Harmonic broadcasting for video-on-demand service

Using conventional broadcasting, if we want to support a 120-minute popular movie every 10 minutes, we need 12 video channels. Assuming the set-top box at the client end can buffer portions of the playing video on a disk, pyramid broadcasting schemes can reduce the bandwidth requirements to 5.7 channels. We present a new scheme which only needs 3.2 channels. For a movie with length D minutes, if we want to reduce the viewer waiting time to D/N minutes, we only need to allocate H(N) video channels to broadcast the movie periodically, where H(N) is the harmonic number of N, H(N)=1+1/2+…+1/N. In order to support video-on-demand service for a popular movie, the new scheme greatly reduces the bandwidth requirements     

Analysis of epoxidized soybean oil by gas chromatography

A fast and cost-effective procedure to quantitate epoxidized soybean oil by means of an external standard method is reported. This procedure is applicable to commercial epoxidized oils, polymer additive packages and polymers—polyvinyl chloride (PVC)—containing epoxidized oils. The epoxidized soybean oil is converted into fatty acid methyl esters with tetramethylammonium hydroxide, and analyzed by capillary gas chromatography with flame-ionization detection. In PVC samples, the epoxidized soybean oil was extracted with toluene and followed by derivatization prior to analysis. The methyl esters of monoepoxyoctadecanoic, diepoxyoctadecanoic and triepoxyoctadecanoic acid were separated with a short capillary column.     

基于改进HRNet的眼底视网膜血管分割算法

针对现有眼底视网膜血管分割算法普遍存在的微小血管细节丢失和病灶信息误判等问题,提出一种基于改进HRNet的血管分割算法.在预处理阶段,利用限制对比度自适应直方图均衡化和自适应的Gamma矫正提高血管与背景对比度;在编码阶段,将HRNet原始卷积替换为可变形卷积,提升卷积对复杂血管形态结构的适应能力;在多尺度特征融合阶段,引入空间金字塔池化和多尺度卷积,扩大感受野同时增强对目标局部特征关注度,改善血管伪影和细微信息丢失的问题.该算法在DRIVE数据库上仿真实验,其准确率、灵敏度和特异性分别为95.79％、80.33％和98.12％.     

基于生成对抗网络的文本生成图像算法

文本生成图像算法对生成图像的质量和文本匹配度有很高的要求. 为了提高生成图像的清晰度, 在现有算法的基础上改进生成对抗网络模型. 加入动态记忆网络、细节校正模块(DCM)、文本图像仿射组合模块(ACM)来提高生成图片的质量. 其中动态记忆网络可以细化模糊图像并选择重要的文本信息存储, 以提高下一阶段生成图像的质量. DCM纠正细节, 完成合成图像中缺失部分. ACM编码原始图像特征, 重建与文本描述无关的部分. 改进后的模型实现了两个目标, 一是根据给定文本生成高质量的图片, 同时保留与文本无关的内容. 二是使生成图像不再较大程度依赖于初始图像的生成质量. 通过在CUB-200-2011鸟类数据集进行研究实验, 结果表明相较之前的算法模型, FID (Frechet inception)有了显著的改善, 结果由16.09变为10.40. 证明了算法的可行性和先进性.     

集群环境中微服务容器资源特征分析及优化

在集群环境中部署微服务已经成为微服务部署的重要方式.由于不同种类服务对于CPU、内存、磁盘等资源的需求不同,导致集群中的节点产生资源碎片、出现资源消耗倾斜.如何提高集群资源利用率、降低集群能耗,成为继保障服务级别协议(service level agreement, SLA)之后的重大挑战.本文以阿里巴巴集团2021年发布的近两万个微服务的详细跟踪为数据样本,从容器资源使用情况、节点部署特征和资源消耗偏好等多个维度出发,分析其集群资源消耗特征,发现集群中出现了资源消耗倾斜的情况.通过进一步分析节点中容器部署情况发现容器资源分配不合理加剧了这一现象.基于此我们提出了一种使用深度双Q网络的模型,依据上游服务资源需求的实时变化,对容器资源分配进行优化.对比实验结果表明该方法可以在保证服务SLA的情况下有效提高容器资源利用率,改善节点资源消耗倾斜的情况.     

