An optimal neural network process model for plasma etching

Neural network models of semiconductor processes have recently been shown to offer advantages in both accuracy and predictive ability over traditional statistical methods. However, model development is complicated by the fact that back-propagation neural networks contain several adjustable parameters whose optimal values are initially unknown. These include learning rate, initial weight range, momentum, and training tolerance, as well as the network architecture. The effect of these factors on network performance is investigated here by means of a D-optimal experiment. The goal is to determine how the factors impact network performance and to derive a set of parameters which optimize performance based on several criteria. The network responses optimized are learning capability, predictive capability, and training time. Learning and prediction accuracy are quantified by the experimental error of the model. The process modeled is polysilicon etching in a CCl ₄-based plasma. Statistical analysis of the experimental results reveals that learning capability and convergence speed depend mostly on the learning parameters, whereas prediction is controlled primarily by the number of hidden layer neurons. An optimal network structure and parameter set has been determined which minimizes learning error, prediction error, and training time individually as well as collectively     

Adaptive neural nets for generation of artificial earthquakeprecursors

A novel methodology for generation of artificial earthquake precursors was tested on Southern California earthquake data in reverse and real time modes. When it was tried as a real time generator of earthquake precursors, it successfully predicted the June, 1992, Landers earthquake. The methodology is based on the use of adaptive neural nets (ANN) that process a set of time-dependent attributes calculated in a moving time-window. The most important of them is a danger function. The structure of the neural net is defined by the properties of input data in the moving time window. Thus, the neural net continuously adapts its structure to the time variant properties of the input attributes. The main problem the authors encountered in training the neural net on the earthquake data was the small size of the training set compared to the number of parameters that describe the structure of the ANN. To prevent instability and over-fitting in the training session, the authors used a technique similar to the damping method in least squares approximation     

基于KPCA-GA-BP神经网络的POI质量预测研究

目前移动网络优化一般基于小区进行网络质量评估及预测，遵循“升维研究，降维实施”的研究思路，提出了兴趣点（point of interest,POI）网络质量的柔性评价体系，但其涉及较多网络关键绩效指标（key performanceindicator,KPI），导致POI网络综合质量评价体系较为庞杂且预测精度不高，为提高POI网络质量预测精准性，采用核主成分分析（kernelprincipalcomponentanalysis,KPCA）算法对反向传播（back propagation,BP）神经网络的输入变量进行相关性压缩，简化了BP神经网络结构，然后通过遗传算法（genetic algorithm,GA）优化了BP神经网络连接权值及阈值参数。与传统BP神经网络预测结果进行对比，在预测准确度方面提高了10.90%，均方误差性能显著降低，对研究POI网络质量的预测可起到较好的支撑作用。     

基于PCA神经网络的葡萄酒评价模型研究

利用主成分分析与RBF神经网络相结合,建立葡萄酒质量评价预报模型,并进行训练和仿真验证。该模型运用SPSS软件对葡萄酒中影响风味指标进行主成分分析,将多变量、非线性的原始数据进行降维,保留原始信息的主要信息,把原来若干个属性变量综合成几个不相关主成分分量;再以计算结果作为RBF网络的输入数据,葡萄酒的感官评价得分作为网络的输出数据,建立葡萄酒主要理化指标与葡萄酒质量的关系模型。结果表明：该评价模型的建立,缩短了葡萄酒评价的周期,克服了品酒师聚集的困难;与传统RBF网络相比,大大简化了网络结构,提高了网络的训练速度和预报精度,为质量评价问题提供了一种的研究思路。     

A signal processing application in genomic research: protein secondary structure prediction

The digital nature of genomic information makes it suitable for the application of signal processing techniques to better analyze and understand the characteristics of DNA, proteins, and their interaction. Prediction of genes, protein structure, and protein function greatly utilize pattern recognition techniques, in which hidden Markov models, neural networks, and support vector machines play a central role. Signal processing offers a variety of methods from pattern recognition and network analysis for the diagnosis and therapy of genetic diseases. In this paper, we focus on protein secondary structure prediction and discuss the problems in single sequence setting.     

