Two-Dimensional Multilabel Active Learning with an Efficient Online Adaptation Model for Image Classification

Conventional active learning dynamically constructs the training set only along the sample dimension. While this is the right strategy in binary classification, it is suboptimal for multilabel image classification. We argue that for each selected sample, only some effective labels need to be annotated while others can be inferred by exploring the label correlations. The reason is that the contributions of different labels to minimizing the classification error are different due to the inherent label correlations. To this end, we propose to select sample-label pairs, rather than only samples, to minimize a multilabel Bayesian classification error bound. We call it two-dimensional active learning because it considers both the sample dimension and the label dimension. Furthermore, as the number of training samples increases rapidly over time due to active learning, it becomes intractable for the offline learner to retrain a new model on the whole training set. So we develop an efficient online learner to adapt the existing model with the new one by minimizing their model distance under a set of multilabel constraints. The effectiveness and efficiency of the proposed method are evaluated on two benchmark data sets and a realistic image collection from a real-world image sharing Web site—Corbis.     

Micromachining of passive planar micromixer on poly (methyl methacrylate) substrate with excimer laser ablation

Excimer laser ablation technique was introduced into this work to fabricate a passive planar micromixer on the PMMA substrate. T-junction shaped and width-changed S-shaped microchannels were both designed in this micromixer to enhance mixing effect. The mixing experiment of distilled water and Rhodamine B with injection flow rate of 500 and 1,500 μm/s validates the mixing effectivity of this micromixer, and indicates the feasibility of excimer laser ablation in the microfabrication of μ-TAS device.     

关于公平交换协议中使用正交验证理论的安全性分析及改进

所谓公平交换协议是指,一个电子商务协议能保证交易双方在执行协议的过程中,任何一方通过欺诈行为、误传消息或者提前终止协议这些不公平方式,都不可能从另一方获取利益.要么任何一方都可以得到对方的信息,要么双方都得不到对方的信息.Ray等学者自2000年先后提出了一系列基于正交验证理论的公平交换协议.这类协议的重要贡献之一在于,基于正交验证理论,通过使用一对相容密钥分别对同一商品加密,使得在不泄露所购买商品的情况下,用户在付款之前能够对所购买商品的正确性进行验证,实现了使用离线TTP的优化公平交换协议.然而,通过对Ray协议进行安全性分析发现,恶意商家可以任意传送错误商品的信息也能轻易通过用户的验证,导致用户为之付款,即该验证过程是不安全的,最终导致该类协议退化为一个使用在线TTP的公平交换协议.另外,给出了改进方案,以避免上述攻击.     

深层网查询接口的自动集成

深层网数据库的访问方式主要是通过查询接口,所以查询接口是外部访问深层网数据库的门户.为了能够同时访问同一领域多个Web数据库,需要对多个Web数据库的查询接口进行集成.因此,提出基于本体的深层网查询接口集成方法.首先构建领域核心本体,在模式匹配过程中,不断完善核心本体;然后,以本体作为媒介,在不同查询接口模式间建立属性映射关系,发现属性间的语义关联;最后,根据本体概念出现的频数生成集成接口.实验表明提出的深层网查询接口自动集成方法是可行的和高效的.     

基于使用控制模型的防恶意下载系统

为了对恶意下载行为实施有效控制,研究通用访问控制框架和使用控制模型(UCON),提出基于流量的UCON。设计并实现基于UCON的防恶意下载系统,阐述其关键技术和实现方法。实验结果表明,该模型能有效控制用户下载行为,具有高安全性和较强的操作灵活性。     

求解昂贵区间多目标优化问题的高斯代理模型

本文将数据挖掘(高斯过程回归建模)和智能进化算法(GA,NSGA-Ⅱ)进行结合,用于解决优化函数未知的昂贵区间多目标优化问题.首先利用高斯过程对采用中点和不确定度表示的未知目标函数和约束函数进行建模,由于相关性和准确性是区间函数模型的两个必备条件,故提出一种融合多属性决策的双层种群筛选策略,并将其嵌入到遗传算法求解高斯模型参数的过程中,第1层根据相关性属性排除候选解集中部分劣解,第2层根据准确性属性排除候选解集中其余超出种群规模的劣解,两属性的权重系数决定两层排除劣解的比例.然后将所建模型作为优化对象的代理模型引导区间NSGA-II算法优化求解,从而获得所需的Pareto前沿.     

