基于高斯密度图估计的自然场景汉字检测

针对自然场景下中文小文本难以定位的问题,提出了基于高斯密度图估计的并行深度网络对自然场景汉字进行检测。首先将中文数据集中的汉字位置信息转换为高斯文字密度图;其次引入一种多级并行连接结构,提高网络细节信息捕捉能力;最后再融合网络中的上采样特征信息得到高精度文字密度图,最终实现对文字区域的定位。在中文数据集CTW(Chinese text in the wild)上进行了实验,实验结果表明提出方法准确率和召回率均有较大提升,证明了该方法的可行性和准确性。     

Spark平台上利用网络加权Voronoi图的分散迭代社区聚类并行化研究

针对当下数据大规模增长对计算能力需求的急剧增长,传统独立运行的机器在大规模网络社区中执行社区检测操作时无法提供所需的数据处理能力的问题,提出一种网络加权Voronoi图的并行分散迭代社区聚类法(NWVD-PDICCM)。利用基于网络加权Voronoi图的分散迭代社区聚类方法(NWVD-DICCM)提取大型网络的有效社区结构。结合并行聚类方法,将DICCM方法的操作从串行过程转换为并行计算。利用执行并行社区聚类时的图分区,通过最小化从属工作者之间的通信来加速该过程。仿真实验结果表明,NWVD-PDICCM可以与一系列计算机架构平台共同运行,并且实现基于Spark平台的并行操作,相比其他几种较新的方法,在大规模网络数据处理能力方面得到显著提升。     

控制传布图到结构图的转换算法

控制传布图是介于数据流图和结构图之间的图形工具,是实现从数据流图到结构图转换的中间产物。本文详细讨论了如何从控制传布图转换为结构图,给出了转换算法,从而,该算法与文[4]所提出的控制传布图算法一起完成了数据流图到初始结构图的半自动转换。     

面向机器翻译的句类依存树库构建及应用

该文以汉英机器翻译为应用目标,以概念层次网络理论的语义网络和句类分析方法为理论基础,探讨了句类依存树库构建的理论和标注实践等问题,描述了构建树库所需的概念类别标注集和句类关系标注集。并通过与已有汉语树库进行对比,以汉语显性轻动词句的标注为例,分析了汉语句类依存树库的特点。该文在应用层面定义了面向汉英机器翻译的融句法语义信息于一体的“句类依存子树到串”双语转换模板,尝试基于汉语句类依存树库提取汉英转换模板。     

单向k-元n-立方体网络

单向[k]-元[n]-立方体是指具有单向边的[k]-元[n]-立方体互连网络拓扑。当网络包含的顶点数目较大时,比起传统的双向[k]-元[n]-立方体,单向?[k]-元[n]-立方体对通信硬件复杂性的要求更低一些。提出了[k]-元[n]-立方体的一个定向,使得定向后的单向[k]-元[n]-立方体[UQkn]有一些良好的性质。证明了[UQkn]是正则的,极大弧连通的,具有迭代结构的且[UQkn]的直径是小的。此外,提出了一个简单的多项式时间路由算法。     

洗牌交换置换网络SEP的一种新的路由算法

1引言与sEP网络的定义 众所周知,Cayley图和Cayley陪集图在计算机互连网络的设计与分析中起着重要的作用[1～3].例如:熟知的环(ring)网络,圆环面(torus)网络,超圆环面(super-torus)网络[7],星图(star graph)网络,超立方体网络(hypercube),立方体连接圈(cubeconnected cycles)网络[2]均可看作是Cayley图.而de Bruijn网络与洗牌交换网络[8]可作为Cayley陪集图的例子.     

计算循环群上4度Bi-Cayley网络直径的快速算法

Cayley图是一类高对称正则图,有许多好性质,被广泛认为是一类理想的互连网络拓扑结构。Bi-Cayley图是Cayley图的一个自然推广,特别地,循环群上4度Bi-Cayley网络[BC(n;±s1,±s2)]是双环网络[DLG(n;±s1,±s2)]的一个自然推广。讨论了循环群[?n]上4度Bi-Cayley网络[BC(n;±s1,±s2)]连通的充分必要条件,并给出了计算该网络直径的一种算法,其时间复杂度为[O(lb n)]。     

