OAC Core: A Machine Learning Assisted Visual Analytics Approach for Understanding Flow Surfaces

OAC Core：一种用于理解流表面的机器学习辅助视觉分析方法

• 批准号: 2104158
• 负责人: Chaoli Wang
• 金额: $ 49.99万
• 依托单位: University of Notre Dame
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2022
• 资助国家: 美国
• 起止时间: 2022-01-01 至 2024-12-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2104158&HistoricalAwards=false
• 关键词: OAC; Core; Machine; Learning; Assisted

Among the popular integration-based techniques for fluid flow visualization, line-based techniques have made significant advances over the years, providing a sharp contrast to surface-based techniques. Surface-based flow visualization is often regarded as one of the most challenging problems in scientific visualization. Flow surfaces can provide better illustrative capabilities and much-improved visualization compared to flow lines. However, existing surface-based flow visualization methods face substantial challenges related to surface selection, visualization, and analytics, calling for creative ideas and enabling solutions. The overarching goal of this project is to develop a machine learning-assisted visual analytics approach for understanding three-dimensional complex flow surfaces. The project contributes to the state-of-the-art flow visualization by providing end-to-end solutions that introduce an innovative deep learning approach to surface feature learning, investigate a new template-based visual interface for flow surface exploration, and explore occlusion reduction and relationship interrogation via visual transformation. The team will work with domain experts and apply the proposed solutions to study cardiovascular flow, aligning with the NSF's mission to promote the progress of science and to advance the national health, prosperity, and welfare.Underlying the proposed work is a novel data-driven and user-centric approach for surface selection and exploration. The project will develop a new framework that supports (1) selection of representative surfaces through feature learning, projection, and clustering powered by graph neural networks, (2) exploration of surface patterns via a principled vocabulary-based method for shape-invariant partial flow surface querying and matching, and (3) comparative analytics of flow surfaces for studying seeding sensitivity or variability via a river-like visual metaphor. Research results will be evaluated through comparison against existing methods, assessment by domain experts, and a formal user study. Furthermore, for cardiovascular flows the project will investigate the hemodynamics of aortic dissection and intracranial aneurysm for solution validation. The project outcomes include (1) source code that implicitly learns surface features using graph neural networks, (2) a surface shape database based on the universal alphabet, and (3) an educational tool for teaching and learning flow surfaces. The general approach will impact other fields of study, including machine learning for visualization, shape analysis, visual analytics, and visualization in education.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

在流行的基于集成的流体流动可视化技术中，基于线的技术多年来取得了重大进展，与基于表面的技术形成了鲜明的对比。基于表面的流场可视化是科学可视化中最具挑战性的问题之一。与流线相比，流动表面可以提供更好的说明能力和更好的可视化。然而，现有的基于表面的流动可视化方法面临着与表面选择、可视化和分析相关的重大挑战，需要创造性的想法和实现解决方案。该项目的总体目标是开发一种机器学习辅助的可视化分析方法，用于理解三维复杂流面。该项目通过提供端到端解决方案，为最先进的流动可视化做出贡献，这些解决方案将创新的深度学习方法引入表面特征学习，研究用于流动表面探索的新的基于模板的视觉界面，并通过视觉转换探索遮挡减少和关系询问。该团队将与领域专家合作，并将所提出的解决方案应用于研究心血管血流，与NSF的使命保持一致，以促进科学进步，促进国家健康，繁荣和福利。所提出的工作是一种新的数据驱动和以用户为中心的表面选择和探索方法。该项目将开发一个新的框架，支持（1）通过图形神经网络提供的特征学习，投影和聚类选择代表性表面，（2）通过基于词汇的原则性方法探索表面模式，用于形状不变的部分流表面查询和匹配，和（3）通过类似河流的视觉隐喻研究播种敏感性或可变性的流动表面的比较分析。研究结果将通过与现有方法的比较，领域专家的评估和正式的用户研究进行评估。此外，对于心血管血流，该项目将研究主动脉夹层和颅内动脉瘤的血流动力学，以进行解决方案确认。项目成果包括：（1）使用图形神经网络隐式学习表面特征的源代码，（2）基于通用字母表的表面形状数据库，以及（3）用于教学和学习流动表面的教育工具。该奖项反映了NSF的法定使命，并通过使用基金会的智力价值和更广泛的影响力审查标准进行评估，被认为值得支持。

项目成果

Reconstructing Unsteady Flow Data From Representative Streamlines via Diffusion and Deep-Learning-Based Denoising

• DOI: 10.1109/mcg.2021.3089627
• 发表时间: 2021-11-01
• 期刊: IEEE COMPUTER GRAPHICS AND APPLICATIONS
• 影响因子: 1.8
• 作者: Gu, Pengfei;Han, Jun;Wang, Chaoli
• 通讯作者: Wang, Chaoli

VCNet: A generative model for volume completion

• DOI: 10.1016/j.visinf.2022.04.004
• 发表时间: 2022-04
• 期刊: Vis. Informatics
• 作者: Jun Han;Chaoli Wang

NeRVI: Compressive neural representation of visualization images for communicating volume visualization results

• DOI: 10.1016/j.cag.2023.08.024
• 发表时间: 2023-08
• 期刊: Comput. Graph.
• 作者: Pengfei Gu;Da Chen;Chaoli Wang

Hierarchical Sankey Diagram: Design and Evaluation

分层桑基图：设计和评估

• DOI: 10.1007/978-3-030-90436-4_31
• 发表时间: 2021
• 期刊: International Symposium on Visual Computing
• 作者: Porter, William P;Murphy, Conor P;Williams, Dane R;O'Handley, Brendan J.;Wang, Chaoli
• 通讯作者: Wang, Chaoli

DL4SciVis: A State-of-the-Art Survey on Deep Learning for Scientific Visualization

• DOI: 10.1109/tvcg.2022.3167896
• 发表时间: 2022-04
• 期刊: IEEE Transactions on Visualization and Computer Graphics
• 影响因子: 5.2
• 作者: Chaoli Wang;J. Han