力。     

侧信道多层感知器攻击中基于贝叶斯优化的超参数寻优

传统侧信道攻击利用加密设备泄露的物理信息来获取密钥,由于其需要大量人为干预,越来越多研究将机器学习算法运用到侧信道攻击中,其中神经网络攻击效果最好,而多层感知器又是神经网络的基础,其中超参数在很大程度上影响最终训练与攻击结果.为实现超参数自动寻优,将贝叶斯寻优的方法应用在侧信道多层感知器攻击中,并提出对离散值的处理方法,发展出能够结合超参数经验的侧信道多层感知器超参数寻优方法.实验对比了人工寻优与贝叶斯寻优两种算法用于侧信道多层感知器攻击中的效率,验证了多层感知器与侧信道攻击相结合及贝叶斯寻优的可行性和高效性.     

提高滤波器带外抑制的新结构——类交指结构

简要介绍了滤波器的耦合方式和提高带外抑制的传统方法,通过改变普通切比雪夫滤波器的耦合结构,用不同于传统的方法同样达到了提高带外抑制的目的.用Ansoft HFSS软件设计8560MHz切比雪夫滤波器实例,通过比较采用传统法和新方法的两个模型的仿真曲线,验证了设计的可行性.     

Nonlinear Poisson regression using neural networks: a simulation study

We describe a novel extension of the Poisson regression model to be based on a multi-layer perceptron, a type of neural network. This relaxes the assumptions of the traditional Poisson regression model, while including it as a special case. In this paper, we describe neural network regression models with six different schemes and compare their performances in three simulated data sets, namely one linear and two nonlinear cases. From the simulation study it is found that the Poisson regression models work well when the linearity assumption is correct, but the neural network models can largely improve the prediction in nonlinear situations.     

Limitations of multi-layer perceptron networks - steps towards genetic neural networks

In this paper we investigate multi-layer perceptron networks in the task domain of Boolean functions. We demystify the multi-layer perceptron network by showing that it just divides the input space into regions constrained by hyperplanes. We use this information to construct minimal training sets. Despite using minimal training sets, the learning time of multi-layer perceptron networks with backpropagation scales exponentially for complex Boolean functions. But modular neural networks which consist of independentky trained subnetworks scale very well. We conjecture that the next generation of neural networks will be genetic neural networks which evolve their structure. We confirm Minsky and Papert: “The future of neural networks is tied not to the search for some single, universal scheme to solve all problems at once, bu to the evolution of a many-faceted technology of network design.”     

一种人工神经网模型及其在专家系统中的应用

本文提出了一种新的人工神经网模型,这一建立于感知机和线性机基础之上的多层网络模型,可用以处理模糊分类的问题,这个模型作为知识库,被应用于一个生产决策专家系统,在学习功能,运行速度,对新问题的适应性和对输入数据的容错性等方面,均优于原来的系统,取得了令人满意的效果。     

Predicting preferences of university students for prepaid vs post paid cell phone service plans

Information and communication technologies have recently become widely used, especially among the younger population. In this study, the factors affecting the preference of undergraduate students for prepaid or postpaid cell phone service plans were analyzed and a multi-layer perceptron type feed forward neural network model was developed to predict the preferences. Using the responses to the questionnaire administered to a group of undergraduate students in Istanbul University, the factors determining the preference for service plan were determined with χ2 test for independence. A classification model based on multi-layer perceptron type neural networks was developed. The classification accuracy of this model was compared to linear regression, LDA, QDA, Naive Bayes and decision tree approaches and shown to be superior.     

基于多层感知机模型的车牌识别技术

车牌识别是对实时图像中的车牌区域进行感知和截取,进行光学字符识别的过程.针对人工检测效率低的弊端,设计了一种应用于Android移动平台的车牌识别检测系统,利用OpenCV视觉开发库进行二次开发,使用支持向量机对图像进行判断,截取有效车牌区,并使用人工神经网络中的多层感知机模型实现字符的识别.测试结果表明,该系统检测速...     

Small hydro power plant identification using NNARX structure

A feedforward multi-layer perceptron neural network structure is developed to model the nonlinear dynamic relationship between input and output of a hydro power plant connected as single machine infinite bus system. Two independent second-order neural network nonlinear auto-regressive with exogenous signal models are used in the study. The structure selection of each independent model is based on various validation tests. The optimal brain surgeon pruning strategy adopted for optimizing the neural network structure. The network performance is studied for fixed and change in operating point.     

Adaptive air-fuel ratio control with MLP network

This paper presents an application of adaptive neural network model-based predictive control (MPC) to the air-fuel ratio of an engine simulation. A multi-layer perceptron (MLP) neural network is trained using two on-line training algorithms: a back propagation algorithm and a recursive least squares (RLS) algorithm. It is used to model parameter uncertainties in the nonlinear dynamics of internal combustion (IC) engines. Based on the adaptive model, an MPC strategy for controlling air-fuel ratio is realized, and its control performance compared with that of a traditional PI controller. A reduced Hessian method, a newly developed sequential quadratic programming (SQP) method for solving nonlinear programming (NLP) problems, is implemented to speed up nonlinear optimization in the MPC.     

Neural network modeling of time-dependent creep deformations in masonry structures

Stresses and deformations in concrete and masonry structures can be significantly altered by creep. Thus, neglecting creep could result in un-conservative design of new structures and/or underestimation of the level of its effect on stress redistribution in existing structures. Brickwork has substantial creep strain that is difficult to predict because of its dependence on many uncontrolled variables. Reliable and accurate prediction models for the long-term, time-dependent creep deformation of brickwork structures are needed. Artificial intelligence techniques are suitable for such applications. A model based on radial basis function neural networks (RBFNN) is proposed for predicting creep and is compared to a multi-layer perceptron neural network (MLPNN) model recently developed for the same purpose. Accurate prediction of creep was achieved due to the simple architecture and fast training procedure of RBFNN model especially when compared to MLPNN model. The RBFNN model shows good agreement with experimental creep data from brickwork assemblages collected over the last 15 years.