One-shot Transfer Learning for Population Mapping

Fine-grained population distribution data is of great importance for many applications, e.g., urban planning, traffic scheduling, epidemic modeling, and risk control. However, due to the limitations of data collection, including infrastructure density, user privacy, and business security, such fine-grained data is hard to collect and usually, only coarse-grained data is available. Thus, obtaining fine-grained population distribution from coarse-grained distribution becomes an important problem. To complete this task, existing methods mainly rely on sufficient fine-grained ground truth for training, which is not often available. This limits the applications of these methods and brings the necessity to transfer knowledge from data-sufficient cities to data-scarce cities. In knowledge transfer scenario, we employ single reference fine-grained ground truth in the target city as the ground truth to inform the large-scale urban structure and support the knowledge transfer in the target city. By this approach, we transform the fine-grained population mapping problem into a one-shot transfer learning problem for population mapping task. In this paper, we propose a one-shot transfer learning framework, PSRNet, to transfer spatial-temporal knowledge across cities in fine-grained population mapping task from the view of network structure, data, and optimization. Experiments on real-life datasets of 4 cities demonstrate that PSRNet has significant advantages over 8 baselines by reducing RMSE and MAE for more than 25%. Our code and datasets are released in Github.