A Prompt Tuning Approach Based on Pre-trained Cross-City Graph Neural Networks for Zero-Shot Flow Generation
Abstract
We introduce a generalizable flow generation modeling approach that leverages an efficient pre-trained mobility model to learn generic mobility prior knowledge across different source cities and a light-weight graph prompt learning scheme to fine tune the pre-trained model for flow generation. We provide an implementation using the U.S. COVID-19 mobility flow dataset as an example.
Fields of Study & Subjects
Traffic Prediction and Management Techniques
Human Mobility and Location-Based Analysis
COVID-19 epidemiological studies
Publication details
- Journal / source
- Figshare
- Publication date
- 3 March 2026
- Language
- EN
- Document type
- Dataset