基于变分自动编码器的动态主题模型

传统动态主题模型的后验分布推断需要复杂的推理过程,仅模型假设的细微变化就需要重新进行推断,时间成本较高,制约了模型的可变性和通用性。为了提高动态主题模型的性能,提出了基于变分自动编码器融合动态因子图进行推断的动态主题模型。该模型对变分下界进行再参数化,生成一个下界估计器,将隐变量转换为一组辅助参数,使得新的参数不依赖于变分参数,用标准随机梯度下降法直接优化变分目标,同时融合动态因子图对状态空间模型进行建模,弱化推断的概率特性,简化优化过程,实现有效的推断。结果表明,提出的模型不仅保证了准确性,而且其简化模型有效降低了推断的时间成本,从而为动态主题模型能有效应用于复杂的时间场景提供更多可能。     

基于卷积神经网络的宠物猫品种分类研究

为了提高宠物猫品种分类的准确率,提出了一种卷积神经网络融合的方法进行特征提取。首先,基于堆叠卷积自动编码器的域自适应技术,采用反卷积操作丰富特征图;其次,利用Inception结构增加网络的宽度来提取多尺度信息的特征图;最后,使用Softmox函数对图像进行分类,在Oxford-ⅢT数据集中进行实验分析。实验结果表明,利用改进后的模型对宠物猫进行分类,准确率高于对比模型,达到了84.56%,损失值为0.015 0。所提出的卷积神经网络融合方法不仅能通过丰富特征图、加深网络深度更好地表达特征,还能提高分类性能和收敛性能,较好地解决了宠物品种识别中由宠物相似所带来的识别率低的问题,还可以推广应用到其他图像相似问题的应用场景中。     

基于SVM模型的上市公司财务困境预警研究

本文运用支持向量机技术,以我国上市公司为研究样本,建立了一种新型的财务困境预警模型.通过对该模型的反复调整,得到了较高的预测精度和较好的推广能力.与其它财务困境预警模型相比,我们所建立的支持向量机模型在对上市公司财务困境预警方面具有便于使用、预测精度高等特点.具有广阔的实际应用前景.     

国外包装技术新进展

电子连续运行编码器美国马胄（Markem）公司研制的Markem9840是一种电子连续运行编码器，设计的打印范围为381mm（轴向）476mm（经问）。它可在各种普通包装材料上连续产生优质的印记，每分钟可打四400次，在薄膜上的速度可高达每分钟102m。该编码器在设计上符合欧洲和北美安全标准。9840编码器容易安装在连续和间歇运行的包装机上，这种DC马达传动的9840＄码器不需要来自母机的任何机械式传动（链条、皮带、齿轮等）。它能根据包装机速匿自动地调节，确保任何日刷速医的日刷质量和对准。9840编码器优点仅多，＠括：快速、简单的设定，键…     

基于支持向量机的安全投入需求预测研究

由于煤炭安全投入影响因素较多,所具有的非线性、复杂性与内在随机性,传统的方法难以准确模拟其变化状态并预测未来需求。神经网络是一种较好解决此类问题的方法,但是本身存在诸多缺陷,而这些缺陷支持向量机（Support Vector Machine．SVM）可以克服。本文基于安全投入预测的非线眭,利用统计学习理论的支持向量机,建立基于支持向量机的预测模型。     

基于最小二乘支持向量机的我国能源消费计算方法

本文提出利用最小二乘支持向量机方法定量计算能源消费量， 选取了人口数量、 经济增长、 技术水平、 产业结构、 固定投资、 路网密度、 对外开放、 政策影响等作为能源消费的影响因素， 将样本的影响因素数据作为输入量， 将样本的能源消费量数据作为输出量， 利用训练好的最小二乘支持向量机模型作为能源消费量定量计算的有效工具。 研究结果表明： 基于最小二乘支持向量机模型的计算结果与样本值拟合精度较高， 能较好且客观地反映各控制因素对能源消费量的影响， 对制定科学有效的能源规划和决策具有重要意义。     

