A data-driven kernel method assimilation technique for geophysical modelling

Incorporating the quantity and variety of observations in atmospheric and oceanographic assimilation and prediction models has become an increasingly complex task. Data assimilation allows for uneven spatial and temporal data distribution and redundancy to be addressed so that the models can ingest massive data sets. Traditional data assimilation methods introduce Kalman filters and variational approaches. This study introduces a family of algorithms, motivated by advances in machine learning. These algorithms provide an alternative approach to incorporating new observations into the analysis forecast cycle. The application of kernel methods to processing the states of a quasi-geostrophic numerical model is intended to demonstrate the feasibility of the method as a proof-of-concept. The speed, efficiency, accuracy and scalability in recovering unperturbed state trajectories establishes the viability of machine learning for data assimilation.     

基于用户QoE的流媒体系统性能测评模型设计

影响流媒体用户服务质量体验QoE(Quality of Experience)的因素有很多,如何对用户QoE进行量化判别是一个复杂的问题。为此,以媒体传输指标MDI为基础,研究了用于测量流媒体用户QoE的基本指标,并提出了相关测量数据的获取方法。以流媒体用户的实际体验质量为依据,提出了一种有效的流媒体系统性能评价模型,该模型为流媒体系统的性能分析和优化调整提供了理论基础,具有广泛的应用价值。     

一种基于QoE效用函数的资源分配算法

针对多用户MIMO-OFDM系统,立足业务体验方,给出了一种最大化用户QoE的资源分配算法。通过设计QoE效用函数,将用户QoE与系统QoS参数关联起来,在发送功率和目标误码率的约束条件下,以最大化用户平均QoE为目标,通过QoE效用函数获取用户当前时刻QoE增量,据此确定用户时频资源分配优先级,进而进行注水功率分配。仿真结果表明,该算法能够充分利用系统资源,有效提高用户平均QoE。     

非平衡数据集分类研究

现实世界中存在着非平衡数据集,即数据集中的一类样本数量远大于另一类。而少数类样本的识别通常是人们首要关心的,将少数类样本误分为多数类要比将多数类样本误分为少数类付出更高的代价。传统的机器学习算法可能会产生偏向多数类的结果,因而对于少数类而言,预测的效果会很差。在对目前国内外非平衡数据集研究现状深入分析的基础上,针对非平衡数据集数据复杂度研究和失衡解决方法研究两个方向相对孤立及缺乏系统性的缺陷,提出了一种非平衡数据集整体解决框架,以满足日益迫切的应用需求。     

一种基于重取样的代价敏感学习算法

大多数非均衡数据集的研究集中于纯重构数据集或者纯代价敏感学习,本文针对数据集类分布非均衡和不相等误分类代价往往同时发生这一事实,提出了一种以最小误分类代价为目标的基于混合重取样的代价敏感学习算法。该算法将两种不同类型解决方案有机地融合在一起,先用样本类空间重构的方法使原始数据集的两类数据达到基本均衡,然后再引入代价敏感学习算法进行分类,能提高少数类分类精度,同时有效降低总的误分类代价。实验结果验证了该算法在处理非均衡类问题时比传统算法要优越。     

基于广义线性模型的QoE评估方法研究

基于简化的广义线性模型,建立了QoE等级概率分布回归模型。该模型能够通过网络QoS参数映射得到用户QoE等级累积概率分布。采用最大似然估计完成了对模型参数的估计,并使用皮尔逊方法对模型的拟合优度进行了检验。Web业务的QoE评估实例验证了方法的有效性。     

针对动态非平衡数据集鲁棒的在线极端学习机

动态数据存在数据量动态改变,数据类别分布非平衡、不稳定等问题,这些问题成为分类的难点。针对该问题,通过对在线极端学习机模型进行拓展,提出鲁棒的权值在线极端学习机算法。为解决动态数据非平衡性,该算法借助代价敏感学习理论生成局部动态权值矩阵,从而优化分类模型产生的经验风险。同时,算法进一步考虑动态数据由于时序性质改变造成的数据分布变化,而引入遗忘因子增强分类器对数据分布变更的敏感性。算法在不同数据分布的24个非平衡动态数据集上测试,取得了较好的效果。     

基于拆分集成的不均衡数据分类算法

为改进SVM对不均衡数据的分类性能,提出一种基于拆分集成的不均衡数据分类算法,该算法对多数类样本依据类别之间的比例通过聚类划分为多个子集,各子集分别与少数类合并成多个训练子集,通过对各训练子集进行学习获得多个分类器,利用WE集成分类器方法对多个分类器进行集成,获得最终分类器,以此改进在不均衡数据下的分类性能.在UCI数据集上的实验结果表明,该算法的有效性,特别是对少数类样本的分类性能.     

