一种基于Spark的不确定数据集频繁模式挖掘算法

如何在海量不确定数据集中提高频繁模式挖掘性能是目前研究的热点.传统算法大多是以期望、概率或者权重等单一指标为数据项集支持度，在大数据背景下，同时考虑概率和权重支持度的算法难以兼顾其执行效率.为此，本文提出一种基于Spark的不确定数据集频繁模式挖掘算法（UWEFP），首先，为了同时兼顾数据项的概率和权重，计算一项集的最大概率权重值并进行剪枝；然后，为了减少对数据集的多次扫描，结合Spark框架的优点，设计了一种具有FP-tree特征的新颖的UWEFP-tree结构进行模式树的构建及挖掘；最后在Spark环境下，以UCI数据集进行实验验证.实验结果表明本文的方法在保证挖掘结果的同时，提高了效率.     

流媒体技术及应用

主要介绍了流媒体技术、原理及其应用，详述了作为流媒体实现的关键技术-流式传输和传输的过程，以及目前支持流媒体传输的协议和使用较多的流式视频格式。     

Formal Analysis of Streaming Downloading Protocol for System Upgrading

For a PC-mobile download system which is embedded with streaming download protocol, there are problems that the data cannot be transmitted correctly from the PC to the mobile, or the transmission is unacceptably slow. To solve these problems, we carry out a formal analysis for the protocol with some timing parameters and a given probability of message loss and unordered data using a probabilistic model checking tool PRISM. We introduce a technique to reduce the state space of the system modeling the protocol which is a network of probabilistic timed automata. The experimental results in PRISM give us a clear explanation to the problems, and are helpful in identifying the optimal parameter settings to meet industrial requirements.     

基于媒体用户访问行为偏好模型的代理缓存算法

目前,代理缓存技术广泛应用于改善流媒体传输的服务质量.文章从实际用户日志文件的分析出发,利用发现的用户浏览流媒体对象时的行为分布模型,提出了一种新的视频流媒体缓存算法.仿真结果证明,该算法可以通过记录很少的用户访问信息获取较高的性能表现.     

Streaming potential measurements on stainless steels surfaces: evidence of a gel-like layer at the steel/electrolyte interface

Electric charges at the surface of a passive stainless steel are generally considered as concentrated either in the passive film itself, or at the metal/passive film interface, or in the electrical double layer at the film solution interface. Rest potential time dependence after immersion of a passive surface in aqueous electrolytes suggests however that slow processes occur in the onset of the surface charge. Specific experiments, such as streaming potential measurements and electrochemical impedance spectroscopy in a thin electrolyte cell, were carried out for understanding better this phenomenon. An AISI 304 type austenitic stainless steel with polished or bright annealed surface finishes was immersed in NaCl aqueous solutions with various pH and chloride concentrations. The streaming potential time evolution shows two steps: a first rapid one (∼2 min) is attributed to the onset of the surface charge. The second step is much slower (approximately several hours) and possibly due to an interphase layer between the passive film and the solution. Following this idea, the whole kinetics is controlled by cation migration across the interphase when the pH is larger than the isoelectric pH (pH_iep), while chloride ions are incorporated in the interphase when pH < pH_iep. Impedance measurements allow determining both the kinetics of charge transport and the thin cell conductivity. When glass is used as reference material for the cell walls instead of stainless steel, the Nyquist plots show a high-frequency response. For stainless steel cell walls, a low-frequency response is observed, attributed to a slow charge reorganisation inside the interphase layer. The charge distribution at metal/electrolyte interface is discussed in terms of a gel-like layer which possibly takes place at the passive film/electrolyte interface.     

PRAND: GPU accelerated parallel random number generation library: Using most reliable algorithms and applying parallelism of modern GPUs and CPUs

The library PRAND for pseudorandom number generation for modern CPUs and GPUs is presented. It contains both single-threaded and multi-threaded realizations of a number of modern and most reliable generators recently proposed and studied in Barash (2011), Matsumoto and Tishimura (1998), L’Ecuyer (1999,1999), Barash and Shchur (2006) and the efficient SIMD realizations proposed in Barash and Shchur (2011). One of the useful features for using PRAND in parallel simulations is the ability to initialize up to 10¹⁹${10}^{19}$ independent streams. Using massive parallelism of modern GPUs and SIMD parallelism of modern CPUs substantially improves performance of the generators.     

Granulation of a hardmetal powder for spark plasma sintering

A fluidized bed granulation method, pressure swing granulation (PSG), was applied to granulation of a hardmetal powder without pressing lubricants for making the upstream process of spark plasma sintering (SPS) more efficient.

The properties of the granules were examined and compared with those of spray dried granules and extruded ones under the present system using a sieve.

Spherical granules between 0.15 and 0.84 mm in diameter difficult to obtain by the spray drying were obtained with high yield. The flowability of the granules was far better than that of spray dried granules and similar to that of extruded ones. Iron contamination and oxidization during pressure swing granulation were tolerable to the real production.     

基于IPTV的网络教学系统的设计

以IPTV技术为支持的远程教育模式具有投入少且效果好的特点,在远程教育中发挥着越来越重要的作用.利用P2P和流媒体技术为支撑,设计和实现了具有IPTV交互特点的远程教育系统.该系统包括课后练习,测试以及答疑等学习活动,实现集在线学习,游戏娱乐,生活服务于一体的互动数字化教育,并有效解决实时课堂传输中视频质量的问题.     

一种基于深度强化学习的Spark Streaming参数优化方法

Spark Streaming作为主流的开源分布式流分析框架,性能优化是目前的研究热点之一。在Spark Streaming性能优化中,业务场景下的配置参数优化是其性能提升的重要因素。在Spark Streaming系统中,可配置的参数有200多个,对参数调优人员的经验要求较高,未经优化的参数配置会影响流作业执行性能。因此,针对Spark Streaming的参数配置优化问题,提出一种基于深度强化学习的Spark Streaming参数优化方法（DQN-SSPO）,将Spark Streaming参数优化配置问题转化为深度强化学习模型训练中的最大回报获得问题,并提出权重状态空间转移方法来增加模型训练获得高反馈奖励的概率。在3种典型的流分析任务上进行实验,结果表明经参数优化后Spark Streaming上的流作业性能在总调度时间上平均缩减27.93%,在总处理时间上平均缩减42%。     

基于国产处理器增量式实时交通流预测算法及实现

针对城市交通难以处理大量数据且实时性差等问题,提出了根据增量式城市交通流数据预测拥堵情况的一种基于国产处理器的L-BFGS(limited-memory BFGS)算法。该算法通过存储向量序列计算Hessian矩阵,改进Two-Loop算法求下降方向,在Spark集群中并行处理时收敛速度快,适用于实时性要求强的城市交通场景。实验结果证明,L-BFGS预测算法完全可以在国产平台上对大规模的实时交通数据流进行快速建模、预测,在改善城市交通管理水平提供有效支撑的同时也丰富了国产芯片的应用领域。