I’m a third-year Ph.D. student at the Mobility Data Analytics Center (MAC) and Causal Learning and Reasoning Group (CLeaR) of Carnegie Mellon University, under the supervision of Dr. Sean Qian and Dr. Kun Zhang. My research interest is mainly in representation learning (causal discovery and nonlinear dynamics embedding), spatiotemporal forecasting, and some social media analytics. I’m currently developing causality-guided AI/ML techniques for modeling and optimization of large-scale dynamical systems with intervention-aware approaches.

In May 2020, I graduated from CMU with M.S. in Machine Learning. I have a Bachelor degree in Aerospace Engineering 🚀 from Beihang University in China. I can be reached at weiran[AT]cmu[DOT]edu.

Research Projects

Hidden Causal Representation Learning

This project aims to recover time-delayed latent causal variables and identify temporally causal latent processes from measured temporal data. Most existing work either focuses on estimating the causal relations between observed variables, or starts from the premise that causal variables are given beforehand. Real-world observations (e.g., image pixels, sensor measurements, etc.), however, are not structured into causal variables to begin with. This project proposes two general conditions with which the latent temporally causal processes can be identified and develop a theoretically-grounded training framework that enforces the assumed conditions through proper constraints.

Fig.1 - We propose LEAP -- a Latent tEmporally cAusal Processes estimation framework built upon the framework of VAEs while enforcing the conditions as constraints for identification of the latent causal processes.

Predictive Real-time Traffic Management in Large-Scale Networks Using Model-based AI

This project proposes to develop theories, models and algorithms of Artificial Intelligence (AI) guided by transportation network flow models, to achieve two main goals: to predict non-recurrent traffic conditions in large-scale networks at least 30 minutes ahead, and to proactively recommend operational management strategies in real-time. The prediction will be made by a machine that learns not only historical multi-source traffic data but also considers operational strategies that are currently being or to be recommended/engaged (called intervention-aware prediction). Operational strategies are made and updated in real time using model-based AI with the ahead-of-curve prediction.

Fig.2 - We propose PEACE (Predict traffic, Early detection of traffic anomalies, Approximate traffic flow physics, Control traffic and Estimate network benefits) for holistic proactive traffic incident management.

User-Centric Interdependent Urban Systems

The complex nature of interrelationships among various urban systems is central to smart cities. There may exist clear spatial and temporal correlations among usage patterns of all urban systems. The objective of this research is to fuse and analyze massive data from transportation, energy, and social media systems to discover the spatio-temporal correlations of usage patterns among those systems.