The 8th NYUAD Transportation Symposium

Speaker Bios and Abstracts

• Yinhai Wang, University of Washington | Edge AI Empowered Smart Hub for Transportation Safety and Infrastructure Intelligence Enhancement

  Biography

  Dr. Yinhai Wang is a professor of transportation engineering at Civil and Environmental Engineering and
  an adjunct professor at the Electrical and Computer Engineering of the University of Washington (UW). He has served as director for Pacific Northwest Transportation Consortium (PacTrans), USDOT University
  Transportation Center for Federal Region 10 since 2012, and for Northwestern Tribal Technical
  Assistance Program Center since 2022, and Washington State Transportation Center (TRAC) since 2023.

  Dr. Wang was an elected governor for the IEEE Intelligent Transportation Systems Society (ITSS) from
  2011 through 2013 and served as the 2018-2019 president of the Transportation & Development
  Institute (T&DI) for American Society of Civil Engineers (ASCE). He is currently chair of the Artificial
  Intelligence (AI) and Advanced Computing Applications Committee of the Transportation Research
  Board (TRB), vice president for Council of University Transportation Centers (CUTC), chair for the ASCE
  Center on Technical Advancement, and the Connected and Automated Transportation Special Section
  Editor for ASCE Journal of Transportation Engineering Part A: Systems. His active research fields include
  traffic sensing, transportation safety, transportation data science, edge computing, traffic operations
  and simulation, smart urban mobility, and AI methods and applications. He is a professional engineer
  registered in Washington, a member of the Washington State Academy of Sciences, a Distinguished
  Member of ASCE, and a fellow of both IEEE and ITE. He has been recognized by numerous awards, such
  as the 2023 ASCE Francis C. Turner Award and the Institute of Transportation Engineers (ITE) Innovation
  in Education Award for 2018.
  

  Abstract

  Artificial Intelligence (AI) has emerged as a transformative force, reshaping conventional
  paradigms and revolutionizing global road networks. This presentation will delve into the
  critical role of AI and edge computing in enhancing road transportation systems, elevating
  infrastructure intelligence, and improving transportation safety. Particularly, the speaker will
  spotlight the Mobile Unit for Sensing Traffic (MUST) technology, developed by the University of
  Washington's Smart Transportation Applications and Research Laboratory (STAR Lab). This
  smart hub amalgamates tailored AI, edge computing, and connected vehicle technologies to
  augment safety and infrastructure insights. Potential impacts of this technology on
  transportation safety and infrastructure intelligence will be discussed. Additionally, the
  presentation will address the opportunities and challenges associated with AI research and
  applications, including issues related to privacy, security, and equity.

• Cathy Wu, MIT | Learning-guided Optimization for Mobility

  Biography

  Dr. Cathy Wu is an associate professor at MIT in LIDS, CEE, and IDSS. She holds a PhD from UC Berkeley, and BS and M.Eng. from MIT, all in EECS, and completed a postdoc at Microsoft Research. Her research aims to leverage machine learning to solve hard optimization problems for next-generation mobility systems. She is broadly interested in leveraging modern computing and AI to advance decision-making. Wu has received a number of awards, including the NSF CAREER, PhD dissertation awards, and publications with distinction. She serves on the Board of Governors for the IEEE ITSS, is a Program Co-chair for RLC 2025, and is an AC/AE for ICML, NeurIPS, and ICRA. She is also helping spearhead efforts towards reproducible research in transportation.

  Abstract

  Critical to designing future transportation systems is solving many hard optimization and control problems. As such, researchers increasingly look towards data-driven methods such as deep reinforcement learning (RL). However, our research suggests that RL's practical application is limited by its brittleness, sometimes failing to train in the presence of small changes in the environment. On the other hand, we show that learning-guided optimization, which hybridizes machine learning and classical optimization, can effectively accelerate state-of-the-art solvers by up to 2-7 times and demonstrate generalizability across problem variations. Applications discussed include mixed autonomy traffic, traffic signal control, vehicle routing problems, multi-robot warehousing, and integer linear programming.
  