基于支持向量机和特征向量提取的人脸识别框架

为了研究支持向量机在人脸识别中的应用,提出了人脸识别框架,该框架首先利用特征向量提取算法对人脸进行特征提取,得到人脸面部纹理特征数据,然后采用支持向量机对提取的数据进行训练,得到人脸模板,并依据人脸模板对人脸进行识别。实验采用ORL人脸数据库作为实验数据,使用LBP算法提取特征向量,使用LIBSVM训练得到人脸模板数据库,当模板人脸数与预测人脸数比值一定时,随着人数增加,其预测的正确率会有所下滑;当人数一定时,人脸模板与预测人脸数值上升,其正确率会有所上升。当选择一个相对合适的模板比例时,正确率将会达到89.29%以上。实验结果表明,提出的框架对于人脸具有良好的识别能力。     

太阳能技术在弯梁变矩抽油机上的应用

为解决弯梁变矩抽油机不能根据油井负荷及产液量自动调整平衡度的问题,利用原抽油机尾游梁主体结构,结合太阳能储能自平衡技术,升级为可自主调节平衡度的弯梁变矩抽油机,现场应用后节电效果良好,故障率降低,具有较好的经济及社会效益。     

基于T-S模型的一类时滞非线性系统模型预测控制

针对具有状态时滞特性的非线性离散系统,利用线性矩阵不等式的方法和Lyapunov稳定性理论,研究了基于状态反馈的非线性系统模型预测控制问题。基于T-S模糊模型对非线性离散系统进行描述并给出一种"min-max"预测控制算法,采用模型预测控制与模糊理论相结合的方法,利用平行分布补偿的原理,通过在每一个采样时刻优化无穷时域的性能指标,来求解在范数有界条件下相应的状态反馈控制律,讨论了预测控制器的设计问题,分析了此设计问题的可解性,给出了状态反馈控制器基于线性矩阵不等式的设计算法,保证了系统的稳定性,通过仿真实例证明了所提控制算法的有效性及系统的稳定性。该方法能够在化工、冶金、机械等具有时滞特性的工业生产过程中得到很好的应用。     

基于轻量级卷积神经网络的苹果表面缺陷检测方法

为了解决目前农业信息领域对苹果表面缺陷检测准确率低的问题,提出一种基于轻量级卷积神经网络的苹果表面缺陷检测方法。首先采集苹果缺陷样本图片制作实验数据集用于模型训练和测试;其次在AlexNet网络结构的基础上,引入深度可分离卷积代替原有网络中的标准卷积运算来进行图像特征的提取;最后利用全局平均池化方法代替原有网络中的全连接层,从而将卷积层输出的多个特征图以自身为单位进行映射得到特征点。实验结果表明:改进后网络对苹果缺陷识别精度达到了98.57%,较改进前提升1.55%;较改进前模型参数量减少99.3%、训练速度提高32.67%、FPS提高33.28%,改进后的轻量级卷积神经网络不仅减少了模型参数量和训练时间,而且提高了检测精度和速度。因此,新的检测方法在减少模型参数量的同时,还可保证模型的检测精度和效率,具有较强的工程实用性,可为苹果缺陷分类提供理论参考。     

Thematic Thinking and Individual Performance in Research and Development

Individuals differ in their preference for processing information on the basis of taxonomic, feature‐based similarity, or thematic, relation‐based similarity. These differences, which have been investigated in a recently emerging research stream in cognitive psychology, affect innovative behavior and thus constitute an important antecedent of individual performance in research and development (R&D) that has been overlooked so far in the literature on innovation management. To fill this research gap, survey and test data from the employees of a multinational information technology services firm are used to examine the relationship between thematic thinking and R&D professionals' individual performance. A moderated mediation model is applied to investigate the proposed relationships of thematic thinking and individual‐level performance indicators. Results show a positive relationship between thematic thinking and innovativeness, as well as individual job performance. While the results do not support the postulated moderation of the innovativeness–job performance relationship by employees' political skill, they show that the relationship between thematic thinking and job performance is fully mediated by R&D professionals' innovativeness. The present study is thus the first to reveal a positive relationship between thematic thinking and innovative performance.     

