影视节目扁平化标签获取技术研究

针对影视节目标签手动采集费时费力,以及传统树状标签体系信息冗余且不全面等问题,提出一种标签自动获取技术。通过数据爬取技术采集与节目相关的互联网原始数据,然后通过文本分析、同义匹配、数据库匹配等技术进行数据分析与挖掘,最终实现对扁平化节目标签的获取。实验结果表明,在选取8～10个标签时,该算法准确率为84.3%~ 92.4%,召回率为53.4%~ 63.1%,说明该算法获取的标签能够很好地对影视节目进行描述。     

Algorithms for semi on-line multiprocessor scheduling problems

In the classical multiprocessor scheduling problems, it is assumed that the problems are considered in off-line or on-line environment. But in practice, problems are often not really off-line or on-line but somehow in between. This means that, with respect to the on-line problem, some further information about the tasks is available, which allows the improvement of the performance of the best possible algorithms. Problems of this class are called semi on-line ones. The authors studied two semi on-line multiprocessor scheduling problems, in which, the total processing time of all tasks is known in advance, or all processing times lie in a given interval. They proposed approximation algorithms for minimizing the makespan and analyzed their performance guarantee. The algorithms improve the known results for 3 or more processor cases in the literature. Project supported by the National Natural Science Foundation of China (Nos. 19701028 and 19971078) and National 973 Research Project of China.     

基于YOLOv2模型的道路目标检测改进算法

针对传统道路目标检测算法推荐窗口冗余、鲁棒性差、复杂度较高的问题,提出基于YOLOv2模型的道路目标检测改进算法。相较于传统的HOG+SVM目标检测算法,YOLO模型优势在于提升了检测速度及准确度,更适用于实时目标检测。比较YOLO V3 与 YOLO V2算法,前者在构造神经网络模型时复杂度较高,故最终选择YOLO V2算法。针对原算法中选取Anchor Boxes时所采用的K-MEANS算法造成的目标物体框冗余问题,以及原算法对于不规则物体以及遮挡物体检测效果较差等问题,提出基于YOLO V2模型的一种改进方法,将K-MEANS算法改进为一种DA-DBSCAN算法,通过动态调整参数的方式大大减少了锚点框冗余问题。实验表明,改进后的模型准确率达到96.76%,召回率达到96.73%,检测帧数达到37帧/s,能够满足实时性要求。     

搜索引擎中分类技术研究

本文提出了一种基于主题采集的Web文档自动分类算法,该算法对朴素贝叶斯分类模型进行了改进。利用该算法,我们实现了一个基于主题信息采集的网页分类系统。文中着重对该系统的页面解析、中文分词和文本分类模块进行了论述,并对改进后的贝叶斯分类方法进行了评估。实验结果表明,该算法对网页分类有较高的准确性。     

一种改进的K-Modes聚类算法

为了改善传统K-Modes聚类算法相异度度量公式弱化了类内相似性,忽略了属性间差异,以及单一属性值的Modes忽视了某一属性可能存在多属性值组合,且算法受初始中心点影响很大的缺点,基于多属性值Modes的相异度度量方法提出MAV-K-Modes算法,并采用一种基于预聚类的初始中心选取方法。使用UCI数据集进行实验,结果表明,MAV-K-Modes算法相比于传统K-Modes算法,其正确率、类精度和召回率都有明显提升,且MAV-K-Modes算法适合于并行化改造。     

Predictive control of a class of bilinear systems based on global off-line models

INTRODUCTION Bilinear systems are a kind of important nonlinear systems with relatively simple structure, and many industrial processes can be described as a bilinear system. Thus research on the control of this kind of systems is very important. On the other hand, model predictive control (MPC) (Clarke et al., 1987) has been widely used in industrial applications and many predictive control methods focusing on bilinear systems are emerging (Bloemen et al., 2001; Fontes et al., 2004; He…     

Subspace identification for continuous-time errors-in-variables model from sampled data

We study the subspace identification for the continuous-time errors-in-variables model from sampled data. First, the filtering approach is applied to handle the time-derivative problem inherent in continuous-time identification. The generalized Poisson moment functional is focused. A total least squares equation based on this filtering approach is derived. Inspired by the idea of discrete-time subspace identification based on principal component analysis, we develop two algorithms to deliver consistent estimates for the continuous-time errors-in-variables model by introducing two different instrumental variables. Order determination and other instrumental variables are discussed. The usefulness of the proposed algorithms is illustrated through numerical simulation.     

