基于机器学习的斜坡堤越浪量预测方法研究

针对斜坡堤越浪量预测方法,分别建立集成神经网络(ensemble neural network,ENN)、随机森林(random for-eset,RF)和支持向量回归机(suppport vector regression,SVR)3种机器学习模型对斜坡堤越浪量进行预测,并利用决定系数R2和均方根误差RMSE来评估模型性能.最后,对3种模型的性能进行分析.结果显示,集成神经网络模型的决定系数R2和均方根误差RM S E分别约为0.96和0.0018,随机森林模型的决定系数R2和均方根误差RMSE分别约为0.97和0.0014,支持向量回归机模型的决定系数R2和均方根误差RMSE分别约为0.94和0.002.对比发现,3种模型的决定系数都达到0.9以上,都具有较高的预测精度,随机森林相比其他两个模型精度更高.     

Microfinance and social capital formation- a social network analysis approach

The microcredit program has emerged as an important poverty alleviation strategy over the last three decades, and several studies have examined its economic impacts on the community well-being. However, far too little attention has been paid to the effects of micro credits on community social connection and solidarity. This paper aims to examine the application of Social Network Analysis (SNA) to explore the impact of the rural microcredit fund on community social capitals. In doing so, the data on interactions of four rural development groups' members before and after the microcredit project implementation were collected using participatory workshops in Neyzar village of Qom province in Iran. The data were analyzed by Ucinet software, and the socio-graphs were produced by the NetDraw application. The results show that, more people have been involved in the social interactions after the project implementation and there was statistically significant increase in density and decrease in centralization of cooperation network. Furthermore, there were no important distinctions in centrality of people with various educational levels before and after the project implementation. Overall, it can be concluded that, the microfinance initiative considerably promotes the community social capital and participation in the rural development activities. Moreover, the SNA techniques are applicable as an impact assessment tool to investigate changes in community social capital.     

移动网络数据中大数据处理的关键技术探索

最近几年以来,我国科技水平持续提升,计算机网络技术也逐渐在各行各业中获得了广泛运用,这也使得信息技术的深度融合发展受到一定的促进,数据规模愈发庞大,在此背景下,开始演变成大数据供不应求的状况。为此,国内陆续推出了一系列相关的管理法规与条例,同时扩大了对大数据的探究创新力度,有效提升了移动通信网络处理数据的速度,这同时促使国内移动通信网络中有关大数据的研究内容更为明晰,由此可知,针对大数据的相关研究具有一定必要性。     

基于MBP算法和深度学习的人脸识别

为了解决在深度学习提取人脸图像特征时,易忽略其局部结构特征和缺乏对其旋转不变性学习的问题,提出了一种基于单演局部二值模式(MBP)与深度学习相结合的高效率人脸识别方法。首先,用多尺度单演滤波器对图像进行滤波,得到幅值和方向信息;其次,用LBP算法和象限比特的方法进行编码,分块计算组合其直方图特征;然后,将提取的单演特征作为深度信念网络(DBN)的输入,逐层训练优化网络参数,得到优异的网络模型;最后,将训练好的DBN网络在ORL人脸数据库上进行人脸识别实验,进行识别率计算,其识别率为98.75%。所提出的方法使用无监督的贪婪算法,隐藏层设定为2层,使用反向传播算法优化网络。相较于已知的人脸识别方法,MBP+DBN算法对光照、表情和部分遮挡变化具有较好的鲁棒性,在人脸识别中识别率较高,具有一定的优势,为图像特征提供了一种新的识别方法。     

基于PSO的信息熵数据融合非均匀分簇路由算法

针对无线传感器网络分簇算法中能量分布不均衡导致的"热区"和簇头负载过重问题,提出了一种基于PSO算法优化簇头选举的非均匀分簇算法。在候选簇头选举和竞争半径计算过程中综合考虑节点动态能量、节点密度和节点距基站距离,将网络进行非均匀分簇,并引入PSO算法进行最终簇头选举。根据节点能量、节点密度和距基站距离确定簇间单跳多跳结合的路由规则,选取代价函数小的节点作为下一跳节点。基于节点信息熵确定融合阈值,进行簇内数据融合剔除冗余数据。仿真结果表明,改进算法的数据传输量比EEUC算法和UCRA算法分别提高了20%和10%,提升了数据的融合效率,有效延长了网络生命周期,簇头能量消耗得到均衡,减少了网络能量消耗,网络的整体性能显著优于其他对比算法。     

