Uncertainty-Aware Data Transformations for Collaborative Reasoning

摘要：从海量、动态和可能不完整的数字数据中获得洞察力的能力对于那些收集这些数据以进行时间关键型决策的人来说是绝对必要的。这项研究正在开发量化、传播和聚合这些数据的数学公式，以支持使用视觉手段的协作推理。本研究的根本重要性在于，考虑到数据转换和将该信息传播到推理阶段所带来的不确定性，为分析师提供更值得信赖的数据视图。可视化分析的过程可以看作是一系列的数据和可视化转换。调查人员正在研究帮助分析人员发现大型数据存储库中的结构关系的转换。这些关系可以表达数据的各个方面，例如不同数据元素之间的成员关系、包含、重叠和双向映射。然而，数据收集和转换步骤都会产生一定程度的不确定性，其中一些是由于转换，而另一些则固有地存在于源数据中。不确定性会降低信息的质量。理解和量化这种不确定性对于协作推理是至关重要的，协作推理是来自几个分析人员的置信水平聚集的产物。本研究将这种转换和不确定性建模为贝叶斯网络。该网络中的节点表示转换及其相关的不确定性，也表示不同协作分析人员的置信水平。利用传播定律和条件概率对变换之间和分析人员之间的不确定性聚合进行建模。这样的模型产生了一个可视化的分析框架，提供了不确定性的显式表示，使分析人员能够为获得的洞察力分配置信度。这个框架有效地帮助减轻了意义构建和视觉推理的复杂性。

Title: Uncertainty-Aware Data Transformation for Collaborative ReasoningPI: Kwan-Liu Ma, University of California at DavisAbstract:The ability to obtain insight from massive, dynamic, and likely incomplete digital data is absolutely essential to those who collect these data for time-critical decision-making. This research is developing mathematical formulations for the quantification, propagation, and aggregation of such data to support collaborative reasoning using visual means. The fundamental importance of this research is to give analysts a more trustworthy view of data with the consideration and incorporation of uncertainty due to data transformations and the propagation of that information to the reasoning stage.The process of visual analytics can be viewed as a series of data and visual transformations. The investigators are studying the transformations that help analysts discover structural relationships in large data repositories. These relationships can express aspects of the data such as membership, inclusion, overlapping, and bijective mappings between different data elements. Both data collection and transformation steps, however, incur certain degrees of uncertainty, some due to the transformations, and others inherently present in the source data. Uncertainty can degrade the quality of the information. Understanding and quantifying this uncertainty is crucial to collaborate reasoning, which is the product of the aggregation of confidence levels from several analysts. This research models such transformations and uncertainties as a Bayesian network. Nodes in this network represent the transformations and their associated uncertainty, and also represent the confidence levels of different collaborating analysts. Aggregation of uncertainty between transformations and among analysts is modeled using propagation laws and conditional probabilities. Such a model results in a visual analytics framework providing an explicit representation of uncertainty to enable analysts to assign confidence levels to the gained insight. This framework effectively helps alleviate the complexity of sense-making and visual reasoning.