• Lily Du, University of Florida | Traffic Flow Model-Based AI for Advanced Traffic Prediction

  Biography

  Dr. Lili Du is a professor in the Civil and Coastal Engineering Department, University of Florida at
  the end of 2024. She was a NRC Renowned Research Associate at Turner-Fairbank Highway Research
  Center from 2023-2024. Before that, she worked as an assistant and then an associate professor at the
  Illinois Institute of Technology (IIT) from 2012 to 2017, and as a post-doctoral research associate for
  NEXTRANS at Purdue University from 2008 to 2012. Dr. Du received her Ph.D. degree in Decision
  Sciences and Engineering Systems with a minor in Operations Research and Statistics from Rensselaer
  Polytechnic Institute in 2008.

  Dr. Du’s research is characterized by integrating operations research, network modeling, game theory, control theory, machine learning, and statistical methods into traffic flow analysis, transportation system analysis, and network modeling. Her current research mainly focuses on the impacts of connected and/or autonomous vehicles and electric vehicles, mobility on demand, smart curb, network resilience, and traffic flow analysis.

  Dr. Du’s research has been published in Transportation Research Part B, Part C, and Part D, IEEE Transactions on ITS, Networks and Spatial Economics. Her research has been funded by National Science Foundation (NSF), State DOT, STRIDE UTC, and Toyota InfoTechnology Center. Dr. Du is a recipient of the NSF CAREER award in 2016. Her recent project, “Driverless City” won the First Nayar Prize at IIT. She is currently chairing TRB ADB30-5 subcommittee on Emerging Technologies in Network Modeling and ASCE-T&DI Artificial Intelligence in Transportation Committee. She serves as an editor for Transportation Research Part B:
  Methodological, an associate editor for IEEE Transactions on Intelligent Transportation Systems, and an
  editor of the Journal of Intelligent Transportation Systems.
  

  Abstract

  In the face of expanding volumes and diverse arrays of transportation data, the potential of machine learning (ML) and artificial intelligence (AI) technologies to enhance traffic prediction, drive transformative informed decisions, and offer innovative, proactive solutions for future transportation system management is significant. However, to fully unlock the power of AI/ML technologies, customization with transportation domain knowledge, particularly traffic flow analysis, is imperative. This is because transportation systems, intricate and dynamic in their physical nature, generate extensive and complex traffic data that demands sophisticated tools for extracting meaningful predictions and insights. While traditional traffic flow models establish a foundational understanding of dynamics, they fall short in adapting to real-time variations. AI/ML's capacity to discern patterns, adapt to dynamic conditions, and process vast datasets positions it as a potent ally in traffic prediction. However, this potentially comes with the caveat of nonsensical results if not tailored to consider traffic flow features. Therefore, AI/ML augmenting traffic flow analysis by leveraging both physical rigor and rich information from historical and real-time data enables anticipatory responses to traffic patterns, congestions, and potential disruptions. This presentation exemplifies this concept by sharing several of our research studies. They integrated ML/AI approaches with traffic flow models, such as CTM and shockwaves, to address challenging issues in real-time traffic speed, travel time, flow prediction, and recurrent and nonrecurrent event detection, such as public events.

• Michael Cassidy, University of California, Berkley | Transportation Science with AI as Supplement Not Substitute: Two Illustrations

  Biography

  Dr. Michael Cassidy is the Robert Horonjeff Professor and Chancellor’s Professor of Civil and Environmental Engineering at the University of California, Berkeley. He received a doctorate in Civil Engineering (majoring in Transportation Engineering) from Berkeley; served for nearly four years as an assistant professor in the School of Civil Engineering at Purdue University in West Lafayette, Indiana; and joined the Berkeley faculty in 1994. He is an associate editor of the journal Transportation Research Part B; a member of the International Advisory Committee for the International Symposium on Transportation and Traffic Theory; and a former director of the University Transportation Center for US Federal Region 9. His research interests focus on transportation planning and management, particularly in the areas of highway traffic, public mass transit, and multi-modal systems.