基于深度监督隐空间构建的语义分割改进方法

现有卷积操作在语义分割任务中难以有效捕捉长距离区域间的关系,导致分割结果不符合人类常识。为此,提出一种基于深度监督隐空间构建的语义分割改进方法。采用“特征图-隐空间-特征图”流程,将图像空间的像素特征转换为隐空间中的节点特征,将区域之间的位置和语义关系转换为节点之间的连接权重,实现了从特征图到隐空间的特征转换。在隐空间构建过程中,使用Kullback-Leibler散度损失函数监督投影矩阵,以避免从特征图到隐空间节点的转换过程中丢失特征;使用InfoNCE损失函数监督节点特征表征与真实标签表征,使得图像特征与标签保持一致。该方法在构建的隐空间上使用图神经网络进行语义推理,学习节点之间的关系,赋予模型学习区域间语义关系的能力,从而改善分割结果中的反常识现象。在公开数据集CityScapes上的实验结果表明,相比基线分割网络,该方法的平均交并比（mIoU）为81.1%,相较于基线分割网络mIoU提升2.6个百分点,能有效提升分割结果。     

边故障[k]元[n]立方体的超级哈密顿交织性

[k]元[n]立方体（记为[Qkn]）是优于超立方体的可进行高效信息传输的互连网络之一。[Qkn]是一个二部图当且仅当[k]为偶数。令[G[V0,V1]]是一个二部图,若（1）任意一对分别在不同部的顶点之间存在一条哈密顿路,且（2）对于任意一点[v∈Vi],其中[i∈{0,1}],[V1-i]中任意一对顶点可以被[G[V0,V1]-v]中的一条哈密顿路相连,则图[G[V0,V1]]被称为是超级哈密顿交织的。因为网络中的元件发生故障是不可避免的,所以研究网络的容错性就尤为重要。针对含有边故障的[Qkn],其中[k4]是偶数且[n2],证明了当其故障边数至多为[2n-3]时,该故障[Qkn]是超级哈密顿交织图,且故障边数目的上界[2n-3]是最优的。     

基于样本的线条画风格转换与定制方法的研究

提出了一种利用线条画作为样本对任意线条画进行风格转换和定制的方法,该方法不仅可以将样本图风格转移到目标图,还可以方便地控制目标图与样本图之间风格的相似程度,该方法主要采用基于细分笔划的风格相似性比较算法,可在多层次上把目标图转换为样本图风格,不仅如此,还通过定义风格的插值操作实现对风格的定制,与已有方法相比,文中方法只需少量的样本笔划即可进行风格转换,并且通过风格定制,使风格转换具较强的交互可控性,文章最后通过实例证明了方法的有效性。     

扩展词组数最小法的假名汉字转换

本文描述了改进后的“词组最小法”、并提出了新算法。它被名为“扩展词组最小法”。重新定义了句子中词组的计算方法。为了实现此目标,从始读到句子假名的全部读入,将词库查询及语法检查的结果以“树”型数据加以保留。采用上述算法后,以假名文字为单位的变换率可达95. 8 %;以词组为单位的变换率可达88. 9 %。     

Bridging the Information Transfer Gap: Measuring Goodness of Information Fit

The ability to determine, a priori, how well an interface design meets the information needs of a user—that is, to ‘bridge the information transfer gap’—is a critical component of any intelligent interface design system. Yet most current approaches either sidestep the problem of computationally measuring ‘goodness of fit’ by compiling design knowledge into a set of pattern matching rules or, at best, use a very limited set of parameters in scoring. We describe a formal representation for capturing the information needs associated with user tasks and the information conveying capabilities of interface elements. A multifaceted scoring technique, using this representation, is presented for assessing a design's goodness of fit against current user requirements. The hallmarks of this approach are the explicit representation of the goals of good user interface design, evaluation of a proposed design against these goals and a computational approach to performing tradeoffs among these goals when incompatible. Finally, we report on the implementation and user acceptance of this design in an adaptive interface application for military attack/scout helicopters.     

藏文音节拼写检查的CNN模型

藏文音节拼写检查是藏语自然语言处理的基本任务,在藏文文字处理、文字识别、文本生成等领域具有广泛的应用。该文首先针对藏文音节的结构提出了音节向量化的方法,即音节矩阵。然后构建了适合于藏文音节拼写检查的CNN模型,使用1 364 880个藏文音节进行训练。最后对68 244个藏文音节进行测试。实验结果显示,藏文音节拼写检查CNN模型的结果优于规则、RNN和LSTM等模型,不仅对符合藏文文法的音节能正确识别外,而且对梵音藏文音节也能有效识别,正确率、召回率以及F值分别为99.52%、99.30%和99.41%。     

Denotation and connotation in the human – computer interface: The ‘Save as …’ command

This paper presents a semiotic technique as a means of exploring meaning and understanding in interface design and use. This is examined through a study of the interaction between the ‘file’ metaphor and ‘save as’ command metaphor. The behaviour of these (from a functional or computational basis) do not exactly match, or map onto, the meaning of the metaphor. We examine both the denotation of a term to the user, i.e. its literal meaning to that person, and the term's connotations, i.e. any other meanings associated with the term. We suggest that the technique applied is useful in predicting future problems with understanding the use of metaphor at the interface and with designing appropriate signification for human-computer interaction. Variation in connotation was expected but a more fundamental difference in denotation was also uncovered. Moreover, the results clearly demonstrate that consistency in the denotation of a term is critical in achieving a good user understanding of the command.     

