SATC: CORE: Medium: Principles and Algorithms for Visual Data Exploration Under Differential Privacy

数据科学的一个基本部分是可视化数据探索，它使用可视化和用户交互来促进从数据中发现新知识和可操作的见解。当前的可视化分析技术很好地支持可视化数据探索。然而，在许多领域，正在探索的数据包括个人的个人事实。因此，对数据的访问可能受到隐私政策或法规的限制。该项目将开发新的可视化数据探索技术，该技术可以支持新知识的发现，同时保证个人隐私得到保护。这些技术将在医疗保健数据、政府管理数据和移动数据的背景下进行研究。该项目将扩展对私人数据的安全有效的探索，允许更广泛的数据科学家社区从更丰富的数据源中产生见解，包括那些以前由于隐私问题而被禁止的数据源。本项目开发的视觉探索方法将为差分隐私模型提供保障，差分隐私模型正在成为个人数据保护的主导标准。在提供差异化隐私保证的同时，实现对数据的精确视觉探索需要隐私算法、可视化技术的新进展，以及仔细的评估方法和人类受试者实验。在差分隐私下支持数据可视化的根本挑战源于隐私算法和可视化技术之间的复杂交互。如果对可视化任务进行定制，则可以更好地设计用于私有数据发布的算法。特殊的可视化方法需要用于嘈杂的私有数据，包括那些传达不确定性和对虚假视觉伪像具有鲁棒性的数据。提出的研究有可能通过以下方式改变私有数据的使用：(i)调查当前的可视化和交互技术应该如何适应由差异私有算法引入的噪声；（ii）开发新的算法，为静态可视化和交互式视觉探索提供更好的视觉准确性；（iii）提供基准和评估标准，以加速私有可视化的创新。这些算法的有效性和价值将通过运行一系列以人为本的评估来进行实证评估。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

A fundamental part of data science is visual data exploration, which uses visualization and user interaction to facilitate the discovery of new knowledge and actionable insights from data. Visual data exploration is well-supported by current visual analytics technology. However, in many domains, the data being explored includes personal facts about individuals. Access to the data may therefore be limited by privacy policies or regulations. This project will develop novel visual data exploration technology that can support the discovery of new knowledge while at the same time guaranteeing that individuals’ privacy is protected. These technologies will be studied in the context of healthcare data, government administrative data, and mobility data. This project will expand the safe and effective exploration of private data, allowing a broader community of data scientists to generate insights from a richer set of data sources, including those previously off-limits due to privacy concerns. The visual exploration methods developed in this project will provide guarantees in the model of differential privacy, which is emerging as the dominant standard for protecting personal data. Enabling accurate visual exploration of data while offering a guarantee of differential privacy requires novel advances in privacy algorithms, visualization technology, as well as careful evaluation methodology and experiments with human subjects. The fundamental challenges of supporting data visualization under differential privacy stem from the complex interaction between privacy algorithms and visualization techniques. Algorithms for private data release can be better designed if they are customized to visualization tasks. And special visualization methods need to be used with noisy privatized data, including those that communicate uncertainty and are robust to spurious visual artifacts. The proposed research has the potential to transform the use of private data by (i) investigating how current visualization and interaction techniques should be adapted in the presence of noise introduced by differentially private algorithms, (ii) developing new algorithms that offer better visual accuracy, for both static visualizations and interactive visual exploration, and (iii) providing a benchmark and evaluation standards to accelerate innovation in private visualization. The effectiveness and value of these algorithms will be evaluated empirically by running a series of human-centered evaluations.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.