关于举办美国石溪大学徐素素教授学术讲座的通知
发布时间: 2021-09-13

题目:迈向自适应智慧城市:协同集成感知、学习和驱动以实现大规模城市监测

Towards a Self-adaptive Smart City: Collaboratively Integrating Sensing, Learning and Actuation for Large-scale Urban Monitoring

时间:20210917日周五9001000

地点:腾讯会议 ID42265298356(若300人满请入直播间)

腾讯直播间:https://meeting.tencent.com/l/QALGLWoVMAcS

报告人:徐素素(美国石溪大学,土木工程系)

徐老师官方主页:https://www.stonybrook.edu/commcms/civileng/people/_core_faculty/xu_susu.php

欢迎广大师生参加!

                              

                                     土木与交通学院

                                                                                                 20210913



报告人简介:

  徐素素博士是石溪大学土木工程系助理教授,还是计算机科学系兼任教员。她本科毕业于清华大学获得学士学位,随后在美国卡内基梅隆大学获得机器学习硕士学位和土木工程博士学位。她曾任美国斯坦福大学博士后研究员和美国Qualcomm技术公司人工智能研究团队的研究科学家。她的研究重点是协同整合群体感知、物理信息机器学习和激励机制,以实现自适应智能城市基础设施系统并提高城市服务的效率、稳健性和可持续性。她在2018年的IEEE机器学习与应用国际大会(ICMLA)上获得了最佳论文奖,并获得了NeurIPS 2018对抗性视觉挑战赛的冠军。她曾入选2019年麻省理工学院CEE明日之星和多德奖学金等荣誉。

https://www.stonybrook.edu/commcms/civileng/people/_core_faculty/xu_susu.php

 

Dr. Susu Xu is an assistant professor at Department of Civil Engineering, and affiliated faculty at Department of Computer Science, Stony Brook University. She received her Ph.D. in Civil Engineering and Master in Machine Learning from Carnegie Mellon University, her bachelor’s degree from Tsinghua University. She has been postdoctoral research fellow at Stanford University and research scientist at the AI research team in Qualcomm Technologies. Her research focuses on collaboratively integrating crowdsensing, physics-informed machine learning, and incentive mechanisms for enabling self-adaptive smart urban infrastructure systems and improving the efficiency, robustness, and sustainability of urban services. She received the Best Paper Award at the IEEE International Conference of Machine Learning and Applications (ICMLA) in 2018, and the champion of NeurIPS 2018 Adversarial Vision Challenge. She is also the recipient of 2019 MIT CEE Rising Star and Dowd Fellowship. 

 

报告摘要:

  随着人口的增长和对高质量城市服务的需求,迫切需要建设一个能够在不断变化的动态下自主调整城市基础设施系统监控和管理策略的“自适应”城市。近年来,传感器网络和5G技术的快速发展使大规模多源数据和实时多智能体控制成为可能。但是大规模且相互依赖的物理基础设施系统对数据驱动的监控和管理策略提出了挑战。在本次演讲中,我将介绍一个整合了资源感知、物理信息学习和大型城市监测和管理激励机制的框架。首先,我将谈谈我在城市人群感知系统方面的工作。该系统利用安装在个人移动设备和车辆上的低成本传感器来自动感知城市时空信息,并实时了解潜在的城市动态(例如人员流动、空气质量、交通拥堵、道路恶化等)。然后,我将介绍关于激励大规模车辆移动和人类活动的新理论和算法,以及时对检测到的城市系统变化做出反应,并长期保持最佳的城市监控系统。我还将介绍中国首个部署在200多辆出租车上的车辆城市群体感知系统,这些系统已经收集了41.8万公里的总里程及5亿个城市数据点。最后,我将简要介绍我在物理信息学习和分布式学习算法开发方面的研究,以便有效理解大规模城市系统。

With increasing populations and demand of high-quality urban services, there is an urgent need of building a “self-adaptive” city which can autonomously adapt its monitoring and management strategies for urban infrastructure systems under constantly changing dynamics. The recent rapid development of sensor networks and 5G technologies are enabling large-scale multi-source data and real-time multi-agent control. But the large-scale and interdependent physical infrastructure systems pose challenges to data-driven monitoring and management strategies. In this talk, I will introduce a framework that collaboratively integrates resource-aware sensing, physics-informed learning and incentive mechanisms for large-scale urban monitoring and management. First, I will talk about my works on urban crowdsensing systems. The system utilizes low-cost sensors mounted on individual mobile devices and vehicles to automatically sense spatio-temporal urban information, and learn real-time underlying urban dynamics (e.g., human mobility, air quality, traffic congestion, road deterioration, etc.). Then I will introduce my new theory and algorithm on incentivizing large-scale vehicle mobilities and human activities to timely react to the detected changes in urban systems and maintain a long-term optimal urban monitoring system. I will introduce the first deployed vehicular urban crowdsensing system built on more than 200 taxis in China, which have already run a total mobile mileage of 418,000km and collected half-billion urban data points. Further, I will briefly mention my research on physics-informed learning and distributed learning algorithms development for efficient understanding of large-scale urban systems.