Collaborative Research: Data Poisoning Attacks and Infrastructure-Enabled Solutions for Traffic State Estimation and Prediction

合作研究：数据中毒攻击和基于基础设施的交通状态估计和预测解决方案

• 批准号: 2326341
• 负责人: Yuan Hong
• 金额: $ 16.99万
• 依托单位: University of Connecticut
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2023
• 资助国家: 美国
• 起止时间: 2023-09-01 至 2026-08-31
• 项目状态: 未结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2326341&HistoricalAwards=false
• 关键词: Collaborative; Research; Data; Poisoning; Attacks

This award will support research to investigate "data poisoning" attacks in transportation systems and develop new defense methods to enhance transportation cybersecurity. With ubiquitous data and widely applied data-driven methods in transportation, data poisoning attacks are becoming a critical cybersecurity threat to traffic state estimation and prediction (TSEP), as well as to decision making related to vehicle fleet management and traffic control. This research will have profound societal benefits and impacts by identifying new data poisoning attacks and developing novel defense methods on essential transportation applications. The research will also help raise awareness of data security and facilitate the development of infrastructure-enabled solutions to strengthen transportation security. The team will integrate research results into existing and new courses and will advise both graduate and undergraduate students, especially students from groups underrepresented in science and engineering research, to participate in cutting-edge research. The project team members will participate in multiple outreach programs by providing inputs in science and engineering from this project to K-12 students, especially high school students. The team will also convey research findings to transportation agencies, the academic community, and industry partners. The researchers will transfer research findings to practice, to make significant impacts in the real world. This research will develop a new paradigm in designing transportation data poisoning attacks and developing innovative defense solutions to ensure transportation data security. Data poisoning attacks are first formulated as sensitivity analysis of optimization problems over data perturbations (attacks). Lipschitz continuity-based analysis methods and semi-derivative based algorithms will be developed to help design attack models that are more general and applicable to transportation applications. The team will also develop approximation schemes of the complex objective functions and/or constraints of learning models and study the transferability of attack methods on deep learning models. To defend against the attacks, an infrastructure-enabled defense framework will be developed by leveraging existing and newly deployed secure infrastructure data/information to detect and mitigate attacks. This new defense framework will help develop a secure data network to effectively defend against different attacks on various applications. The research will also provide useful insights to study attacks and develop novel defense methods in other engineering and science fields.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.

该奖项将支持研究调查交通系统中的“数据中毒”攻击，并开发新的防御方法来增强交通网络安全。随着数据的无处不在和数据驱动方法在交通领域的广泛应用，数据中毒攻击正成为交通状态估计和预测(TSEP)以及与车队管理和交通控制相关的决策的关键网络安全威胁。这项研究将对识别新的数据中毒攻击和开发针对关键交通应用的新防御方法产生深远的社会效益和影响。这项研究还将有助于提高人们对数据安全的认识，并促进基础设施支持的解决方案的开发，以加强交通安全。该团队将把研究成果整合到现有的和新的课程中，并将建议研究生和本科生，特别是来自科学和工程研究中代表性不足的群体的学生，参与尖端研究。项目团队成员将通过向K-12学生，特别是高中生提供该项目的科学和工程方面的投入，参与多个外展项目。该团队还将向交通机构、学术界和行业合作伙伴传达研究成果。研究人员将把研究成果转化为实践，在现实世界中产生重大影响。这项研究将在设计交通数据中毒攻击和开发创新的防御解决方案以确保交通数据安全方面开发一种新的范式。数据中毒攻击首先被描述为对数据扰动(攻击)的优化问题的敏感性分析。将开发基于Lipschitz连续性的分析方法和基于半导数的算法，以帮助设计更通用和适用于交通应用的攻击模型。该团队还将开发复杂目标函数和/或学习模型约束的近似方案，并研究攻击方法在深度学习模型上的可转移性。为了防御攻击，将开发一个基础设施启用的防御框架，利用现有和新部署的安全基础设施数据/信息来检测和缓解攻击。这一新的防御框架将有助于开发一个安全的数据网络，以有效防御针对各种应用的不同攻击。这项研究还将为在其他工程和科学领域研究攻击和开发新的防御方法提供有用的见解。这一奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

项目成果

Text-CRS: A Generalized Certified Robustness Framework against Textual Adversarial Attacks

• DOI: 10.1109/sp54263.2024.00053
• 发表时间: 2023-07
• 期刊: ArXiv
• 作者: Xinyu Zhang;Hanbin Hong;Yuan Hong;Peng Huang;Binghui Wang;Zhongjie Ba;Kui Ren

Local Differentially Private Heavy Hitter Detection in Data Streams with Bounded Memory

• DOI: 10.1145/3639285
• 发表时间: 2023-11
• 期刊: Proceedings of the ACM on Management of Data
• 作者: Xiaochen Li;Weiran Liu;Jian Lou;Yuan Hong;Lei Zhang;Zhan Qin;Kui Ren

DPI: Ensuring Strict Differential Privacy for Infinite Data Streaming

• DOI: 10.48550/arxiv.2312.04738
• 发表时间: 2023-12
• 期刊: ArXiv
• 作者: Shuya Feng;Meisam Mohammady;Han Wang;Xiaochen Li;Zhan Qin;Yuan Hong