基于用户体验延迟的QoE层次评估模型

针对Web服务中难以获得反映用户体验的数据,且难以对用户体验质量（QoE）进行评估的问题,在分析了Web数据和现有评价方法的基础上,提出了一种融入用户体验延迟ED的QoE层次评价方法。该方法基于层次分析法,并结合人类生理主观感受,能够更加客观地衡量用户体验。通过相关案例验证了该方法的可行性。     

具有用户体验保障的资源优化分配算法

在研究智能电视用户体验质量(Quality of Experience,QoE)量化的基础上,提出一种具有用户体验保障的资源分配模型,并针对该模型提出两种资源分配算法:RA_BAT算法和RA_GHEU算法.实验结果表明,基于回溯法的RA_BAT能够求得问题的最优解,可作为算法比较的参照系,而启发式算法RA_GHEU可在极短的运行时间内求出接近于最优的解,适合用于智能电视资源分配的实时处理.     

IP网络中视频流QoE评价模型

网络视频业务的兴起使网络运营商和服务提供商更加关注视频的用户体验（QoE）,然而视频用户体验（QoE）值由于其主观性且评价过程复杂,难以在视频流传输中实时获取。通过实验分析了视频传输过程中服务质量（QoS）参数变化对视频QoE的影响,建立了客观、可测量的QoS参数与视频QoE之间映射模型,用可量化的QoS参数来评定视频QoE受网络性能的影响程度,以评估网络视频质量,该模型形式简单,能够实时监测视频质量。实验结果表明,该模型的评价结果能较好反映视频QoE。     

结合前馈调参与迭代学习的数据驱动控制方法

在前馈控制中,需要尽可能的去除前馈控制器对系统模型的需求,同时保证高精度和鲁棒性.本文提出了一种数据驱动的将迭代前馈调参与迭代学习控制进行结合的方法,通过引入基函数参数化的前馈控制器和输入整形滤波器,使用梯度下降法求解最优系统前馈控制器,消除期望轨迹引入的扰动;通过迭代学习控制,消除系统重复性扰动,进一步提高控制精度.算法具有不依赖系统模型,高精度,适用于变轨迹任务的优点.文中给出了相应的仿真,并应用到一个直线电机系统,通过实验验证了算法的有效性.     

Subjective evaluation of critical success factors for a QoE aware adaptive system

For many years video content delivery has established itself as the killer application. Improving QoE on adaptive streaming is focusing many efforts in the quest for optimized methods and metrics to allow a QoE driven adaptation. Questions such as whether adaptive systems based on Scalable Video Coding improve subjective quality and in which situations or to what degree are still open issues. Tolerance and indifference thresholds for each type of content, conditions or viewer category, with regard to adaptive systems are critical success factors that are yet unresolved. We compare the performance of a complete adaptive system with the traditional, i.e. non-adaptive, approach in subjective terms. Results of surveying 75 participants show that the adaptation improves QoE under most of the evaluated conditions. Tolerance thresholds for triggering adaptation events have been identified. Users accustomed to Internet video are more critical than users that only watch TV. The under 35 year old subset among the available population is generally more satisfied with the adaptive system than the older subset.     

机器生成语言的质量评价方法综述

生成语言的质量评价很大程度上影响着自然语言生成的研究,已成为制约该领域发展的瓶颈问题。通过对机器翻译、自动文摘、对话系统、图像标题生成和机器写作等广义自然语言生成任务的语言质量评价方法的汇总,介绍了人工评价和自动评价的特点、优缺点和开放评价资源,分析了不同任务的不同评价角度和适用面。不同评价方法的对比分析,可为方法融合和关键问题的探索提供借鉴。整体上机器生成语言质量评价还局限于语言形式的比较,在语义表达的准确性、衔接连贯性等深层评价上存在诸多挑战。结合评价难点问题和现有研究的推进情况,分析了生成语言质量评价的研究趋势。     

