Survey of Federated Learning Models for Spatial-Temporal Mobility Applications

TL;DR

This survey analyzes Federated Learning (FL) for spatial-temporal mobility, highlighting privacy-preserving contributions across trajectory prediction, traffic flow, and POI-based tasks. It compares FL approaches to centralized baselines, discusses heterogeneity and personalization, and catalogs datasets, metrics, and evaluation practices. The authors propose open-source resources and a road map addressing semantic context, Byzantine resilience, communication efficiency, fairness, reproducibility, and deployment frameworks. The study underscores the rapid growth of FL in mobility applications and aims to guide researchers and practitioners toward standardized benchmarks and practical deployments with privacy guarantees. Overall, the work maps the landscape of FL for ST mobility and outlines concrete directions to close gaps between theory and real-world deployment.

Abstract

Federated learning involves training statistical models over edge devices such as mobile phones such that the training data is kept local. Federated Learning (FL) can serve as an ideal candidate for training spatial temporal models that rely on heterogeneous and potentially massive numbers of participants while preserving the privacy of highly sensitive location data. However, there are unique challenges involved with transitioning existing spatial temporal models to decentralized learning. In this survey paper, we review the existing literature that has proposed FL-based models for predicting human mobility, traffic prediction, community detection, location-based recommendation systems, and other spatial-temporal tasks. We describe the metrics and datasets these works have been using and create a baseline of these approaches in comparison to the centralized settings. Finally, we discuss the challenges of applying spatial-temporal models in a decentralized setting and by highlighting the gaps in the literature we provide a road map and opportunities for the research community.

Figures (5)

• Figure 1: Number of FL Surveys in comparison with the growing number of spatial-temporal FL methodological papers. The shaded areas present the percentage of the surveys that are application specific.
• Figure 2: Federated Learning general workflow. On the left, a cross-device setting where each client is a device (, smartphone) and on the right is cross-silo setting where each client represents an organization. The main difference between the two is the magnitude both in terms of clients and local data.
• Figure 3: Entropy Experiments.
• Figure 4: Figure from FedTSE yuan2022fedtse showing the interactions between Edge Computing (EC) Server and Road Side Units (RSU) acting as a cross-silo unit. The LSTM model weights are aggregated using FedAvg, and a Deep Reinforcement Learning (DRL) Agent to maximize reward.
• Figure 5: Figure from CNGNN meng2021cross presents (a) the server-side graph neural network with a systematic overview of training steps: (1) Federated learning of on-node models. (2) Temporal encoding update. (3) Split Learning of GN. (b) Client $i$ Auto-encoder architecture.