改进的遗传谐振匹配网络学习算法

针对简单自适应匹配（SFAM）网络结构冗余的问题提出一种改进遗传谐振匹配网络学习算法.通过引入群体适应度均值和标准差来自适应调整交叉概率和变异概率;针对自适应谐振网络自身在训练过程中可能出现的类冗余问题,在遗传操作中引入裁剪算子,通过定义置信度因子依据规则对训练过程中网络可能出现的冗余类标志节点进行删除,降低网络结构的...     

线性模型选阶的一个新准则

本文给出了新的线性模型选阶准则Λ。由Λ推广得到的Λα。在适当选取α之后，能够浙近正确地选出真模型的阶数，文中讨论了ＡＩＣＣ和Λα的选阶比较问题。     

Machine learning in data envelopment analysis: A smart mechanism for indicator selection

Indicator selection has been a compelling problem in data envelopment analysis. With the advent of the big data era, scholars are faced with more complex indicator selection situations. The boom in machine learning presents an opportunity to address this problem. However, poor quality indicators may be selected if inappropriate methods are used in overfitting or underfitting scenarios. To date, some scholars have pioneered the use of the least absolute shrinkage and selection operator to select indicators in overfitting scenarios, but researchers have not proposed classifying the big data scenarios encountered by DEA into overfitting and underfitting scenarios, nor have they attempted to develop a complete indicator selection system for both scenarios. To fill these research gaps, this study employs machine learning methods and proposes a mean score approach based on them. Our Monte Carlo simulations show that the least absolute shrinkage and selection operator dominates in overfitting scenarios but fails to select good indicators in underfitting scenarios, while the ensemble methods are superior in underfitting scenarios, and the proposed mean approach performs well in both scenarios. Based on the strengths and limitations of the different methods, a smart indicator selection mechanism is proposed to facilitate the selection of DEA indicators.     

Can out‐of‐sample forecast comparisons help prevent overfitting?

This paper shows that out‐of‐sample forecast comparisons can help prevent data mining‐induced overfitting. The basic results are drawn from simulations of a simple Monte Carlo design and a real data‐based design similar to those used in some previous studies. In each simulation, a general‐to‐specific procedure is used to arrive at a model. If the selected specification includes any of the candidate explanatory variables, forecasts from the model are compared to forecasts from a benchmark model that is nested within the selected model. In particular, the competing forecasts are tested for equal MSE and encompassing. The simulations indicate most of the post‐sample tests are roughly correctly sized. Moreover, the tests have relatively good power, although some are consistently more powerful than others. The paper concludes with an application, modelling quarterly US inflation. Copyright © 2004 John Wiley & Sons, Ltd.     

基于过拟合神经网络的混沌伪随机序列(英文)

结合神经网络和混沌映射的特点,提出了一种基于过拟合BP神经网络的混沌伪随机序列产生方法.以logistic映射和Henon映射为例,测试结果表明,该方法能克服有限精度效应对混沌系统的影响,改善混沌伪随机序列的性能,为获得高性能的伪随机序列提供了一种新的思路.     

基于分布式概念格的分类规则挖掘

以概念格为分类模型,引入知识合并思想,并针对大规模数据的分类求解以及过拟合问题引入剪枝策略,从而得到分类剪枝概念格模型,在此基础上提出了基于分布式概念格模型的强分类规则提取算法;通过理论证明了算法的正确性,并通过实验证明了算法的可行性。     

支持向量机算法在若干熔盐相图中间相预报中的应用

用支持向量机(SVM)算法与原子参数方法相结合预报了KNO₃-KBr、KNO₃-KI、Cs,Li,Er Cl等含卤化物系的中间化合物形成情况.若干预报已得到实验证实.用留一法对比了SVM算法和若干传统的模式识别算法对AX-BX系和AX₂-BX₂系形成中间化合物,含稀土氯化物的盐系形成A₃B₂C₁₉型和A₂BCl₅型化合物等的预报正确率,结果表明:SVM算法所建立的数学模型的预报正确率比Fisher法和KNN法高.     

基于神经网络模型评价高技术项目投资风险

简要回顾了人工神经网络的发展历史,介绍了信息计量学的概念和人工神经网络算法在信息计量学领域中的应用,警示了在人工神经网络应用中的"过拟合"现象。在对高技术项目投资风险因素分析的基础上,建立了能够预测项目投资风险的线性神经网络模型,模型避免了"过拟合"现象的发生。通过对有关数据进行实例计算,预测结果表明,该神经网络模型稳定可靠,所获得的结果是令人满意的。     

基于改进的SqueezeNet的人脸识别

针对众多基于卷积神经网络的人脸识别技术在追求提高人脸识别率上,忽视了网络模型输入参数,导致模型输入参数多、训练时间长和无法在内存小的硬件上运行等问题,提出一种基于改进的Squeeze Net的人脸识别模型。改进的Squeeze Net模型保留了原网络模型中的小卷积核去提取图片特征,还采用首尾池化层分别引入对应的后续卷积层进行特征融合,提取细微的人脸纹理特征来稳定模型收敛性,防止小的卷积核在复杂的人脸训练集上产生过拟合。针对分类函数Softmax的改进,采用L2范数约束的方法,将最后一层的特征约束在一个球面内,减少相同特征间距,提高网络收敛能力。通过两种改进后的Squeeze Net模型在与其他的先进模型对比,在不降低人脸识别准确率的前提下,具有输入参数少、模型易于收敛和能够运行在内存小的硬件设备的优势。结果在CASIA-WebFace和ORL人脸库上得到了有效性的实验验证。     

基于粗集理论的新决策树剪枝方法

提出了一种基于粗糙集理论的新决策树剪枝方法.在剪枝的过程中,不仅考虑了树的分类精度,而且还考虑了生成树的深度对剪枝的影响;最后针对具体的数据集对新方法进行了验证,得到了较好的效果.