CAREER: SaTC: Towards Machine-learnable Enhancing Framework for Local Differential Privacy

以数据为中心的应用程序的流行需要收集和分析可能包含用户敏感数据的信息。虽然局部差分隐私（LDP）可以量化地控制信息泄漏，以保护用户隐私，它是具有挑战性的，以解决一个动态的威胁景观，攻击者可以利用LDP设计的特点操纵分析结果。该项目的新颖之处在于利用机器智能沿着其他使能技术来理解LDP隐私、安全和实用性之间复杂的相互作用，并开发一个可机器学习的LDP增强框架来解决它们之间的矛盾。该项目的成功将为未来铺平道路，在未来，无处不在的机器智能可以在各种具有挑战性的场景中理解和关注用户隐私，安全和实用需求。该项目更广泛的意义和重要性在于扩大妇女和代表性不足的学生对STEM的参与;鼓励跨学科、跨部门伙伴关系以及技术转型和孵化;促进多样性和公平意识的技术发展。这个CAREER项目旨在开发一种新的人工智能辅助的隐私增强框架，可以更好地解决LDP安全性，实用性和隐私之间的紧张关系。将产生可推广的理论和原则，以指导开发的机器代理感知部署环境，并学习对所观察到的对抗行为、预期效用和隐私目标的最佳响应。该代理还可以自我解释它所采取的行动，并可以与用户和开发人员进行交互，从而提高算法的透明度和问责制。此外，人类利益相关者将参与系统的成长周期，使其能够随着时间的推移而发展。该项目还旨在开展基于研究的教育活动，以加强内政部的网络安全相关教育和辅导计划，同时加强大学及其他级别的网络安全劳动力培训。同时，该项目旨在通过与不同背景的K-12学生开展外展活动，促进STEM职业发展。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审查标准进行评估，被认为值得支持。

The prevalence of data-centric applications demands collecting and analyzing the information that may contain sensitive data of users. While local differential privacy (LDP) can quantifiably control information leakage to protect user privacy, it is challenging to tackle a dynamic threat landscape where the attacker can manipulate the analytical results by exploiting the characteristics of LDP design. The project’s novelties are to leverage machine intelligence along with other enabling techniques to understand the complex interplay between LDP privacy, security, and utility, and develop a machine-learnable LDP enhancing framework to address their tension. The success of the project will pave the way toward a future where ubiquitous machine intelligence can understand and attend to user privacy, security, and utility demands in various challenging scenarios. The project’s broader significance and importance are broadening the participation of women and underrepresented students in STEM; encouraging interdisciplinary, cross-sector partnership, and technology transformation and incubation; and promoting diversity-and-equity-aware technology development. This CAREER project aims to develop a novel AI-assisted privacy-enhancing framework that can better address the tension between LDP security, utility, and privacy. Generalizable theories and principles will be produced to guide the developed machine agent to sense the deployment environment and learn optimal responses to the observed adversarial actions, expected utility, and privacy goals. The agent will also be self-explainable for the actions it takes and can interact with both users and developers, thereby improving algorithmic transparency and accountability. In addition, human stakeholders will get involved in the growing cycle of the system, allowing it to evolve over time. The project also aims to conduct research-informed educational activities that will strengthen cybersecurity-related education and mentoring programs in the home department while enhancing cybersecurity workforce training at the university level and beyond. Meanwhile, the project seeks to promote STEM careers by engaging in outreach activities with K-12 students with diverse backgrounds.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.