面向目标用户的深度学习模型可视化综述

深度学习模型在某些场景的实际应用中要求其具备一定的可解释性, 而视觉是人类认识周围世界的基本工具, 可视化技术能够将模型训练过程从不可见的黑盒状态转换为可交互分析的视觉过程, 从而有效提高模型的可信性和可解释度. 目前, 国内外相关领域缺少有关深度学习模型可视化工具的综述, 也缺乏对不同用户实际需求的研究和使用体验的评估. 因此, 本文通过调研近年来学术界模型可解释性和可视化相关文献, 总结可视化工具在不同领域的应用现状, 提出面向目标用户的可视化工具分类方法及依据, 对每一类工具从可视化内容、计算成本等方面进行介绍和对比, 以便不同用户选取与部署合适的工具. 最后在此基础上讨论可视化领域存在的问题并加以展望.     

网络拓扑可视化类库TopVisLibrary的设计与实现

以信息可视化参考模型为基础,分析了网络拓扑可视化过程中的主要操作及数据形态变化,并以此为依据,建立了网络拓扑可视化对象模型,设计实现了网络拓扑可视化类库TopVisLibrary.该库的编程模型和数据接口简单清晰,有助于灵活搭建不同的网络拓扑可视化应用,同时对其它可视化工具的设计实现有一定的借鉴意义.此外,网络拓扑可视化类库的设计方法也可为其它库的设计提供参考.     

多源网络安全数据时序可视分析方法研究

近年来网络安全数据时序可视化是网络安全的热门话题,它主要是通过提供科学的数据可视化工具,提高网络安全,提升工作人员维护计算机的能力,进而将问题的发现、问题的识别、问题的掌握,全都一手掌握.网络安全可视化同新时代的一个新兴交叉研究,该研究目前已经在网络监控、检查异常等方面取得了不小的进展.文本主要是为了应对多源网络安全数据时序可视分析需求,进行了统一的方法研究,通过实验论证证明了网络安全数据可视分析方法的有效性.     

Interactive simulation and visualization

Most researchers who perform data analysis and visualization do so only after everything else is finished, which often means that they don't discover errors invalidating the results of their simulation until post-processing. A better approach would be to improve the integration of simulation and visualization into the entire process so that they can make adjustments along the way. This approach, called computational steering, is the capacity to control all aspects of the computational science pipeline. Recently, several tools and environments for computational steering have begun to emerge. These tools range from those that modify an application's performance characteristics (either by automated means or by user interaction) to those that modify the underlying computational application. A refined problem-solving environment should facilitate everything from algorithm development to application steering. The authors discuss some tools that provide a mechanism to integrate modeling, simulation, data analysis and visualization     

A survey of network anomaly visualization

Network anomaly analysis is an emerging subtopic of network security. Network anomaly refers to the unusual behavior of network devices or suspicious network status. A number of intelligent visual tools are developed to enhance the ability of network security analysts in understanding the original data, ultimately solving network security problems. This paper surveys current progress and trends in network anomaly visualization. By providing an overview of network anomaly data, visualization tasks, and applications, we further elaborate on existing methods to depict various data features of network alerts, anomalous traffic, and attack patterns data. Directions for future studies are outlined at the end of this paper.     