全球价值链下中国增加值贸易的核算及网络特征研究

基于投入产出模型和社会网络分析，利用世界投入产出表相关数据，对2000—2014年中国的增加值贸易进行核算，并分析全球增加值贸易的网络特征。研究结果显示：（1）从出口目的地来看，中国向美国、日本、德国等国家的增加值出口较大；（2）从行业来看，中国的纺织业、除汽车和摩托车之外的其他产品批发业、采矿业等行业的增加值出口较大；（3）根据网络密度的计算结果，世界范围的各国（地区）贸易联系程度在增强；（4）中国的相对点度数和点强度有上升的趋势，而日本和美国有下降的趋势，说明中国在世界增加值贸易格局中的地位在提升，不过美国仍在世界增加值贸易格局中占据主导地位；（5）核心边缘分析结果表明核心国家（地区）的数目经历了先增加后减少的过程，边缘国家（地区）数目则先减少后增加。其中，中国的核心度一直在增加，日本和美国的核心度呈现下降的趋势。因此，为了扩大增加值贸易，增强国际贸易的话语权，中国政府有必要采取调整进出口税率等政策，并重视与美国等国家的双边贸易合作，以实现双赢的结果。     

Do ongoing networks block out new friends? Reconciling the embeddedness constraint dilemma on new alliance partner addition

Research Summary: This study addresses a theoretical dilemma regarding how alliance network constraint (reflected by network cohesion) affects a firm’s alliance formation with new partners. Using a network pluralism approach, we separate a firm’s ego alliance network into two activity‐based networks—an exploratory network and an exploitative network—based on the primary value chain activity involved in each alliance. We argue that the cohesion of exploratory or exploitative networks has an inverted U‐shaped effect on the addition of new partners in the same activity‐based network, and a positive effect on the addition of new partners in the other network. Results based on data from the biotechnology industry largely support our predictions with one exception. Our study contributes to both scholarly understanding of network embeddedness and alliance practice. Managerial Summary: The structure of firms’ ongoing alliance networks may have paradoxical implications for their efforts to search for and form alliance with new partners. That is, when a firm’s alliance partners are tightly connected with each other, the cohesive network tends to both encourage and impede the focal firm to add new partners. We resolve this dilemma by showing that when a firm is deeply entrenched in a cohesive alliance network conducting a certain type of activities (e.g., R&D activities), it may not easily add new R&D alliance partners. However, it may still be able to escape from the cohesive R&D alliance network by seeking new partners conducting other activities (e.g., manufacturing activities).     

The problem of employees' network centrality and supervisors' error in performance appraisal: A multilevel theory

This article reveals an unexplored paradox for HR managers: the centrality of an employee in the social network benefits performance but hampers performance appraisal because it affects supervisors' rating errors. Central employees can be erroneously rated high on performance even when they are not high performers because supervisors tend to overappraise their performance. A distinction is made between rating precision, which depends on supervisors' uncertainty regarding employees' performance, and rating accuracy, which depends on supervisors' bias in favor of employees. Employee centrality is posited to be beneficial to precision but deleterious to accuracy because it regulates the diffusion of positive information, status, and power, all of which distort supervisors' capacity and motivation to accurately appraise performance. It is then argued that rating errors caused by network centrality affect aggregate perceptions of justice in organizations. When employees are highly connected to each other in a dense network, organizations have a strong and positive justice climate. Yet when some employees are more central than others in a centralized network, organizations have a negative and weak justice climate. The article contributes to the literature because it identifies an unexplored dark side of network centrality and offers recommendations for HR managers to cope with its deleterious consequences and for scholars to study them.     

Conditional risk measures in a bipartite market structure

In this paper, we study the effect of network structure between agents and objects on measures for systemic risk. We model the influence of sharing large exogeneous losses to the financial or (re)insurance market by a bipartite graph. Using Pareto-tailed losses and multivariate regular variation, we obtain asymptotic results for conditional risk measures based on the Value-at-Risk and the Conditional Tail Expectation. These results allow us to assess the influence of an individual institution on the systemic or market risk and vice versa through a collection of conditional risk measures. For large markets, Poisson approximations of the relevant constants are provided. Differences of the conditional risk measures for an underlying homogeneous and inhomogeneous random graph are illustrated by simulations.     

A gossip based information fusion protocol for distributed frequent itemset mining

The computational complexity, huge memory space requirement, and time-consuming nature of frequent pattern mining process are the most important motivations for distribution and parallelization of this mining process. On the other hand, the emergence of distributed computational and operational environments, which causes the production and maintenance of data on different distributed data sources, makes the parallelization and distribution of the knowledge discovery process inevitable. In this paper, a gossip based distributed itemset mining (GDIM) algorithm is proposed to extract frequent itemsets, which are special types of frequent patterns, in a wireless sensor network environment. In this algorithm, local frequent itemsets of each sensor are extracted using a bit-wise horizontal approach (LHPM) from the nodes which are clustered using a leach-based protocol. Heads of clusters exploit a gossip based protocol in order to communicate each other to find the patterns which their global support is equal to or more than the specified support threshold. Experimental results show that the proposed algorithm outperforms the best existing gossip based algorithm in term of execution time.