基于决策树的日志分析方法

为了解决服务器运行过程中由于性能故障造成服务质量下降的问题,提出一种基于决策树的日志分析方法,以服务器日志文件中记录服务器关键性能指标的数据为研究对象,利用决策树中常用的ID3、C4.5和CART 3种算法预测服务器未来性能指标发展趋势。实验结果表明,在实际运行过程中,C4.5算法对服务器性能指标数据预测的准确率和召回率最好,分别达到了92.23%和95.37%,在3种决策树算法中拥有最高的准确率与召回率,且相比传统开发人员从日志文件中寻找故障的方法,准确率提高了20%左右,因此能够更好地预测服务器系统性能指标发展趋势。通过该方法可提前感知系统运行状况,并及时作出调整,从而有效降低实际生产过程中服务器故障发生概率,提高服务质量。     

基于时效性的爬虫调度

搜索引擎作为互联网信息获取的入口，实现高效、准确的信息获取非常重要，爬虫作为搜索引擎的上游，其重要性不言而喻，特别是大数据时代信息更新频繁，如何在第一时间获取新闻是实现爬虫时效性的重要因素。为了充分利用有限资源，提升带宽利用率，设计一种基于历史数据预测的爬虫调度算法。该算法通过抓取网站历史，更新频次积累数据，使用随机森林回归建立模型，并在系统中实现爬虫调度。实验结果表明，该策略在抓取新链的命中率上提升了46%，平均成本降低了11%，平均抓取延时降低了14%。     

Simulation of sludge blanket height in clarifiers

Sludge blanket height (SBH) is an important parameter in the clarifier design, operation and control. Based on an overview and classification of SBH algorithms, a modified SBH algorithm is proposed by incorporating a threshold concentration limit into a relative concentration sharp change algorithm to eliminate the disturbance of compression interfaces on the correct simulation of SBH. Pilot-scale test data are adopted to compare reliability of three SBH algorithms reported in literature and the modified SBH algorithm developed in this paper. Calculated results demonstrate that the three SBH algorithms give results with large deviation (> 50%) from measured SBH, especially under low solid flux conditions. The modified algorithm is computationally efficient and reliable in matching the measured data. It is incorporated into a one-dimensional clarifier model for stable simulation of pilot-scale experimental clarifier data and into dynamic simulation of a full-scale wastewater treatment plant (WWTP) clarifier data. Project supported by the National Natural Science Foundation of China (Grant No.501380107)     

融合位置相似性度量的快消品电商网站推荐算法

提出一种融合位置相似性度量的协同过滤推荐算法（CF-FLSM）。算法融合位置相似性度量进行加权计算用户间的兴趣相似度,从而为目标用户产生推荐结果。将CF-FLSM应用于一个具体的快消品电商网站,得出的推荐结果与传统使用余弦相似性的协同过滤推荐算法（CF）相比,精确率和召回率分别提高了3.74%和3.91%。     

基于 YOLOv3 算法的教室学生检测与人数统计方法

教室环境中人群密度高、重叠部分较大的特点会给人数统计工作带来困难。基于深度学习 YOLOv3 目标检测算法对学生目标进行检测,并加入 Deep-Sort 算法为核心实时追踪方法,对 YOLOv3 算法检测到的学生目标进行一段时间的跟踪,从而克服传统视频人数统计方法中忽略视频上下帧关联信息的缺点,并且能更好地解决视频遮挡问题。目标检测方法中的损失函数用 tan 方损失函数代替原有交叉熵损失函数,跟踪算法中的卡尔曼滤波算法采用 Levenberg-Marquardt 对修正后的轨迹预测予以优化。最后,对该方法进行性能评价与对比实验,包括是否加入上下文信息对比以及与 SVM 目标检测算法作对比。实验结果表明,此方法在测试集中,加入上下文信息后准确率达 93.4%,召回率达 81.4%。对比 SVM,该方法在教室视频人数统计中准确率提升2.1%,召回率提升 8.9%。     

基于公平性和QoS保障的OFDMA-WLAN系统资源分配(英文)

为了满足OFDMA-WLAN系统下行通信中多用户的不同业务需求,提出一种基于公平性和QoS服务保障的资源分配算法.不同的QoS要求被转换成不同的速率要求来计算QoS满意等级;优化目标被修改为公平性驱动的优化函数以提供公平性保障;复杂的资源分配问题被划分为信道分配和功率分配问题,并通过二分图匹配和注水法得到分配结果.与其他算法相比,所提出的算法牺牲了较少的数据速率换取更高的公平性和QoS满意度.仿真结果表明所提算法具有保障QoS和公平性的能力,且在QoS、公平性和速率之间权衡折中时表现更好.     

Robust video foreground segmentation and face recognition

Face recognition provides a natural visual interface for human computer interaction (HCI) applications. The process of face recognition, however, is inhibited by variations in the appearance of face images caused by changes in lighting, expression, viewpoint, aging and introduction of occlusion. Although various algorithms have been presented for face recognition, face recognition is still a very challenging topic. A novel approach of real time face recognition for HCI is proposed in the paper. In view of the limits of the popular approaches to foreground segmentation, wavelet multi-scale transform based background subtraction is developed to extract foreground objects. The optimal selection of the threshold is automatically determined, which does not require any complex supervised training or manual experimental calibration. A robust real time face recognition algorithm is presented, which combines the projection matrixes without iteration and kernel Fisher discriminant analysis (KFDA) to overcome some difficulties existing in the real face recognition. Superior performance of the proposed algorithm is demonstrated by comparing with other algorithms through experiments. The proposed algorithm can also be applied to the video image sequences of natural HCI. Project supported by the National Natural Science Foundation of China (Grant No.60872117), and the Leading Academic Discipline Project of Shanghai Municipal Education Commission (Grant No.J50104)     

