CAREER: Efficient Structural Analysis of Multivariate Fields for Scalable Visualization

可视化在研究和工业中发挥着至关重要的作用，它为用户提供数据的图形界面，为他们提供解释、评估和决策的直观基础。然而，快速增长的规模、维度和所产生的科学数据的多尺度复杂性创造了一个普遍的分析挑战，而现有的可视化技术并没有适当地解决这个挑战。特别是，对固有的多变量和多场物理问题的研究通常被简化为单个标量或向量场的可视化，从而忽略了丰富的重要信息。PI将通过开拓大规模数据集的高效可视化分析的综合方法来解决当前艺术状态的局限性。提出的方法的核心是几何显著性的新定义，这将允许自动识别科学数据中的显著流形。为此，PI将统一并包含来自数学和计算机视觉的各种概念，为多领域问题的结构分析和视觉表示创建一个原则性和通用的模型。将设计利用并行架构的创新数据结构和稀疏采样策略，以实现大规模和多变量数据集的有效处理。最后，这些计算基础将为一个以用户为中心的可视化分析框架提供动力，PI将在流体动力学、材料工程、高能物理和心血管研究等多学科合作的背景下进行评估。这一研究成果将有利于科学界，因为它为广泛科学问题的计算或测量数据集的有效分析和可视化提供了一个严格和可扩展的框架。PI将通过开源门户网站分发创建的软件工件，并将其集成到领先的可视化工具中，以促进其传播。PI将在主要会议上组织教程和研讨会，以提高用户社区对已开发技术的认识，他将提供基准数据集和样本结果，以促进可视化社区的协作努力。除了研究之外，PI还将在本科和研究生阶段开设新课程，让学生了解数据分析在科学和工程中的重要性，以及高级可视化在这一背景下的作用。最后，这些教育活动将自然地补充对代表性不足的少数民族和当地K-12课程的推广努力。

Visualization plays a crucial role in research and industry by offering users a graphical interface to their data that affords them an intuitive basis for interpretation, assessment and decision making. Yet, the rapidly growing size, dimensionality, and multi-scale complexity of the produced scientific data create a pervasive analysis challenge that is not properly addressed by existing visualization technology. In particular, the investigation of inherently multivariate and multifield physical problems is typically reduced to the visualization of a single scalar or vector field, thereby neglecting a wealth of important information.The PI will address the limitations of the current state of the art by pioneering a comprehensive approach for the efficient visual analysis of large-scale datasets. Central to the proposed approach is a novel definition of geometric saliency that will permit the automatic identification of remarkable manifolds in scientific data. To that end, the PI will unify and subsume a variety of concepts from mathematics and computer vision to create a principled and versatile model for the structural analysis and visual representation of multifield problems. Innovative data structures and sparse sampling strategies leveraging parallel architectures will be devised to enable the efficient processing of large and multivariate datasets at scale. Lastly, these computational foundations will power a user-centric visual analysis framework that the PI will assess in the context of multidisciplinary collaborations spanning fluid dynamics, materials engineering, high-energy physics, and cardiovascular research.This research effort will benefit the scientific community by contributing a rigorous and scalable framework for the effective analysis and visualization of computational or measured datasets across a broad range of scientific problems. The PI will distribute the created software artifacts through an open source web portal and integrate them in leading visualization tools to facilitate their dissemination. The PI will organize tutorials and workshops at premier conferences to raise the awareness of the user community about the developed technology and he will provide benchmark datasets and sample results to promote a collaborative effort in the visualization community. Beyond research, the PI will create new courses at both the undergraduate and graduate levels to expose students to the critical importance of data analysis in science and engineering and to the role of advanced visualization in this context. Finally, these education activities will naturally complement an outreach effort toward underrepresented minorities and local K-12 programs.