Query generation for retrieving data from distributed semistructured documents using a metadata interface

We describe a method for generating queries for retrieving data from distributed heterogeneous semistructured documents, and its implementation in the metadata interface DDXMI (distributed document XML metadata interchange). The proposed system generates local queries appropriate to local schemas from a user query over the global schema. The system constructs mappings between global schema and local schemas (extracted from local documents if not given), path substitution, and node identification for resolving the heterogeneity among nodes with the same label that often exist in semistructured data. The system uses Quilt as its XML query language. An experiment is reported over three local semistructured documents: ‘thesis’, ‘reports’, and ‘journal’ documents with ‘article’ global schema. The prototype was developed under Windows system with Java and JavaCC.     

Case study: Using its style tools to implement IBM's CUA-2 user interface style

In an empirical case study of software tools, two participants used the ITS style designer's language to implement a general purpose, executable, rule-based user interface style. This language allows style designers to select, combine and modify rule prototypes in order to construct a rule-based user interface style. The participants implemented the entire IBM CUA-2 user interface style, plus nine additional human-computer interaction techniques, in 5–7 person-weeks. This is impressive productivity. Typically the time to complete a single CUA-2 application is measured in person-years, not person-weeks. The style implemented here is reusable by any ITS-implemented application. The achievement reported here shows that about half the work of all ITS-implemented CUA-2 applications has been completed in a few person-weeks. This result demonstrates the power and productivity of the ITS approach and tools. The results of this case study showed that key ‘ready-mades’ (e.g. named attribute groups) provided in the style designer's toolkit could be used intact, which is important for importing or exporting interaction techniques from one style to another style. The results generalize to other computer-literate designers who may want to use these tools to create other user interface styles.     

Predicting user preferences of environment design: a perceptual mechanism of user interface customisation

It is a well-known fact that users vary in their preferences and needs. Therefore, it is very crucial to provide the customisation or personalisation for users in certain usage conditions that are more associated with their preferences. With the current limitation in adopting perceptual processing into user interface personalisation, we introduced the possibility of inferring interface design preferences from the user’s eye-movement behaviour. We firstly captured the user’s preferences of graphic design elements using an eye-tracker. Then we diagnosed these preferences towards the region of interests to build a prediction model for interface customisation. The prediction models from eye-movement behaviour showed a high potential for predicting users’ preferences of interface design based on the paralleled relation between their fixation and saccadic movement. This mechanism provides a novel way of user interface design customisation and opens the door for new research in the areas of human–computer interaction and decision-making.     

Expressive line selection by example

An important problem in computer generated line drawing is determining which set of lines produces a representation that is in agreement with a user’s communication goals. We describe a method that enables a user to intuitively specify which types of lines should appear in rendered images. Our method employs conventional silhouette-edge and other feature-line extraction algorithms to derive a set of candidate lines, and integrates machine learning into a user-directed line removal process using a sketching metaphor. The method features a simple and intuitive user interface that provides interactive control over the resulting line selection criteria and can be easily adapted to work in conjunction with existing line detection and rendering algorithms. Much of the method’s power comes from its ability to learn the relationships between numerous geometric attributes that define a line style. Once learned, a user’s style and intent can be passed from object to object as well as from view to view.     

Training data selection with user’s physical characteristics data for acceleration-based activity modeling

This paper proposes an activity recognition method that models an end user’s activities without using any labeled/unlabeled acceleration sensor data obtained from the user. Our method employs information about the end user’s physical characteristics such as height and gender to find and select appropriate training data obtained from other users in advance. Then, we model the end user’s activities by using the selected labeled sensor data. Therefore, our method does not require the end user to collect and label her training sensor data. In this paper, we propose and test two methods for finding appropriate training data by using information about the end user’s physical characteristics. Moreover, our recognition method improves the recognition performance without the need for any effort by the end user because the method automatically adapts the activity models to the end user when it recognizes her unlabeled sensor data. We confirmed the effectiveness of our method by using 100 h of sensor data obtained from 40 participants.