Cross-City Transfer Learning for Deep Spatio-Temporal Prediction

Abstract

Spatio-temporal prediction is a key type of tasks in urban computing, e.g., traffic flow and air quality. Adequate data is usually a prerequisite, especially when deep learning is adopted. However, the development levels of different cities are unbalanced, and still many cities suffer from data scarcity. To address the problem, we propose a novel cross-city transfer learning method for deep spatiotemporal prediction tasks, called RegionTrans. RegionTrans aims to effectively transfer knowledge from a data-rich source city to a data-scarce target city. More specifically, we first learn an inter-city region matching function to match each target city region to a similar source city region. A neural network is designed to effectively extract region-level representation for spatio-temporal prediction. Finally, an optimization algorithm is proposed to transfer learned features from the source city to the target city with the region matching function. Using citywide crowd flow prediction as a demonstration experiment, we verify the effectiveness of RegionTrans. Results show that RegionTrans can outperform the state-of-theart fine-tuning deep spatio-temporal prediction models by reducing up to 10.7% prediction error. * Equal contribution Preprint. Work in progress.In literature, existing deep learning approaches are often designed to predict citywide phenomenon as a whole [18,19], and thus it is hard to enable region-level knowledge transfer. To this end, rather than adopting the existing deep neural networks for citywide spatio-temporal prediction (e.g. ), we propose a novel deep transfer learning method. First, we design a region matching function to link each target city region to a similar source region based on the short period of service data or correlated auxiliary data if applicable. Then, in our proposed network structure, to catch the spatio-temporal patterns hidden in the service data, ConvLSTM layers [12] are firstly stacked. Afterward, to encode region representation, we newly add a Conv2D layer with 1 × 1 filter, which is the key and fundamental component of our network to make region-level transfer feasible. Finally, the discrepancy between region representations of the inter-city similar regions is minimized during the network parameter learning for the target city, so as to enable region-level cross-city knowledge transfer. With crowd flow prediction as a showcase [18,19], we verify the feasibility and effectiveness of RegionTrans. Briefly, this paper has the following contributions.(i) To the best of our knowledge, this is the first work to study how to facilitate deep spatio-temporal prediction in a data-scarce target city by transferring knowledge from a data-rich source city.(ii) We propose a novel deep transfer learning method RegionTrans for spatio-temporal prediction tasks by region-level cross-city transfer. RegionTrans first computes inter-city region similarities, and then stacks ConvLSTM and Conv2D (1 × 1 filter) layers to extract region-level representations reflecting spatio-temporal pattern...

Knowledge Adaption for Demand Prediction based on Multi-task Memory Neural Network

“…Due to the insufficiency of sources and external data, some works utilized developed previously learning methods to adapt the knowledge to new tasks or domains for prediction [20,24,27]. An intercity region matching function was learned by Wang et al [20] to match two similar regions from the source domain to the target domain for crowd flow prediction.…”

“…Due to the insufficiency of sources and external data, some works utilized developed previously learning methods to adapt the knowledge to new tasks or domains for prediction [20,24,27]. An intercity region matching function was learned by Wang et al [20] to match two similar regions from the source domain to the target domain for crowd flow prediction. For air quality prediction, a Flexible multi-modal transfer Learning (FLORAL) method was proposed by Wei et al [24] through learning semantically related dictionaries from the source domain and adapting it to the target domain.…”

Smart City Development With Urban Transfer Learning

“…As a concrete example, suppose we want to build inter-city similarity on crowd flow dynamics but we cannot find enough crowd flow historical records for some cities, then we may rely on social media check-ins to construct the inter-city similarity, while the inter-modality correlation of check-in and crowd flow is actually learned from the cities with both rich historical check-in and crowd flow records. In such a way, we can transfer the crowd flow prediction model from one source city to a target new city (cross-city) with the help of check-ins as the proxy (cross-modality) [15]. Later in this article, we will illustrate several urban transfer learning applications, and readers will see that most adopt both cross-modality and cross-city transfer.…”

“…As deep leaning has become the state-of-the-art solution for crowd flow prediction [2], our focus is a deep transfer learning mechanism. Figure 3 (a) shows the overall design of our mechanism RegionTrans [15]. Briefly, the applicability of RegionTrans lies in the fact that there are usually similar regions between cities (e.g., CBD), and thus the crowd flow patterns of such inter-city similar regions can be transferred.…”

“…Social media check-ins are collected from Foursquare [13]. The results show that, compared to the state-of-the-art methods only considering the target city data, RegionTrans can reduce the prediction error by up to 26% [15].…”