Talk Information

• Topic: Modeling Urban Traffic Data with Matrix and Tensor Factorization Approaches
• Date: October 21, 2024
• 2024 INFORMS Annual Meeting, Seattle, USA
• Session format: Invited Session
• Time: 10:45 AM - 12:00 PM

Talk Summary

In urban systems, large amounts of human movement data, such as urban mobility and traffic flow, are readily available for implementing downstream tasks and supporting decision-making. Since these data are characterized by spatiotemporal dimensions and demonstrate data patterns of the systems, it is important to reformulate urban system problems with machine learning. In this talk, we introduce two scentific problems: 1) imputing missing values from partially observed traffic data, and 2) discovering spatial and temporal patterns from time-varying urban systems. Since these spatiotemporal data always (empirically) reveal low-rank properties and spatiotemporal correlations, we start from a sequence of low-rank matrix and tensor methods such as matrix/tensor factorization. To reinforce the model for building temporal correlations and dynamics, we integrate time series convolution and autoregression into the framework. In the modeling process, we intend to highlight the importance of temporal modeling in the low-rank matrix and tensor methods.