基于周期信号的迭代学习控制在逆变器中的应用

提出一种基于周期信号的迭代学习控制方法,并将其应用于电力逆变器,分析了跟踪误差沿迭代方向的收敛条件,给出了控制律的时域表达形式以及参数的确定方法.由于该方法理论上可实现跟踪误差收敛到零,因而使逆变器输出电压跟踪精度大幅提高.通过MATLAB/Simulink仿真,该方法的有效性得到验证;此外,与传统PID控制相比,仿真结果显示出该方法具有较好的负载适应能力以及优越的跟踪性能.     

GPU平台上面向性能和功耗的分支优化

强大的计算能力使得GPGPU在通用计算领域得到了广泛的应用。然而,GPGPU的SIMT(Single Instruction Multiple Threads)工作方式,使其执行效率受到应用中不一致分支行为(Branch Divergence)的严重影响。虽然人们提出了线程交换方法来减小分支带来的性能损失,但这种方法往往会引入额外的访存操作,不仅在一定程度上减少了线程交换优化的性能收益,还增加了功耗。首先举例说明线程交换范围对程序性能和功耗的影响;然后提出了一种减少线程交换所引入的额外访存操作的方法。实验表明,对于Reduction程序,当交换范围为256时,在性能平均损失为4%的情况下功耗降低幅度最大为7%;而对于Bitonic程序,当交换范围为256和512时,在没有功耗开销的情况下,性能分别最大提升了6.4%和5.3%。     

面向Cassandra数据库的高效动态数据管理机制

Cassandra数据库是当前通用的数据库之一,同时也被Apache列为重点发展的顶级项目。针对Cassandra分布式数据库系统而言,大量的写请求会造成过多分散的SStable结构以及过高的数据冗余度,进而造成用户读取请求响应速度下降。该问题可以通过系统自动触发的局部数据合并机制或人为干预的整体数据合并机制来解决。然而,不合时机的系统自动局部合并过程会严重降低用户正在执行的读取操作的性能,而过长时间的人为整体数据合并过程又会长时间地占用系统资源,严重制约系统的整体性能。针对此问题,提出了一种面向Cassandra数据库的动态数据管理机制。首先,实时监测系统环境,将数据按照写入时间和大小进行分层分级管理,对合并的时机、参与合并的文件及合并过程分别制定相应的执行策略;其次,通过特定优化手段减少数据的合并时间,以降低合并过程对系统性能的影响。测试结果表明,该管理机制优化了Cassandra数据库的合并过程,提升了系统整体的读取响应速度。     

Weighted principal component analysis-based service selection method for multimedia services in cloud

Cloud computing has rendered its ever-increasing advantages in flexible service provisions, which attracts the attentions from large-scale enterprise applications to small-scale smart uses. For example, more and more multimedia services are moving towards cloud to better accommodate people’s daily uses on various smart devices that support cloud, some of which are similar or equivalent in their functionality (e.g., more than 1,000 video services that share similar “video-play” functionality are present in APP Store). In this situation, it is necessary to discriminate these functional-equivalent multimedia services, based on their Quality of Service (QoS) information. However, due to the abundant information of multimedia content, dozens of QoS criteria are often needed to evaluate a multimedia service, which places a heavy burden on users’ multimedia service selection. Besides, the QoS criteria of multimedia services are usually not independent, but correlated, which cannot be accommodated very well by the traditional selection methods, e.g., traditional simple weighting methods. In view of these challenges, we put forward a multimedia service selection method based on weighted Principal Component Analysis (PCA), i.e., Weighted PCA-based Multimedia Service Selection Method (W_PCA_MSSM). The advantage of our proposal is two-fold. First, weighted PCA could reduce the number of QoS criteria for evaluation, by which the service selection process is simplified. Second, PCA could eliminate the correlations between different QoS criteria, which may bring a more accurate service selection result. Finally, the feasibility of W_PCA_MSSM is validated, by a set of experiments deployed on real-world service quality set QWS Dataset.