Collaborative Research: Frameworks: MobilityNet: A Trustworthy CI Emulation Tool for Cross-Domain Mobility Data Generation and Sharing towards Multidisciplinary Innovations

在人工智能时代，获取大规模数据对推进科学和工程研究起着至关重要的作用。然而，大多数这些移动数据通常是专有的，因此除非数据所有者发布，否则科学和工程社区无法访问或获得昂贵的数据。最近，在Data Science for Social Good的背景下，一些数据所有者愿意与公众分享他们的移动数据，以释放他们的价值，但仍然存在重大的隐私问题。该项目背后的关键科学驱动力是科学和工程社区缺乏清洁的多样化移动数据访问与生成机器学习和实际差异隐私的进步之间的差距。为了弥合这一差距，项目团队构建了一个可信赖的网络基础设施（CI）仿真工具，称为MobilityNet，以(1)通过强调MobilityNet的可信度（即实用性、隐私性和公平性）的隐私保护生成机器学习，基于真实数据合成真实的移动数据；(2)通过与19个合作伙伴的合作，与科学和工程社区共享这些数据，以进行多学科创新。通过计算机科学、交通工程、城市和区域规划、地理、流行病学和经济学等多个科学和工程学科的研究，MobilityNet产生的数据有可能产生重大的科学和社会影响。该项目设计了MobilityNet，这是一个创新的行为启发的生成式CI工具，用于可靠的移动数据合成（即平衡效用、隐私和公平性）。合成数据将通过三个CI组件生成，这些组件建立在经过验证的模型（如生成式机器学习和差分隐私）的基础上，从技术和人的角度：(1)跨域真实数据管理，以解决数据偏差；(2)社会知情的真实数据解释，以解决数据隐含性；(3)保护隐私的合成数据生成，以解决数据敏感性。MobilityNet的关键创新首先以社会科学为基础，从用户研究中获得见解，以便更好地理解设计的CI工具在跨域设置中对效用、隐私和公平性的影响；然后，它在CI设计和实施（例如，数据管理和生成）方面具有一系列社会知情的技术优点；进一步用可测量的指标进行评估；它最终通过综合数据共享产生影响，使科学和工程社区受益。MobilityNet的研究愿景将有助于国家CI生态系统的成功。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。

In the era of Artificial Intelligence, accessing large-scale data plays a pivotal role in advancing Science and Engineering research. However, most of these mobility data are often proprietary and thus cannot be accessed or are costly to acquire by the Science and Engineering community unless released by data owners. Recently, under the context of Data Science for Social Good, some data owners are willing to share their mobility data with the public to unlock their values, but there are significant privacy concerns that remain in the way. The key science driver behind this project is the gap between the lack of sanitized diverse mobility data access for the Science and Engineering community and the advance of generative Machine Learning and practical differential privacy. To bridge this gap, the project team builds a trustworthy cyberinfrastructure (CI) emulation tool called MobilityNet to (1) synthesize realistic mobility data based on real data, via privacy-preserving generative machine learning emphasizing MobilityNet's trustworthiness (i.e., utility, privacy, and fairness); and (2) share these data with the Science and Engineering community for multidisciplinary innovations by working with 19 partners. The data generated by MobilityNet have the potential for significant scientific and societal impacts via research in multiple Science and Engineering disciplines such as Computer Science, Transportation Engineering, Urban and Regional Planning, Geography, Epidemiology, and Economics. This project designs MobilityNet, an innovative behavior-inspired generative CI tool for trustworthy mobility data synthesis (i.e., balancing utility, privacy, and fairness). The synthetic data will be generated via three CI components built upon validated models (such as generative machine learning and differential privacy) from both technical and human aspects: (1) cross-domain real data curation to address data bias, (2) socially-informed real data interpretation to address data implicitness, and (3) privacy-preserving synthetic data generation to address data sensitivity. The key innovation of MobilityNet is first grounded in social science, where insights are drawn from user studies to better understand the impact of the designed CI tool on utility, privacy, and fairness in a cross-domain setting; it is then materialized with a set of socially-informed technological merits on CI design and implementation (e.g., data curation and generation); it is further evaluated with measurable metrics; it finally creates impacts through synthetic data sharing to benefit the Science and Engineering community. The research vision in MobilityNet will contribute to the success of the national CI Ecosystem.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.