[250225] New publication in Transportation Science

I'm thrilled to share that our paper, 'Scalable Dynamic Mixture Model with Full Covariance for Probabilistic Traffic Forecasting,' has been accepted for publication in Transportation Science!

I’m thrilled to share that our paper, “Scalable Dynamic Mixture Model with Full Covariance for Probabilistic Traffic Forecasting,” has been accepted for publication in Transportation Science!

This work was part of my postdoctoral research at McGill University in collaboration with Professor Nicolas Saunier, Mr. Vincent Zheng, Professor Martin Trépanier, and Professor Lijun Sun.

In this study, we introduce a Dynamic Mixture Model with a Full Spatiotemporal Covariance Matrix to better capture the time-varying distribution of spatiotemporal characteristics in network traffic data.

To overcome the limitations of conventional MSE/MAE-based training methods, which assume an independent and isotropic Gaussian distribution, we developed a Mixture Density Network-based training method with time-varying mixture weights. Additionally, we estimate the full spatial and temporal covariance matrix during model training—while ensuring scalability to handle large spatiotemporal datasets efficiently.

Preprint available at: https://lnkd.in/edDfXs5y