An atmospheric visual analysis and exploration system

Meteorological research involves the analysis of multi-field, multi-scale, and multi-source data sets. In order to better understand these data sets, models and measurements at different resolutions must be analyzed. Unfortunately, traditional atmospheric visualization systems only provide tools to view a limited number of variables and small segments of the data. These tools are often restricted to two-dimensional contour or vector plots or three-dimensional isosurfaces. The meteorologist must mentally synthesize the data from multiple plots to glean the information needed to produce a coherent picture of the weather phenomenon of interest. In order to provide better tools to meteorologists and reduce system limitations, we have designed an integrated atmospheric visual analysis and exploration system for interactive analysis of weather data sets. Our system allows for the integrated visualization of 1D, 2D, and 3D atmospheric data sets in common meteorological grid structures and utilizes a variety of rendering techniques. These tools provide meteorologists with new abilities to analyze their data and answer questions on regions of interest, ranging from physics-based atmospheric rendering to illustrative rendering containing particles and glyphs. In this paper, we will discuss the use and performance of our visual analysis for two important meteorological applications. The first application is warm rain formation in small cumulus clouds. Here, our three-dimensional, interactive visualization of modeled drop trajectories within spatially correlated fields from a cloud simulation has provided researchers with new insight. Our second application is improving and validating severe storm models, specifically the Weather Research and Forecasting (WRF) model. This is done through correlative visualization of WRF model and experimental Doppler storm data.     

大规模网络安全数据协同可视分析方法研究

网络安全可视化是近年来网络安全研究的热点,它通过提供有效的信息可视化工具,提升网络安全分析师在解决网络安全问题过程中的感知和认知能力,从而发现模式、识别异常和掌握趋势。为了应对大规模、多数据源的网络安全协同可视分析需求,研究了基于统一格式的事件元组和统计元组的数据融合模型,并提出了擅长事件关联分析的雷达图和擅长统计时序对比分析的对比堆叠流图的设计方法。最后使用该原型系统对2013国际可视分析挑战赛（visual analytics science and technology challenge,VAST Challenge）中网络安全数据可视分析竞赛提供的数据集进行了分析,通过实验和讨论验证了该网络安全数据协同可视分析方法的有效性。     

Software Architecture Visualization: An Evaluation Framework and Its Application

In order to characterize and improve software architecture visualization practice, the paper derives and constructs a qualitative framework, with seven key areas and 31 features, for the assessment of software architecture visualization tools. The framework is derived by the application of the Goal Question Metric paradigm to information obtained from a literature survey and addresses a number of stakeholder issues. The evaluation is performed from multiple stakeholder perspectives and in various architectural contexts. Stakeholders can apply the framework to determine if a particular software architecture visualization tool is appropriate to a given task. The framework is applied in the evaluation of a collection of six software architecture visualization tools. The framework may also be used as a design template for a comprehensive software architecture visualization tool.     

An experiment with affirm and HDM

Comparing the usefulness of methodologies for software development can be especially difficult when the services offered are based on different philosophies. Two systems, Affirm and HDM, were compared for their application to operation system security analysis. The assessment technique was to specify and analyze for security flaws on both systems a miniature example of a security kernel. The specification languages are at the opposite poles of the range from algebraic axioms to transition specifications. The types of security properties that could be verified with the tools available were access policy invariants and information flows. One theorem prover was highly interactive and the other nearly automatic. We found that the example could be specified satisfactorily and recognizably on both systems with a comparable amount of effort. The security analyses, on the other hand, led to very different verification tasks and different results. The two results were complementary rather than contradictory, and some additional experimentation, guided by theoretical suspicions, showed the exact relationship between them.     

水务海洋核心机房可视化安全管理研究与实践

本文以水务海洋核心机房为样板,分析了当前机房管理现状,提出了政府部门核心机房可视化安全管理模式和技术实现方法的研究目标,以机房可视化安全管理平台为载体,介绍了机房可视化安全管理的总体架构,详细描述了机房物理环境、IT设施设备、网络安全的可视化安全管理设计思想与应用效果展现。该实践是水务海洋信息化进程中充分利用可视化展现方式服务安全管理的积极、有益探索。     

A Time Model for Time-Varying Visualization

The analysis of unsteady phenomena is an important topic for scientific visualization. Several time-dependent visualization techniques exist, as well as solutions for dealing with the enormous size of time-varying data in interactive visualization. Many current visualization toolkits support displaying time-varying data sets. However, for the interactive exploration of time-varying data in scientific visualization, no common time model that describes the temporal properties which occur in the visualization process has been established. In this work, we propose a general time model which classifies the time frames of simulation phenomena and the connections between different time scales in the analysis process. This model is designed for intuitive interaction with time in visualization applications for the domain expert as well as for the developer of visualization tools. We demonstrate the benefits of our model by applying it to two use cases with different temporal properties.     