采用序贯学习的改进型Elman神经网络预测研究

针对石油化工领域监测系统中,广泛存在的非线性时间序列预测问题。将Elmann神经网络方法引入化工过程监测的在线预测。并在其理论框架基础之上,改进了Elman神经网络的内部结构,引入了串行学习机制,可以根据实时数据对网络进行在线训练,提高网络预测精度。通过对某芳烃厂实时数据在线预测仿真,表明该方法能够准确地在线预测未来数据,同时具有训练速度快、结构简单、适应性强的优点。     

对角型广义RBF神经网络与非线性时间序列预测

径向基函数（RBF）神经网络在非线性时间序列预测方面发挥着重要作用。本文提出了对角型广义RBF神经网络模型,并利用贝叶斯阴阳（BYY)谐和学习算法进行隐层单元个数的选择和参数初始值的设置,且建立了同步LMS算法进行参数学习。进一步,将对角型广义RBF神经网络应用于非线性时间序列预测,得到了预测准确率高和速度快的效果。      

基于AdaBelief残差神经网络的超材料结构逆设计

基于深度学习的超材料器件设计得到了前所未有的发展,但在基于2维材料的反设计中,传统的人工神经网络难以解决在小采样空间内陷入局部最优值的问题,且随着结构的复杂性增加,需要耗费大量的计算成本.针对这些缺陷提出了一种基于AdaBelief优化算法的残差神经网络,选择基于石墨烯的多层交替薄膜结构的设计来验证该网络的有效性,采用...     

INVESTIGATION OF PROPERTIES OF INFRARED ABSORPTION OF α-HELIX PROTEIN MOLECULES

The mechanism and properties of infrared absorption of α-helix protein molecules are studied by a theory of bio-energy transport established on the basis of molecular structure. From the vibrational energy-spectra of molecules obtained from this theory we know that the infrared lights with wavelengths of 2 µm ?7 µm can be absorbed by α-helix protein molecules. This is basically consistent with experimental data of infrared absorption of collagen and hemoglobin and bivine serum albumen (BSA) proteins with α-helix structure. From these results we account further for the mechanism and properties of biological effect of infrared lights absorbed by the living systems, i.e., the energy of infrared lights is directly absorbed by the amide-Is in amino acid residues in the protein molecules, which results in vibration of amide-1 and change of conformation of proteins and transport of bio-energy from one place to other along the protein molecular chains in human beings and animals. This is a kind of non-thermally biological effect.     

Aero-optic imaging deviation prediction based on ISSA-ELM

When the aircraft is moving at high speed in the atmosphere, aero-optical imaging deviation will appear due to the influence of aero-optical effect. In order to achieve real-time compensation during the flight of the aircraft, it is necessary to analyze and predict the obtained imaging deviation data. In order to improve the search speed and accuracy of the prediction algorithm and the ability to jump out of local optimum, in this paper, an improved sparrow search algorithm optimized extreme learning machine (ISSA-ELM) neural network model is proposed to predict the aero-optical imagine deviation. Finally, the performance of ISSA-ELM, ELM neural network and SSA-ELM neural network was tested. The results showed that compared with ELM and SSA-ELM algorithms, the convergence speed of ISSA-ELM was significantly enhanced, and the accuracy of data prediction was also significantly improved.     

On the predictability of next generation mobile network traffic using artificial neural networks

Though the introduction of the new 4th Generation mobile access technologies promises to satisfy the increasing bandwidth demand of the end‐users, it poses in parallel the need for novel resource management approaches at the side of the base station. To this end, schemes that try to predict the forthcoming bandwidth demand using supervised learning methods have been proposed in the literature. However, there are still open issues concerning the training phase of such methods. In the current work, the authors propose a novel scheme that dynamically selects a proper training set for artificial neural network prediction models, based on the statistical characteristics of the collected data. It is demonstrated that an initial statistical processing of the collected data and the subsequent selection of the training set can efficiently improve the performance of the prediction model. Finally, the proposed scheme is validated using network traffic collected by real, fully operational base stations. Copyright © 2013 John Wiley & Sons, Ltd.     