Primal-dual interior-point algorithms with dynamic step-size based on kernel functions for linear programming

In this paper, primal-dual interior-point algorithm with dynamic step size is implemented for linear programming (LP) problems. The algorithms are based on a few kernel functions, including both self-regular functions and non-self-regular ones. The dynamic step size is compared with fixed step size for the algorithms in inner iteration of Newton step. Numerical tests show that the algorithms with dynamic step size are more efficient than those with fixed step size. Project supported by Dutch Organization for Scientific Research (Grant No.613.000.010)     

Polynomial-time interior-point algorithm based on a local self-concordant finite barrier function

The choice of self-concordant functions is the key to efficient algorithms for linear and quadratic convex optimizations, which provide a method with polynomial-time iterations to solve linear and quadratic convex optimization problems. The parameters of a self-concordant barrier function can be used to compute the complexity bound of the proposed algorithm. In this paper, it is proved that the finite barrier function is a local self-concordant barrier function. By deriving the local values of parameters of this barrier function, the desired complexity bound of an interior-point algorithm based on this local self-concordant function for linear optimization problem is obtained. The bound matches the best known bound for small-update methods. Project supported by the National Natural Science Foundation of China (Grant No.10771133), the Shanghai Leading Academic Discipline Project (Grant No.S30101), and the Research Foundation for the Doctoral Program of Higher Education (Grant No.200802800010)     

基于NURBS模型的自由曲线加工轨迹自适应生成方法(英文)

为实现NURBS曲面快速高精度实时差补,提出了基于修正型sigmoid函数的动力学模型,给出了最大速度、弓高误差、加工曲线的曲率半径和插补周期之间的约束条件.该模型在满足jerk、加速度、速度均连续的前提下,将常用的三次多项式S型以及三角多项式S型动力学模型的15个分段数减少至3个.在此基础上,提出采用差商代替导数的优化Adams算法,避免了常用的Taylor展开所遇到的高阶求导计算,求取了差补周期参数.最后通过减少低次零值B样条基函数的计算,对De Boor-Cox递推算法进行了简化设计,提出了精简型De Boor-Cox算法,缩减了计算量.仿真分析表明,所提算法可根据加工路径有效控制进给速度,在保证加工精度的同时,使计算量得到减少,提高了运算速度.实验结果显示本加工方法可以正确计算目标参数,并适合应用于实际加工系统.     

机动目标跟踪的TIMM算法

针对如何避免或减少由于目标的机动运动所带来的估计误差问题,提出了具有门限的相互作用多模型估计(TIMM)的新算法.该算法主要应用于雷达目标自动跟踪系统中.TIMM算法在相互作用多模型估计器(IMM)算法的基础上引入门限控制器来提高跟踪精度,该算法同样适用于其他各种改进的IMM算法.通过2个不同的例子,对由TIMM和IMM这2种算法产生的均方根误差进行比较.仿真结果表明,同IMM算法相比较,TIMM算法可减少估计误差,从而提高机动目标的跟踪性能.     

Developmental and individual differences in children's on-line representations of dynamic social events

Developmental and clinical differences in children's on-line understanding of ongoing social events were examined. Boys with (n = 38) and without (n = 41) Attention Deficit Hyperactivity Disorder (ADHD) were compared. Prior social schemes and integrative processes were hypothesized to influence children's representation of incoming social cues. Younger children and children with ADHD were expected to differ from older children and children without ADHD, respectively, in what they encoded and how they organized the encoded information. Differences in on-line representation were postulated to underlie differences in children's recall and social reasoning. Children's prior social schemes, looking time and on-line verbal protocols, recall, and social reasoning were assessed. Younger boys and ADHD boys showed less integrated on-line representations, accounting for poorer recall and reasoning.     

A cross-layer design approach on spectrum allocation and resource scheduling in cognitive PMP networks

We propose the spectrum allocation and resource scheduling algorithms in cognitive point to multipoint （PMP） networks with rapid changes of spectrum opportunities and present a media access control （MAC） protocol based on these algorithms. The objective of spectrum allocation is to make efficient use of the spectrum while maintaining the transceiver synchronization on frequency and time in the network. The objective of resource scheduling is to guarantee the quality of service （QoS） requirements of different kinds of connections and to minimize the total energy consumption in the network as well. By sensing only a small set of possible channels in each slot based on the state transition probability of each channel, our spectrum allocation algorithm achieves high spectrum efficiency in the network. The resource scheduling problem is divided into three sub problems and we derive optimal solutions to these problems by greedy algorithm and convex optimization. The simulation results show that our algorithm can make efficient use of the spectrum and the network resources at a cost of low computational complexity.