Multi-view visualizations for emergency communities of volunteers

Information and communication technologies might empower emergency communities of volunteers by assisting community participation and improving their capacity to respond to unexpected events. However, designing technology for such purpose places unique visualization challenges that go beyond the current state of research on public participation tools and related technologies. Empowering these communities requires developing representations that enable collaborative reflection, promote mutual visibility of volunteers’ efforts and sustain a shared view of the community. Similarly, it is necessary to envision visualizations that facilitate sense making of large, simultaneous and distributed pieces of heterogeneous information with different levels of credibility and priority. Accordingly, this paper identifies and characterizes these challenges to propose a multi-view and multi-abstraction-level visualization approach for emergency communities of volunteers. In particular, it combines time-oriented visualizations, space-filling visualization techniques, interaction mechanisms and coordinated maps to support community participation as well as collaborative and individual sense making. The application of these visualization techniques is discussed through the development of a set of design prototypes.     

A unified approach to adapt scientific visualization systems to third-party solvers

This paper presents a new unified approach to adapt scientific visualization systems to third-party solvers implemented on different software and hardware platforms. This approach allows building multiplatform visualization systems, enables automatic conversion of input and output data from any solver into a rendering-compatible format, and provides real-time generation of high-quality images. The automated adaptation of visualization systems to third-party solvers is based on ontological engineering methods. Multiplatform portability is provided by the automatic generation of a graphical user interface (GUI) for each particular operating system and by preprocessing the data to be rendered by using heuristic-based tools, which ensures compatibility with different hardware and software platforms, including desktop computers and mobile devices. In addition, an original anti-aliasing algorithm is proposed to ensure high quality of resulting images. Based on the proposed approach, a multiplatform scientific visualization system called SciVi is developed, which is successfully used for solving various real-world scientific visualization problems from different application domains.     

GIS可视化在跨区电网中的应用

阐述GIS(地理信息系统)可视化技术在跨区电网中的应用。首先分析传统技术在跨区电网可视化过程中的不足,并给出满足跨区电网特定需求的可视化方法和交互设计,主要包括分级过滤、多视角显示、多回路显示等技术。进而阐述基于上述方法的可视化工具的设计,重点考虑工具的开放性、性能及二次开发能力。最后研究基于可视化工具的应用构建。     

Visualization systems on the Information Technology-Based Laboratory

The Center for Promotion of Computational Science and Engineering has developed a detailed design and prototype of the Information Technology Based Laboratory (ITBL). A software group forms the basic ITBL framework and makes up this infrastructure. The infrastructure gives registered users access to any supercomputer within the ITBL for the purpose of numerical simulations and supplies them with tools for discussion sessions among ITBL user groups or communities. When a result is speculated, it's important for every member of the community to see, at the same time, images of the simulation results. To achieve this, we developed visualization systems on the ITBL infrastructure software equipped with efficient remote visualization and collaborative visualization functions. We based the two core visualization systems-the parallel tracking steering (Patras)/ITBL and the application visualization system (AVS)/ITBL-on the existing visualization tools Patras and AVS/Express.     