Electronic and junction properties of poly(2,5-dimethoxyaniline)-polyethylene oxide blend/metal Schottky diodes

Schottky barrier diode devices were fabricated in a sandwich configuration with a blend film consisting of a conducting polymer, poly(2,5-dimethoxyaniline), PDMA in an insulating matrix, polyethylene oxide (PEO). The influence of two different dopants, sulphate anion (SA) or methane sulfonate anion (MSA), in the electronic properties of the device was followed using the devices: ITO/PDMA (SA)-PEO/Al and ITO/PDMA (MSA)-PEO/Al. Current (I)-Voltage (V) characteristics were recorded for making a comparative evaluation of the electronic and junction properties of the devices. The junction and electronic parameters were analyzed and compared in the light of differences in the electronic state, morphology and transport of carriers. The device turn on voltage was found to be higher for Al/ PEO-PDMA (MSA)/ITO (∼3.0 V) in comparison to Al/ PEO-PDMA (SA)/ITO (∼2.85 V). The electronic states of PDMA doped with SA or MSA dopant were ascertained by optical UV-Visible spectroscopy. AC-impedance measurements were made for the devices and the values of bulk resistance (R_c), depletion resistance (R_d) and depletion layer width (W) were deduced through a proposed equivalent circuit. W for the device with PEO-PDMA (MSA) is more (∼24 nm) than the device with PEO-PDMA (SA) (∼8.5 nm). The observed higher Φ_B for PDMA-PEO (MSA) is consistent with this observation.     

An Enhanced Drawing Reproduction Graphical Password Strategy

Passwords are used in the vast majority of computer and communication systems for authentication. The greater security and memorability of graphical passwords make them a possible alternative to traditional textual passwords. In this paper we propose a new graphical password scheme called YAGP, which is an extension of the Draw-A-Secret (DAS) scheme. The main difference between YAGP and DAS is soft matching. The concepts of the stroke-box, image-box, trend quadrant, and similarity are used to describe the images characteristics for soft matching. The reduction in strict user input rules in soft matching improves the usability and therefore creates a great advantage. The denser grid granularity enables users to design a longer password, enlarging the practical password space and enhancing security. Meanwhile, YAGP adopts a triple-register process to create multi-templates, increasing the accuracy and memorability of characteristics extraction. Experiments illustrate the effectiveness of YAGP.     

几何活动轮廓模型中停止速度场的异性扩散

几何活动轮廓(GAC)模型广泛应用于计算机视觉和图像分析领域,特别是用于定位目标边界.然而,基于GAC模型的图像分割有演化时间长和边界泄漏两个缺点.一方面,停止速度场在同质区域一般不够平滑,这导致活动轮廓不能快速演化到希望的目标边界;另一方面,停止速度场在目标边界上不为0,导致活动轮廓不能停止于目标边界,活动轮廓继续演化进入目标边界内(边界泄漏).针对这两个问题,提出了一种对停止速度场进行各向异性扩散的方法.它基于提出的各向异性扩散模型.然后把各向异性扩散后的停止速度场应用于GAC模型进行图像分割.实验结     

基于主Agent信念修正的推测计算及其资源协商

基于假设推理(abduction-based)的推测计算(speculative computation)是在资源信息不能及时到达时,利用缺省假设进行计算的过程.在计算过程中,如果应答和信念不一致,则主Agent将修正它的信念.为了实现目标,在有限时间内使推测计算的结果更精确,主Agent要通过协商获得尽可能多的实际信息,协商是降低决策风险的主要途径.在介绍假设推理和推测计算的基本原理的基础上,提出了基于时间约束的推测计算扩展框架、基于时间约束的进一步协商框架和基于信念修正的协商算法,并将进一步协商框架和协商算法嵌入到推测计算的过程中,在协商过程中赋予主Agent更强的信念修正能力.最后,在货物运输领域的实验中,证实了基于信念修正的推测计算的有效性.