  Abstract

  Two idealized case studies highlight how AI can affect urban travel, for better or worse. The first case study illustrates the pitfalls of allowing algorithms to generate solutions that are unchecked by domain knowledge and human expertise. The illustration pertains to Adaptive Cruise Control, or ACC, which automatically adjusts a vehicle’s speed in response to changing traffic conditions. Careful field experiments show that in congested traffic, cars operating with ACC adopt separations (i.e., spacings) with the vehicles ahead that are appreciably larger than those chosen by human drivers. The larger spacings mean that less-compacted queues will occupy greater road space in the future, when ACC is a ubiquitous feature of self-driving and human-driven vehicles alike. Simulations of various such futures portend disaster for cities. We discuss why a better future requires greater reliance on traffic science.

  The second illustration addresses the long-standing challenge of synchronizing traffic signals on grids of two-directional streets via application of domain knowledge and human creativity. The synchronization strategy produced in this fashion incorporates unconventional ideas. These include common timing patterns for all signals on a grid; prioritizing traffic movements pointed toward workplaces in the morning, and away from those places in the afternoon; and adaptively toggling neighboring signals between a forward progression mode and a second synchronization mode suitable for congestion. Simulations show that the strategy dramatically outperforms the state-of-the-practice computer program and can perform very well on irregular, real-world grids. We describe how AI-generated algorithms might enhance implementation of this strategy. We also discuss why it would be unrealistic to expect AI to uncover this sort of unconventional strategy in the absence of human innovation.
  

• Sean Qian, Carnegie Mellon University | Data-friendly Mesoscopic Network Modeling: Learning, Prediction, and Decision Making

  Biography

  Dr. Sean Qian is H. John Heinz III Professor of Civil and Environmental Engineering at Carnegie Mellon University. He is jointly appointed at the Department of Civil and Environmental Engineering, Heinz College of Information Systems and Public Policy, and Department of Electrical and Computer Engineering. He directs the Mobility Data Analytics Center (MAC) at CMU. In 2020, Qian founded a CMU technology spinoff firm, TraffiQure Technologies, to commercialize AI/ML technologies in the infrastructure and mobility service domain. Qian's research interest lies in large-scale dynamic network modeling and AI/ML applications for multi-modal transportation systems, in development of intelligent transportation systems (ITS) solutions, and in understanding infrastructure system interdependency. His research has been supported by a number of public agencies and private firms, such as NSF, U.S. DOE, U.S. DOT, PennDOT, Maryland SHA, IBM, Honda Research Institute, Fujitsu Research, Hitachi Rail, Benedum Foundation, and Hillman Foundation. Prof. Qian serves an associate editor for Transportation Research Part C: Emerging Technologies, Transportation Science, Transportmatrica B, and Journal of Public Transportation, and an editorial board editor for Transportation Research Part B: Methodological, and is an active member of the Network Modeling Committee and AI Committee of TRB. He is the recipient of the NSF CAREER award in 2018 and Greenshields Prize from the Transportation Research Board in 2017. Qian was a postdoctoral researcher in the Department of Civil and Environmental Engineering at Stanford University from 2011 to 2013, and received his PhD degree in Civil Engineering at the University of California, Davis in 2011 and his M.S. degree in Statistics at Stanford University in 2012.
  

