Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

COLA: Cross-city Mobility Transformer for Human Trajectory Simulation

Yu Wang, Tongya Zheng, Yuxuan Liang, Shunyu Liu, Mingli Song

TL;DR

COLA tackles cross-city human trajectory simulation under data scarcity and privacy constraints by transferring universal mobility patterns through a Half-open Transformer that separates private city-specific and shared city-universal modules. The model-agnostic transfer framework follows a meta-learning paradigm with Meta Clone, Internal Update, and Meta Update across source cities, initializing the target city model from a meta shared backbone and finishing with Post-hoc Adjustment to align predictions with the target city’s long-tail distribution. Empirical results across four real-world cities show COLA outperforms state-of-the-art single-city and cross-city baselines on multiple distributional and preference-oriented metrics, with ablations confirming the value of half-open attention and post-hoc calibration. The work demonstrates practical applicability in privacy-preserving data synthesis, data augmentation, and epidemic modeling, offering a scalable approach to generate realistic mobility data while respecting city-specific characteristics.

Abstract

Human trajectory data produced by daily mobile devices has proven its usefulness in various substantial fields such as urban planning and epidemic prevention. In terms of the individual privacy concern, human trajectory simulation has attracted increasing attention from researchers, targeting at offering numerous realistic mobility data for downstream tasks. Nevertheless, the prevalent issue of data scarcity undoubtedly degrades the reliability of existing deep learning models. In this paper, we are motivated to explore the intriguing problem of mobility transfer across cities, grasping the universal patterns of human trajectories to augment the powerful Transformer with external mobility data. There are two crucial challenges arising in the knowledge transfer across cities: 1) how to transfer the Transformer to adapt for domain heterogeneity; 2) how to calibrate the Transformer to adapt for subtly different long-tail frequency distributions of locations. To address these challenges, we have tailored a Cross-city mObiLity trAnsformer (COLA) with a dedicated model-agnostic transfer framework by effectively transferring cross-city knowledge for human trajectory simulation. Firstly, COLA divides the Transformer into the private modules for city-specific characteristics and the shared modules for city-universal mobility patterns. Secondly, COLA leverages a lightweight yet effective post-hoc adjustment strategy for trajectory simulation, without disturbing the complex bi-level optimization of model-agnostic knowledge transfer. Extensive experiments of COLA compared to state-of-the-art single-city baselines and our implemented cross-city baselines have demonstrated its superiority and effectiveness. The code is available at https://github.com/Star607/Cross-city-Mobility-Transformer.

COLA: Cross-city Mobility Transformer for Human Trajectory Simulation

TL;DR

COLA tackles cross-city human trajectory simulation under data scarcity and privacy constraints by transferring universal mobility patterns through a Half-open Transformer that separates private city-specific and shared city-universal modules. The model-agnostic transfer framework follows a meta-learning paradigm with Meta Clone, Internal Update, and Meta Update across source cities, initializing the target city model from a meta shared backbone and finishing with Post-hoc Adjustment to align predictions with the target city’s long-tail distribution. Empirical results across four real-world cities show COLA outperforms state-of-the-art single-city and cross-city baselines on multiple distributional and preference-oriented metrics, with ablations confirming the value of half-open attention and post-hoc calibration. The work demonstrates practical applicability in privacy-preserving data synthesis, data augmentation, and epidemic modeling, offering a scalable approach to generate realistic mobility data while respecting city-specific characteristics.

Abstract

Human trajectory data produced by daily mobile devices has proven its usefulness in various substantial fields such as urban planning and epidemic prevention. In terms of the individual privacy concern, human trajectory simulation has attracted increasing attention from researchers, targeting at offering numerous realistic mobility data for downstream tasks. Nevertheless, the prevalent issue of data scarcity undoubtedly degrades the reliability of existing deep learning models. In this paper, we are motivated to explore the intriguing problem of mobility transfer across cities, grasping the universal patterns of human trajectories to augment the powerful Transformer with external mobility data. There are two crucial challenges arising in the knowledge transfer across cities: 1) how to transfer the Transformer to adapt for domain heterogeneity; 2) how to calibrate the Transformer to adapt for subtly different long-tail frequency distributions of locations. To address these challenges, we have tailored a Cross-city mObiLity trAnsformer (COLA) with a dedicated model-agnostic transfer framework by effectively transferring cross-city knowledge for human trajectory simulation. Firstly, COLA divides the Transformer into the private modules for city-specific characteristics and the shared modules for city-universal mobility patterns. Secondly, COLA leverages a lightweight yet effective post-hoc adjustment strategy for trajectory simulation, without disturbing the complex bi-level optimization of model-agnostic knowledge transfer. Extensive experiments of COLA compared to state-of-the-art single-city baselines and our implemented cross-city baselines have demonstrated its superiority and effectiveness. The code is available at https://github.com/Star607/Cross-city-Mobility-Transformer.
Paper Structure (28 sections, 1 theorem, 13 equations, 19 figures, 4 tables, 1 algorithm)

This paper contains 28 sections, 1 theorem, 13 equations, 19 figures, 4 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Overall Framework of COLA
  5. Half-open Transformer
  6. Post-hoc Adjustment
  7. Experiment
  8. Experimental Setup
  9. Datasets.
  10. Baselines.
  11. Evaluation Metrics.
  12. Experiment Settings.
  13. RQ1: Overall Performance
  14. RQ2: Ablation Studies
  15. Performance across different cities.
  16. ...and 13 more sections

Key Result

Proposition 1

Suppose that the probability density function of locations follows Zipf's law $\pi(x) \sim ax^{-\gamma}, \gamma > 0$, $x \in \mathbb{N}^{+}$ is the index of a location, the post-hoc adjustment dynamically scales the pair-wise probabilities of two locations as: $\frac{\tilde{\mathbf{y}}_{t+1}^{m,i}}{

Figures (19)

  • Figure 1: Motivation and challenges of human trajectory simulation across cities.
  • Figure 2: The overall framework of COLA. (i) Initialize the shared parameters of the source model with the meta model. (ii) Optimize the source model with its internal loss. (iii) Update the meta model based on the gradient evaluated on the source city. (iv) Initialize the shared parameters of the target model with the meta model updated by all source cities. (v) Optimize the target model with its internal loss. (vi) Simulate human trajectories with Post-hoc Adjustment technique.
  • Figure 3: Half-open attention in the Transformer.
  • Figure 4: The performance comparison of COLA on GeoLife with different combinations of source cities. All experimental results are conducted over five trials for a fair comparison.
  • Figure 5: Location prediction in the fully simulated scenario.
  • ...and 14 more figures

Theorems & Definitions (1)

  • Proposition 1