A Framework for Unified Network Security Management: Identifying and Tracking Security Threats on Converged Networks

A comprehensive network security management system must coordinate detection and scanning tools for converged networks; derive fully-integrated attack and network models; perform vulnerability and multi-stage attack analysis; support large-scale attack visualization; and possibly orchestrate strategic responses to unwarranted actions that cross network boundaries. We present an architecture that embodies these principles. The unified network security management system described in this paper gleans data from a suite of detection tools for various networking domains. Aggregate real-time network data supplies a comprehensive modeling framework used for further analysis, correlation, and visualization. The resulting system not only provides network administrators with a heads-up cockpit display of their entire network, it also supports guided response and predictive capabilities for multi-stage attacks in converged networks.Jerald Dawkins is Founder and Chief Scientist of Digital Enterprise Security Associates, LLC located in Tulsa, Oklahoma. His academic and professional endeavors have provided him with a background in computer security, attack management, risk analysis, and software engineering. He received his B.S. (Computer Science) degree from Fort Lewis College in 1999 and his M.S. and Ph.D. (Computer Science) from the University of Tulsa in 2003 and 2005, respectively.Kevin Clark is a Masters student at the University of Tulsa. He has been involved with research focusing on Security Risk Metrics, Automated Attack Generation and Analysis, and Attack Visualization.Gavin Manes is a Research Assistant Professor at the Center for Information Security and the University of Tulsa. His research interests are information assurance, digital forensics, telecommunications security, and critical infrastructure protection.     

Integrating an urban green space typology into procedural 3D visualization for collaborative planning

Urban green spaces offer multiple ecosystem services (ES), which provide a variety of benefits to human well-being. Yet in urban planning they are not taken into account systematically. Recently new tools have been developed integrating ES into procedural modeling and visualization to raise stakeholder awareness for the explicit ES trade-offs that have to be made. These tools yet do not allow fast and comprehensive integration of ES provision in urban environments. In this paper we show how urban green space typologies can be linked to ES provision for facilitating collaboration between stakeholders of different backgrounds. Based on a generic typology green spaces were mapped and linked with information on potentially provided ES and their parameters. Further, pattern designs of the green space types were described with a form-based code. Both the map of green space types and the pattern designs were integrated into the parametric modeling and visualization chain of Esri CityEngine resulting in 3D visualizations of the green space patterns and correlating ES indicators. The green space typology allows for integrating different kinds of knowledge from both science and practice communities. The procedural model enables rapid interactive visualization of urban patterns and calculation of simple indicator values on the provision of ES. The simple approach for mapping green space types with low data requirements and the generic green space design patterns allow for transferability to other places and application to large areas. The developed approach is simple and fast yet comprehensive to communicate the vital importance of all green space types within the urban environment.     

基于GMF的WS-Security安全策略配置工具研究与实现

现有的基于WS-Security(Web服务安全)的安全工具在技术层面上提供了Web服务安全的基础设施,但是这些工具需要用户拥有丰富的安全知识,并且没有提供展现业务上下文的手段,给业务用户对安全设施的使用带来了困难。基于MDA的思想设计了一个与业务流程结合的安全策略模型。在这个安全模型中,将描述业务流程的应用模型和描述协同应用安全信息的安全模型结合起来,并基于安全模型使用GMF框架实现了相应的配置工具。业务人员可以在可视化的业务流程之上,基于预定义的安全策略模型,配置Web服务安全策略。工具自动将配置的安全策略转换为WS-Security Policy规范文档。     

用于脑网络连接加权图比较的可视分析系统设计与实现

人脑功能网络的研究是近十年生物学领域的重要课题,可视化工具作为数据分析的重要手段,在脑科学研究中有着举足轻重的地位;然而现有的脑功能网络可视化工具存在信息获取效率低、功能单一等问题;针对以上问题,设计并实现了一款用于脑网络连接加权图比较的可视分析系统,帮助研究人员探索不同群组间的差异;首次提出并使用一种用于脑网络连接加权图比较的新可视化方法,针对该方法的用户评估表明,改进后的可视方法在做对比分析任务时更有效;此外,系统将数据挖掘与可视化相结合,增强了群组间差异的表现形式;并且提供了多视图协同等一系列交互方式供研究人员自主探索;最后使用了两组公开数据集进行案例分析,验证了系统的有用性和高效性.