Online group buying: Some insights from the business-to-business perspective

Extant marketing research has examined online group buying (OGB) from the consumer behavior perspective. The goal of this article is to broaden understanding of OGB by offering insights into the topic from the business-to-business (B2B) perspective. In doing so, it offers insights to help stimulate greater interest in and scholarly work on better understanding and improving B2B relationships and outcomes in OGB. Three discovery-oriented qualitative studies were conducted and analyzed using content analysis with an interpretive approach and contingency theory as a theoretical lens. The findings reveal that marketing exchanges through OGB offers businesses the opportunity to raise corporate and product brand awareness, gain an immediate pool of large groups of customers, expand the customer base, sell other products, contribute to better cash flow, encourage positive word of mouth, and increase sales during non-peak sales periods. However, marketing exchanges through OGB may also be challenging for businesses in terms of maintaining business profitability, managing costs, withstanding competition, obtaining support from OGB website operators, and managing customer expectations of continued discounts. These findings should help businesses that promote and sell their products through OGB to become better informed of the potential benefits and pitfalls of engaging in OGB, and thus improve their navigation in the OGB marketplace to achieve desired marketing outcomes.     

Empirical study of the subjective meanings of culture between American and Chinese

Subjective culture is a group's characteristic way of perceiving its social environment. This study analyzes cultural meanings through free verbal association derived from the Associative Group Analysis technique. More than one hundred important and relevant stimulus words were identified. Respondents were asked to list all the ideas associated with each stimulus word. The purpose is to find out the differences and similarities in the subjective meaning of selected words among American, Hong Kong and PRC Chinese. The research results provide empirical data on the understanding of culture-specific meanings of words. It is cultural experience that produces the unique distribution of association in a particular word. The spontaneously emerging response clusters offer deep insights into what is salient and what is dominant in the group's subjective meanings of selected themes.The authors are from The Chinese University of Hong Kong.     

Nanotechnology for enhancing food security in India

A framework for assessment of the potential of nanotechnology for enhancing food security in India is developed. Agricultural productivity, soil health, water security, and food quality in storage and distribution are identified as the primary determinants of food security that can be impacted by developments in nanotechnology. The framework is developed in two stages: (i) mapping nanotechnology to agri-food thematic areas across the agricultural value chain and (ii) from the thematic areas to the food security determinants. Using published literature and patents data, a model to organize and map nanotechnology research to the agri-food thematic areas and food security determinants is developed through a specially designed database. The model allows identification and prioritization of potential areas for nanotechnology applications to enhance food security. Comparisons of this technology with green revolution technologies and agricultural biotechnology indicate a possibility of greater and faster impact on all components of the agri-value chain with concurrent social, ethical, legal and environmental implications. There is a need for investments in capacity building and development of an agri-nanotechnology infrastructure in India, and for ex ante assessment of its implications for society.     

The maintenance of macro-vocabularies in an industry: The case of the France's recorded music industry

Different studies have proved the importance of language and vocabularies in structuring shared perceptions among stakeholders within an industry. Little is known about the discursive work involved in the evolution of these vocabularies. The use of corpus linguistics provides insights in the process through which an industry stabilises itself and how incumbents engage in discursive work to maintain their interests. We explore the jolt the Internet provoked in the French recorded music industry between 1988 and 2008. We argue that one major explanation for the industry inertia is that the technology of the Internet and the new distribution channels it opened up were framed by discourses, characterized by a relatively stable vocabulary which repeated established words such as right and artistic work. The discursive work of incumbent actors supports the stabilization of this vocabulary. When new words appeared, they were incorporated into existing vocabulary. This dual process of repetition and incorporation was facilitated by the fact that many of the words central to the debate were relatively empty and ambiguous floating signifiers. This paper represents a promising avenue to better account for the place of words in the institutional work and more specifically institutional maintenance.     

A systematic technique for new product idea generation: the external brain

The author outlines a novel computer-based technique to generate ideas for new products. Essentially the technique uses the so-called Dialog databank, claimed to have access to 152 million units of information, to search for all publications in the database which combine a root concept and words such as idea, innovation, novelty, and so on. The computer can then be instructed to print out the titles of a restricted number defined by the searcher from each database accessed. The searcher can then study the contents of each publication at leisure. In principle other similar databanks can be used.
The technique is illustrated by two hypothetical cases: new uses for water and new uses for metal-powder technology.
The author claims that the technique will generate a large number of ideas in a short time at low-cost, can be used selectively (eg to search within a stated time-period), will link otherwise unrelated disciplines, and is well suited to brainstorming. The main drawbacks are that much of the information may be irrelevant, it cannot provide entirely new ideas though it can link existing concepts in a new way, and it cannot print out the total text from an identified source.     

