III: Small: Collaborative Research: Towards Interactive Data Visualization Management Systems

交互式可视化是一种从数据中探索和获取见解的强大方式。随着从业者可用数据的复杂性和规模不断增长，现有系统发现越来越难以维持高度交互的体验。这个项目的目标是开发一个交互式的视觉数据探索系统，设计和优化，考虑到人类的感知。本研究的目的是将人类感知建模为感知功能。这些功能可以帮助系统避免不必要的计算可视化结果，这些结果比最终用户可以感知到的更准确。通过开发和使用这些功能，系统可以为商业智能、数据驱动科学和医疗保健分析等领域的大型数据集提供高度精确的交互式可视化。此外，这些努力的研究成果将有助于开发研究生水平的新课程主题。交互式可视化系统中一个经常被忽视的元素是循环中的人。尽管随着时间的推移，数据大小和计算能力急剧增加，但人类的感知极限仍然相对稳定。虽然以前的工作已经提出了有效动画的指导方针[Heer等人，VCG 2007；Fisher等人，ICGA 2012]和M4等项目[Jugel等人，VLDB 2014]使用感知洞察力来证明近似算法的合理，我们的目标是建立一个多层数据分析系统，该系统将交互式可视化客户端与后端数据管理系统统一起来，并明确考虑到人类感知模型。这些模型可用于开发感知感知的优化，例如：(1)自动近似感知上无法区分的数据转换，(2)将交互（例如，向右拖动滚动条）生成的查询建模为单个会话，并跨整个查询集进行优化，以及(3)应用面向交互的缓存和重写策略以最小化延迟。最终，这些技术可以确保数据探索的高帧率交互，而不会对用户从可视化中获得的见解产生负面影响。该项研究的进一步信息、出版物和结果可在该项目的网站（http://perceptvis.github.io）上获得。

Interactive visualizations are a powerful way to explore and draw insights from data. As the data available to practitioners continues to grow in complexity and size, existing systems find it harder and harder to maintain a highly interactive experience. The goal of this project is to develop an interactive visual data exploration system designed and optimized to take human perception into account. The aim of this research is to model human perception as perceptual functions. These functions help the system avoid unnecessarily computing visualization results that are more accurate than what can be perceived by the end user. By developing and using these functions, the system can provide highly accurate yet interactive visualizations for large datasets in domains such as business intelligence, data-driven sciences, and healthcare analytics. In addition, research results from these efforts will contribute towards the development of new curriculum topics at the graduate level.A commonly overlooked element of interactive visualization systems is the human in the loop. Although data sizes and computational capabilities have dramatically increased over time, human perceptual limits have remained relatively constant. Although previous work has presented guidelines for effective animations [Heer et al., VCG 2007; Fisher et al., ICGA 2012] and projects such as M4 [Jugel et al., VLDB 2014] used perceptual insights to justify approximation algorithms, our goal is to build a multi-layered data analysis system that unifies interactive visualization clients with backend data management systems, and explicitly takes human perceptual models into account. These models can be used to develop perceptually-aware optimizations such as (1) automatically approximate data transformations that are perceptually indistinguishable, (2) model queries generated by an interaction (e.g., dragging a scrollbar to the right) as a single session and optimize across the entire set of queries, and (3) apply interaction-oriented caching and rewrite strategies to minimize latency. Ultimately, these techniques can ensure high frame-rate interactions for data exploration without negatively impacting the insights that users draw from their visualizations. Further information, publications and results of this research are available at the project web site (http://perceptvis.github.io).