  Abstract

  With the availability of various data sources across all modes of transportation systems, it remains a challenge how to take advantage of those diverse spatio-temporal data to best understand travel patterns across those modes in high spatio-temporal resolutions. In a mesoscopic network modeling framework, we formulate and solve for spatio-temporal passenger and vehicular flows in a multi-modal network explicitly considering solo-driving, public transit, parking, curb use, and ride-sharing. Vehicular flows, namely vehicles in different classifications, are integrated in a holistic dynamic network loading (DNL) model. We further develop a general formulation of heterogeneous flow in their respective choices of modes, facilities, and time. Through a computational graph approach, the travel behavior models and network characteristics can be jointly learned from a generic set of data, e.g., time-varying counts, speeds, census, transit data, and curb use data. Machine learning (ML) techniques are employed to optimally tune generic parameters to fit the multi-source data. This framework has been applied in many use cases for regions, cities, and communities to make optimal decisions in transportation planning. The mesoscopic modeling approach can also be applied to real-time traffic operations, particularly early anomaly detection, and proactive traffic management.

• Lijun Sun, McGill University, Montreal | Probabilistic Deep Learning for Spatiotemporal Mobility/Traffic Data

  Biography

  Dr. Lijun Sun is an associate professor and William Dawson Scholar in the Department of Civil Engineering at McGill University, Montreal (QC), Canada. He received his PhD in Civil Engineering (Transportation) from the National University of Singapore in 2015 and earned his Bachelor's degree in Civil Engineering from Tsinghua University (Beijing, China) in 2011. His research focuses on developing statistical and machine learning techniques, tools, and applications to address the efficiency, resilience, uncertainty, and sustainability issues in urban transportation systems. He is now leading the smart transportation lab at McGill and the current research program of the lab centers on modeling spatiotemporal urban mobility and traffic data, stochastic modeling of human driving behavior, probabilistic time series forecasting, Bayesian statistics, and tensor analysis. He serves as associate editor for Transportation Science and Transportation Research Part C–Emerging Technologies.

  Abstract

  Deep learning has driven remarkable progress in modeling traffic and mobility data, primarily through backpropagation-based learning, which enables the training of large-scale models. Recently, advances in probabilistic deep learning have underscored the importance of uncertainty quantification, showing that probabilistic approaches can yield more robust and interpretable solutions for many transport modeling tasks. In this talk, I will present studies that leverage deep learning with specifically designed probabilistic loss functions to parameterize various transport challenges. I will cover applications in traffic time series forecasting, bus travel time and trajectory prediction, network-wide travel time forecasting, and bike-sharing demand forecasting—all of which highlight the value of probabilistic models in spatiotemporal contexts.

• Alexandre Bayen, Berkeley Space Center, CITRIS, and Bantano Institute | Mixed Autonomy Traffic at Scale

  Biography

  Dr. Alexandre Bayen is the Associate Provost for the Berkeley Space Center, the Director of CITRIS and the Banatao Institute, and Liao-Cho Innovation Endowed Chair and professor of electrical engineering and computer sciences and of civil and environmental engineering at UC Berkeley. He received an Engineering Degree in applied mathematics from the Ecole Polytechnique, France, in 1998, and an MS and PhD in aeronautics and astronautics from Stanford University in 1999 and 2004, respectively. He was a visiting researcher at NASA Ames Research Center from 2000 to 2003. Between January 2004 and December 2004, he worked as the research director of the Autonomous Navigation Laboratory at the Laboratoire de Recherches Balistiques et Aerodynamiques, (Ministere de la Defense, Vernon, France), where he holds the rank of Major.

  Abstract

  This lecture will present the story of the MegaVanderTest, a test involving 100 self-driving vehicles, which ran the week of November 18, 2022, on I24 in Nashville, TN. The MegaVanderTest is the test that achieved the largest concentration of self-driving vehicles collaboratively controlling traffic on a single stretch of freeway in the history of self-driving vehicles. The talk will first cover the architecture built and deployed by the CIRCLES team. It will present some of the algorithms populating the planning layer of the system, mostly based on optimal control and imitation learning of Kernel-based expert controllers. It will also present some of the algorithms populating the local control regulation layer, based on deep-reinforcement learning and MPC. Finally, it will present the way the algorithms had to run in the field, due to the fact that most modern ACC architectures do not always open access to all sensed quantities collected by onboard instrumentation.
  

