Zhichen Liu
PhD candidate
University of Michigan
Zhichen Liu is a PhD candidate in Civil Engineering at the University of Michigan, specializing in AI-enabled transportation network analysis under the supervision of Professor Yafeng Yin. Her research aims to reshape the transportation network equilibrium modeling with emerging multi-source data and machine learning techniques, ultimately supporting scalable and robust decision-making in connected transportation systems. Her works have been published in leading journals and she has been invited to speak at NeurIPS, INFORMS, and the Transportation Research Board Annual Meeting
As an active academic community member, Zhichen serves as a reviewer for top-tier journals and TRB Standing Committee on Freight Transportation Planning and Logistics. She also serves on the board of the Michigan Transportation Student Organization. Her efforts have been acknowledged with an Engineering Graduate Fellowship from the University of Michigan and the prestigious China National Scholarship.
Transportation network equilibrium analysis enables policy-makers to diagnose and treat recurrent congestion and lays the foundation for transportation planning and operation. The traditional equilibrium analysis framework was initiated in the 1950s and limited by the lack of travel data. My research seeks to revolutionize the way we analyze transportation networks by developing an end-to-end framework that directly builds transportation network equilibrium models from multi-source high-volume data and supports scalable and robust decision-making in transportation systems. I use a combination of game theory, machine learning, and optimization theory to achieve this goal.
One of my published papers [1] is the first to integrate the learning of travel behaviors into an end-to-end equilibrium analysis framework, with deep neural networks automatically discovering route choice preferences from empirical data. A later paper [2] extends the end-toend framework to learn both supply-side and demand-side components. This novel framework advances the traditional method by unifying the use of model-based and learning-based approaches, as well as offering more flexibility for handling the complexities of real-world transportation systems.
My latest research [2] investigates applying the proposed framework to the real world. To address the scalability and robustness challenges, I adopt recent developments in differential optimization to develop scalable approximation algorithms. The proposed algorithm is robust with theoretical guarantees and is being validated with real-world connected vehicle trajectory data from Ann Arbor, Michigan.
Moving forward, I aim to expand my work into offering data-informed decisions for all participants in a connected transportation system, including autonomous vehicles [3] and mobility service providers [4]. By reshaping the transportation network equilibrium analysis with emerging multi-source data and learning techniques, I believe my research will help to understand and better manage the complex cyber-physical transportation system.
References
[1] Liu, Z., Yin, Y., Bai, F., Grimm, D.K. 2023. ‘End-to-end learning of user equilibrium with implicit neural networks’. Transportation Research Part C: Emerging Technologies.
[2] Liu, Z., Yin, Y., Bai, F., Grimm, D.K.2023. ‘A Unified Framework for End-to-End Transportation Network Equilibrium Modeling’. International Symposium on Transportation Data & Modelling.
[3] Liu, Z., Yin, Y. ‘End-to-end Game-Theoretic Planning for Autonomous Vehicles’. Working paper.
[4] Liu, Z., Vignon, D., Yin, Y. ‘Reducing Emission in Ridesourcing System with Differential matching’. Working paper.