利用无监督预训练的轨迹深度关联

针对时空相似度算法关联轨迹的局限性,采用深度学习方法进行轨迹关联,并提出了一种基于无监督预训练的匹配神经网络训练方式。利用Geohash向量嵌入对轨迹信号做特征工程处理,构建自注意力机制神经网络结构,使用无标注轨迹数据基于遮蔽预测任务进行模型预训练;然后构建孪生匹配网络结构,加载预训练模型参数;最后使用标注轨迹对数据基于均方差损失函数微调预训练模型参数得到轨迹对匹配模型。采用Geolife GPS轨迹数据集作为评估数据集进行模型训练与测试,实验结果显示,利用无监督预训练的轨迹关联方法较现有最优算法匹配准确率提高了5个百分点,达到了96.3%,充分证明了该方法的有效性。目前轨迹关联领域基于深度学习预训练模型的研究较少,该方法具有重要的参考意义。     

Predicting Protein Function by Frequent Functional Association Pattern Mining in Protein Interaction Networks

Predicting protein function from protein interaction networks has been challenging because of the complexity of functional relationships among proteins. Most previous function prediction methods depend on the neighborhood of or the connected paths to known proteins. However, their accuracy has been limited due to the functional inconsistency of interacting proteins. In this paper, we propose a novel approach for function prediction by identifying frequent patterns of functional associations in a protein interaction network. A set of functions that a protein performs is assigned into the corresponding node as a label. A functional association pattern is then represented as a labeled subgraph. Our frequent labeled subgraph mining algorithm efficiently searches the functional association patterns that occur frequently in the network. It iteratively increases the size of frequent patterns by one node at a time by selective joining, and simplifies the network by a priori pruning. Using the yeast protein interaction network, our algorithm found more than 1400 frequent functional association patterns. The function prediction is performed by matching the subgraph, including the unknown protein, with the frequent patterns analogous to it. By leave-one-out cross validation, we show that our approach has better performance than previous link-based methods in terms of prediction accuracy. The frequent functional association patterns generated in this study might become the foundations of advanced analysis for functional behaviors of proteins in a system level.      

广义RBF神经网络在煤矿冲击地压预测上的应用

本文将广义径向基函数（RBF）神经网络应用于华丰煤矿实测的煤矿中冲击地压数据的建模和短期预报。在网络设计上,本文采用了贝叶斯阴阳（BYY)和谐学习算法进行网络隐单元个数的确定和参数初始值的选取,而在参数学习上,本文则采用了同步LMS学习算法。实验结果表明,这种基于广义RBF神经网络的预测方法在精度和速度上有了显著的优势,能够满足在工程应用中的实际要求。      

基于BP＿Adaboost模型的乳腺癌诊断预测方法研究

乳腺癌是全球女性发病率位居首位的恶性肿瘤,研究基于神经网络模型的乳腺癌诊断预测方法的目的是将临床与机器学习相结合,有助于医疗工作者更加快速准确地判断出患病与否,同时解决现有模型中存在的过拟合以及漏诊率和误诊率过高的问题,并提高预测模型的准确率。本文采用加州大学欧文分校(UCI)数据集,共669个样本,其中包含357个良性样本和212个恶性肿瘤样本,共计10个特征训练预测模型。将10个神经网络模型采用Adaboost方法相结合,即通过Adaboost算法组合多个弱分类器从而形成一个强分类器,最终输出一个具有更高准确率、有较强的自学习能力、自适应能力且泛化性能优良的集成预测模型。结论表明,该模型的预测准确率达到98.5507%,同时准确率(AUC)为0.9966,说明所建模型区分度较好,可以反映模型的诊断价值,且非常稳定,具有非常好的验证效果,为临床应用提供进一步的技术支持和保障。     

