基于EMD距离的稀疏自编码器

KL散度在机器学习领域被广泛地用于模型损失函数之中来度量分布的距离。在稀疏自编码器中,KL散度被用作损失函数的惩罚项来度量神经元输出与稀疏参数的距离,使得神经元输出趋近稀疏参数,从而抑制神经元的激活以得到稀疏编码。在WGAN中,Wasserstein距离被用于解决GAN的梯度消失和模式塌陷问题,使得GAN的训练更加稳定。得益于Wasserstein距离在GAN中的成功应用,提出了基于EMD距离的稀疏自编码器SAE-EMD。实验结果表明,相比于使用KL散度与JS散度,使用EMD距离作为惩罚项的稀疏自编码器可以使得真实样本与重构样本之间的重构误差减小,并且随着惩罚参数的增大,编码更加稀疏。     

面向微博内容的信息抽取模型研究

社会媒体是人们用来分享意见、见解、观念和经验的平台或工具,目前已经发展成具有重大影响力的新媒体。而微博作为社会媒体的一个重要部分,对信息的传播起到了很大的作用。面向微博内容的信息抽取就是要从充满噪音的、零碎的、非结构化的微博内容的自由文本中提取有价值的结构化的信息,以利于从微博内容中有效地获取信息。提出了一种基于因子图的微博事件抽取方法来准确地抽取微博中所反映的事件。最后通过实验验证了该方法在性能和准确性上都比其他的方法要高。     

仿真结果检验和因子分析的两步仿真法

建立多状态两步仿真检验法,可以对均值仿真结果进步检验,确定仿真结果与试验结果之间的误差是由于随机因素引起的偶然误差,还是由于仿真不正确引起的系统误差,进而判断仿真结果正确与否.同时提出多因子两步仿真分析法,能够对仿真结果的各个影响因素(因子)逐一进行分析,并确定哪些因素已被正确仿真,哪些因素尚未被正确模拟,从而指导仿真软件的编制和仿真结果的修正.文中方法适用于正态分析、非正态分布以及形式未知的情况,且事先不必对工程实际中遇到的随机变量分布作任何假设,具有精度高的特点.     

浅析发动机调速率高的原因

本文阐述了发动机调速率高带来的危害,并以我公司生产的KM160为例,介绍了调速器的工作原理,以及在解决该问题时所采取的技术措施,通过单因素法、正交试验法优选出最佳方案组合。经过试验和生产验证这些改进措施行之有效,在以后的批量生产中没有出现调速率高的问题,同时也为解决其它机型提供了借鉴。     

Analysis of factors influencing the Henry Hub natural gas price based on factor analysis

Henry Hub as an important transaction hub for natural gas sets the gas price standard in the USA. In this paper, the factors influencing Henry Hub natural gas prices are analyzed, and the major factors determining the price levels in the period between January 1997 and December 2016 are studied. It is found that economic conditions, total energy demand, US dollar exchange rate and gas consumption are the major factors. The mechanism of each factor influencing the Henry Hub natural gas price is also explored in the paper.     

The validation of the E-Victimisation Scale (E-VS) and the E-Bullying Scale (E-BS) for adolescents

This study investigates the psychometric properties of the E-Victimisation Scale (E-VS) and E-Bullying Scale (E-BS) designed to assess Cyber Bullying among Chinese adolescents. Participants were 484 adolescents aged between 11–16 years randomly recruited from high schools within a region. Exploratory Factor Analysis (EFA) and Confirmatory Factor Analysis (CFA) were applied to investigate the factorial structure of these scales. Reliability was examined by Cronbach’s alpha coefficients by sex. The convergent validity was investigated by correlations among these scales and the Centre for Epidemiological Studies-Depression for Children as well as the Zung’s Anxiety Scales. A single-factor model for the E-VS and a 2-factor model for the E-BS were resulted from the EFA with large factor loadings and about 47% and 56% of variance explained respectively. Cronhach’s alpha values provided evidence for good internal reliability with values ranging from 0.55 to 0.96. Correlations between the E-VS and Depression as well as Anxiety scales showed positive and significant relationships, however, the E-BS was only related to Depression. Psychometric evidence has shown that both E-VS and E-BS are valid instruments for measuring Cyber bullying behaviour and victimisation. Further studies are required on the test–retest reliability, discriminate validity, responsiveness, as well as normative information for standardisation.     

基于最佳鉴别变换和分类器组合的人脸自动识别

人脸识别技术在商业和法律上有广泛的应用前景 ,在安全监控中也大有用武之地 .其主要任务是利用已有的人脸图象库 ,识别静止的或视频图象中的一张或多张人脸 .从抽取具有统计不相关的模式特征着手 ,通过基于小波变换的图象分解和 KL 变换等处理 ,避开人脸识别的小样本集的局限 ,并通过运用具有统计不相关性的最佳鉴别变换 ,来抽取人脸的有效鉴别特征 .同时 ,利用多特征多分类器组合的方法对图象进行识别 .该方法在 ORL 人脸图象库上进行实验 ,得到识别错误率为 2 %的实验结果 ,这是目前在此人脸图象数据库上所得到的最好的实验结果 .而且本方法对人脸的姿态、表情等条件具有一定的不敏感性     

大功率LED驱动电路研究设计

根据大功率LED的供能要求,从EMI滤波、功率因素校正、半桥谐振转换三个方面着手,以FAN6961和FSFR2100为控制芯片,设计了一款大功率的高效率LED驱动电路,在90～264VAC的线路输入和满载下,功率因数高于93%,效率高于85%,并具有低输入电流谐波失真和低EMI.     

基于最佳邻域重构指数的水下高光谱目标检测

水下机器人仅通过传统光学相机获取图像很难在复杂水下环境中或目标物具有保护色的情况下检测到目标,而通过高光谱技术进行水下目标检测可以改善这一情况;由于直接运用传统高光谱检测方法难以满足水下机器人对水下目标检测的要求,提出了一种基于最佳邻域重构指数(ONRIF)的高光谱目标检测方法,该方法通过线性重构的思想进行邻域寻优,选出信息量高且波段相关性低的波段组合,并使用所选波段的融合图像进行目标检测;结果表明,与直接对原始水下海产品高光谱图像进行检测相比,该方法在保证检测效果的前提下,大量减少了检测时间和数据冗余程度;还提出了一种在相同环境下对同类目标物的单波段快速采集检测方法,大大提高了采集数据的速度,可以满足水下机器人对海产品检测的需求.     

Psychometric equivalence of an improved Basic Language Morningness (BALM) Scale using industrial population within comparisons

An existing 13-item composite circadian rhythm ‘morningness’ scale was simplified into a ‘(BA)sic (L)anguage (M)orningness’ (BALM) scale at a seventh grade (12–13 years) reading level. It was given to 150 each of continuously rotating shiftworkers and fixed-schedule dayworkers in an industrial manufacturing plant. Multiple factor analyses confirmed three prior identified factors, morningness/effort, morning alert, and evening. Morningness scores for dayworkers were higher than for shiftworkers (p>000l). The BALM scale revision is suitable for most worker levels. A suggested psychometrically ‘purer’ 9-item morningness scale omits the inconsistent evening factor scale items.