基于稳健性分析的轮转机折页机构优化设计

随着印刷机械速度的提高,在印刷机械设计时,高速与高稳定性已成为相互制约的两个方面。轮转机是印刷设备中速度最高的大型印刷设备,其折页机构对机器运转速度的制约最为显著。为提高轮转机折页机构的速度和稳定性,提出运动精度和动力精度两个稳健性指标。通过对高速运行状态下轮转机折页机构工作状态进行动力学分析,建立其动态响应与设计参数的动力学模型。在动力学模型基础上,建立基于稳健性分析的优化设计模型,以寻求稳健性最佳的设计变量,达到改善轮转机折页机构动力学性能,在保证其折页精度的前提下,提高机构长期高速运转稳定性的目的。设计实例表明,在设计变量稳定的情况下,稳健设计与轨迹优化设计的机构运动性能差别不大,但稳健设计的机构动力学性能要远远优于轨迹优化设计。在设计变量发生微小改变的状态下,运用稳健设计方法所得参数能使机构保持良好的稳定性。     

高频视角下创业板市场的非对称尾部风险度量

创业板汇集了大批高成长性的科技企业,面临着高度不确定的市场风险。本文考虑创业板市场的非对称性和厚尾性,在偏t分布下利用5分钟高频数据构建已实现GARCH模型衡量其波动率,同时考虑到微观结构噪声影响,在RV的基础上引入对噪声稳健的BPV已实现测度,从而提升高频VaR模型的预测准确性。对创业板指和深证成指的实证结果表明,已实现GARCH模型在参数显著性和模型预测准确性方面均优于低频下的ARMA-fGARCH-VaR模型;基于RV和BPV的已实现模型没有显著差异;相比于深市主板,创业板面临的尾部风险更大。同时,为更好地监测与管控创业板市场风险,从政府和创业企业两个层面提出对策建议。     

French agriculture : Trends, outlook and policies

This article presents a general review of past trends, which are characterized by rapid modernization, making up in part for marked retardation. France has thus become a major food exporter. This situation and the general economic circumstances require more realistic farm policies. A plea is made for continued expansion, based not on export restitutions but on prices closer to stabilized world levels. With improved training and advisory services, as well as intensified structural policies, the better areas can stand lower prices. Disfavoured areas cannot do so and require special measures. Above all, fuzzy agrarianistic philosophies must give way to more realism and outward-looking attitudes.     

Mapping the Topic Landscape of JPIM, 1984–2013: In Search of Hidden Structures and Development Trajectories

During the three decades since its inception in 1984, the JPIM has shaped the evolution of innovation research as a scientific field. It helped create a topic landscape that is not only more diverse and rich in insights, but also more complex and fragmented in structure than ever before. We seek to map this landscape and identify salient development trajectories over time. In contrast to prior citation‐based studies covering the first two decades of JPIM research, we benefit from recent advances in natural language processing and rely on a topic modeling algorithm to extract 57 distinct topics and the corresponding most common words, terms, and phrases from the entire full‐text corpus of 1008 JPIM articles published between 1984 and 2013. Estimating the development trajectory of each topic based on yearly publication counts in JPIM allows us to identify “hot,” “cold,” “revival,” “evergreen,” and “wall‐flower” topics. We map these topics onto the Product Development and Management Association (PDMA) Body of Knowledge categories and discover that these categories differ significantly not only in terms of their internal topic diversity and relative prevalence, but also—and arguably more importantly—in terms of their publication and citation trajectories over time. For instance, the PDMA category “Codevelopment and Alliances” exhibits only moderate topic diversity (7 out of 57 topics) and prevalence in JPIM (161 out of 1008 articles). That said, it is among the most dynamic categories featuring two evergreen topic (“Users and Innovation” and “Tools and Systems for Technology Transfer”) and three hot topics (“Open Innovation,” “Alliances and Cooperation,” and “Networks and Network Structure”) as well as a sharply growing annual number of citations received. Our findings are likely to be of interest to all those who are keen to (re)discover JPIM's topic landscape in search of hidden structures and development trajectories.