An improved deadbeat rectifier regulator using a neural netpredictor

This paper proposes a new input current reference prediction scheme for the deadbeat control of a three-phase rectifier used in AC/DC/AC converters. The inherent lag in deadbeat control is compensated by predicting the reference resulting in better performance. The proposed predictor consists of a neural net which is trained on-line and predicts the slow varying and periodic trends of the current reference plus a linear first order predictor which predicts the fast variations of the current reference time signal. A CRITIC decides if the neural net training is sufficient and therefore whether or not to use the prediction in the control loop. The learning rule used allows neural net weights to be trained whenever a parameter change causes an increased prediction error. This predictive-regulator is shown to result in improved performance in steady state, in the presence of input voltage imbalance or load variations     

一种优化的神经网络树异常入侵检测方法

本文提出了一种基于优化神经网络树(ONNT)的异常检测方法,在提高异常检测精确率的同时,增强异常检测模型学习结果的可理解性、可解释性。ONNT是一种具有二叉树结构的混合学习模型,二叉树的节点分裂遵循信息增益率准则;其中间节点嵌入了结构简单的感知器神经网络,能够根据当前节点上给定的子样本集和教师信号,选择较小的特征子集构建相对简单的局部决策曲面。本文提出的异常检测方法包括两个方面的性能优化:1）通过优化神经网络树(NNT)的中间节点,降低局部决策曲面的复杂度,从而使中间节点能在可接受的计算代价内表示成低复杂度的布尔函数或规则集,为实现学习结果的可解释性提供基础;2）通过优化学习模型的整体结构,降低所有中间节点的规则析取式的前件复杂度,从而提高学习结果的可理解性。实验的数值结果表明,与基于NNT的异常检测方法相比,本文提出的方法能够以简单的中间节点和相对精简的整体结构提高检测结果的可解释性和可理解性;与其他同类方法相比,基于ONNT的异常检测方法具有较高的检测精确率,且在一定程度上给出了对异常检测具有重大影响的一些特征信息。      

基于随机森林回归分析的PM2.5浓度预测模型

针对神经网络算法在当前PM2.5浓度预测领域存在的易过拟合、网络结构复杂、学习效率低等问题,引入RFR(random forest regression,随机森林回归)算法,分析气象条件、大气污染物浓度和季节所包含的22项特征因素,通过调整参数的最优组合,设计出一种新的PM2.5浓度预测模型—RFRP模型.同时,收集了西安市2013-2016年的历史气象数据,进行模型的有效性实验分析.实验结果表明,RFRP模型不仅能有效预测PM2.5浓度,还能在不影响预测精度的同时,较好地提升模型的运行效率,其平均运行时间为0.281 s,约为BP-NN(back propagation neural network,BP神经网络)预测模型的5.88％.     

基于CNN特征的宽频段智能测向方法

当前电磁环境日益复杂,利用机器学习方法实现快速且精确的宽频段无线电测向逐渐成为研究的热点。使用卷积神经网络基于端到端的方式完成宽频段测向的方法能够在一定程度上解决宽频段相位模糊的问题,但卷积运算后特征维数大大增加,稀疏的特征影响了最后一层全连接前馈神经网络的分类效果。针对这一问题,提出将无线电测向分为特征学习任务和方向预测任务,使用卷积神经网络作为特征提取器,将通过多层卷积运算得到的结果视为二次提取的特征,作为方向预测任务的输入;针对二次提取特征的稀疏性,提出使用主成分分析算法对特征进行降维,并将稀疏性降低后的特征作为后续分类器的输入。此外,针对特征的特点,探索了几种分类模型作为分类器的效果,包括决策树、随机森林、径向基函数神经网络和K-近邻。实验结果表明,使用主成分分析算法对特征进行降维能够提升训练和测试效率;采用K-近邻构成分类器的准确度明显高于原卷积神经网络的准确度;若需要兼顾准确度和测向效率,采用随机森林构成分类器的效果最好。