• Eugene Vinitsky, New York University | Fixing Evaluation of Mixed Human-autonomous Systems Using Reinforcement Learning

  Biography
  
  Dr. Eugene Vinitsky is an assistant professor of Civil and Urban Engineering and an affiliated professor of Computer Science Engineering. He received his PhD from UC Berkeley in Controls Engineering with a specialization in Reinforcement Learning after which he worked as a research scientist in the Apple Special Projects Group. His work focuses on automation and control in multi-agent systems with a specific focus in using reinforcement learning to improve performance in multi-agent transportation systems. He has also spent time at DeepMind, Facebook AI Research, and Tesla.

  Abstract

  Microsimulation is a key tool used to study transportation impacts. These microsimulators rely on low-dimensional behavioral models that do not always correctly reflect the potential impact of a new intelligent traffic control strategy. This can show up as a large simulation-to-reality gap that slows down the deployment of new intelligent congestion control strategies. We propose to improve on existing behavioral models by using abundant behavioral data to learn models of human behavior. Unfortunately, these imitation-based models, while reproducing qualitative behaviors, tend to have high crash rates that prevent their easy use in micro-simulators. We investigate whether the combination of imitation and reinforcement learning at scale can give us the best of both worlds: human-like behavior and human-level crash rates. We design a new simulator, GPUDrive, that allows us to scale RL training to billions of steps and show promising evidence that this approach can scale.
  

• Saif Jabari, NYU Abu Dhabi | Behind the Algorithms: An Exploration of the Vulnerabilities in Deep Neural Networks

  Biography
  
  
  Dr. Saif Jabari is an associate professor of Civil and Urban Engineering at NYUAD. His research focuses on developing advanced computational methods for urban traffic operations. These techniques integrate high-resolution traffic data with principles of traffic physics and artificial intelligence to address the rapidly evolving needs of the field. Specifically, his work leverages the potential of autonomous vehicle technologies to transition traffic management from reliance on aggregated metrics towards real-time, high-resolution, data-driven control.

  Abstract

  Autonomous vehicles (AVs) rely on deep neural networks (DNNs) for critical tasks such as environment perception — identifying traffic signs, pedestrians, and lane markings — and executing control decisions like braking, acceleration, and lane changing. However, DNNs are vulnerable to adversarial attacks, including structured perturbations to inputs and misleading training samples that can degrade performance. This presentation begins with an overview of adversarial training, emphasizing the impact of input sizes on DNNs' vulnerability to cyberattacks. Subsequently, I will share our recent findings that explore the hypothesis that DNNs learn piecewise linear relationships between inputs and outputs. This conjecture is crucial for developing both adversarial attacks and defense strategies in machine learning security.

• Takahiro Yabe, New York University | Predicting Human Mobility Patterns During Climate Disasters Using Behavior Data and AI

  Biography
  
  Dr. Takahiro Yabe is an assistant professor at New York University, where he directs the Resilient Urban Networks Lab. His research develops computational methods to understand collective social dynamics during disruptions such as disasters and pandemics. He was previously a postdoctoral associate at the MIT Institute for Data, Systems, and Society (IDSS) and Media Lab with Alex ‘Sandy’ Pentland. He received his PhD in Civil Engineering from Purdue University and Master's and Bachelor's Degrees from the University of Tokyo. He is a recipient of the Emerging Researcher Award from the Complex Systems Society.

  Abstract

  As climate change intensifies, cities face increasingly severe and unprecedented disaster events, making it challenging to rely on their own historical data for disaster planning. The lack of prior data has hindered our ability to build accurate urban simulation models that predict human behavior (e.g., where do people evacuate to, and when do they return?) and their socio-economic impacts on cities in future disaster scenarios. I will share our research progress on leveraging large-scale mobility data to unravel universal laws of recovery across cities and disasters, as well as computational modeling frameworks that enable accurate forecasts and sharing of human mobility behavior patterns in hypothetical disaster scenarios.