HTTP移动流媒体QoE管理综述

随着移动通信网络和技术的快速发展,多媒体信息的大量涌现,人们对多媒体信息服务的要求也越来越高。基于HTTP协议的移动流媒体由于拥有广泛运营基础的Web服务器网络环境,部署简单,适用范围广泛等优点成为研究的热点。简要介绍了HTTP移动流媒体的相关特点,重点分析了影响HTTP移动流媒体用户体验质量的各层因素,归纳总结了评价方法和现有的工具,并对 HTTP 移动流媒体相应的优化进行了介绍,最后对HTTP移动流媒体QoE管理进行了总结和展望。     

面向大规模特征选择的自监督数据驱动粒子群优化算法

大规模特征选择问题的求解通常面临两大挑战:一是真实标签不足,难以引导算法进行特征选择;二是搜索空间规模大,难以搜索到满意的高质量解。为此,提出了新型的面向大规模特征选择的自监督数据驱动粒子群优化算法。第一,提出了自监督数据驱动特征选择的新型算法框架,可不依赖于真实标签进行特征选择。第二,提出了基于离散区域编码的搜索策略,帮助算法在大规模搜索空间中找到更优解。第三,基于上述的框架和方法,提出了自监督数据驱动粒子群优化算法,实现对问题的求解。在大规模特征数据集上的实验结果显示,提出的算法与主流有监督算法表现相当,并比前沿无监督算法具有更高的特征选择效率。     

Simultaneous hysteresis-dynamics compensation in high-speed,large-range trajectory tracking: A data-driven iterative control

In this article, a data-driven difference-inversion-based iterative control (DDD-IIC) approach is proposed to compensate for both nonlinear hysteresis and dynamics of Hammerstein systems. Simultaneous hysteresis-dynamics compensation is needed in control of Hammerstein systems such as smart actuators, where effects of hysteresis and dynamics coexist and become pronounced in high-speed, large-range output tracking. Challenges, however, arises as hysteresis modeling, as needed in many existing control methods, can be complicated and prone to uncertainties, and the hysteresis and the dynamics are coupled and tend to change due to the variations of the system conditions (e.g., the aging of smart actuators). The proposed DDD-IIC technique aims to achieve simultaneous hysteresis-dynamics compensation with no need for modeling hysteresis and/or dynamics, and with both precision tracking and good robustness against hysteresis/dynamics variations. The convergence of the DDD-IIC algorithm in the presence of random output disturbance/noise is analyzed. It is shown that when the noise is negligible, exact tracking is achieved and the size of hysteresis accounted is given by the Golden ratio. The proposed DDD-IIC method is demonstrated via experiments of high-speed large-range output tracking on two different types of smart actuators with symmetric and asymmetric hysteresis behavior, respectively.     

基于核空间中K-近邻的不均衡数据算法

*The Scientific Research Program Funded by Education Department of Shaanxi Province under Grant No.12JK0748(陕西省教育厅科技计划项目);the Science and Technology Research Project of Shangluo University under Grant No.13sky024(商洛学院科学与技术研究项目).
　　法插入样本;最后在新的训练集上确定最终决策函数。在人工数据集和4组UCI数据集上进行了实验,结果表明了该算法对不均衡数据集进行降维采样的有效性。     

Developing a Quality of Experience (QoE) model for Web Applications

Web‐based service providers have long been required to deliver high quality services in accordance with standards and customer requirements. Increasingly, however, providers are required to think beyond service quality and develop a deeper understanding of their customers' Quality of Experience (QoE). Although models exist that assess the QoE of Web Application, significant challenges remain in: (1) Defining QoE factors from a Web engineering perspective, (2) quantifying the relationship between so‐called “objective” and “subjective” factors of relevance, and (3) dealing with limited data available in relation to subjective factors. In response, the work here presents a novel model (and associated software instantiation) that integrates factors through Key Performance Indicators and Key Quality Indicators. The mapping is incorporated into a correlation model that assesses the QoE of Web Applications, with a consideration of defining the factors in term of quality requirements derived from web architecture. The data resulting from the mapping is used as input of the proposed model to develop artefacts that quantify and predict QoE using Machine Learning. The development of proposed model is framed and guided by Design Science Research approach with the purpose of enabling providers to more informed decisions regarding QoE and/or to optimise resources accordingly. Although the work is oriented at developing an artefact that has clear utility for practice, the nascent design theory underpinning the work is developed and discussed.