Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Semi-decentralized Training of Spatio-Temporal Graph Neural Networks for Traffic Prediction

Ivan Kralj, Lodovico Giaretta, Gordan Ježić, Ivana Podnar Žarko, Šarūnas Girdzijauskas

TL;DR

The paper tackles real-time traffic prediction over distributed sensor networks by introducing semi-decentralized training of Spatio-Temporal Graph Neural Networks across edge cloudlets. It develops a simulation framework and compares four training schemes—centralized, traditional FL, server-free FL, and Gossip Learning—on METR-LA and PeMS-BAY, demonstrating that semi-decentralized approaches achieve competitive accuracy with enhanced scalability and fault tolerance. Key findings note significant geographic variation in model performance and nontrivial communication costs due to the large receptive fields of ST-GNNs, yet the planar graph structure helps keep cloudlet costs bounded as the network scales. The work contributes a practical assessment of distributed ST-GNN training for smart mobility and highlights avenues for reducing overhead, personalizing cloudlets, and optimizing cloudlet deployment.

Abstract

In smart mobility, large networks of geographically distributed sensors produce vast amounts of high-frequency spatio-temporal data that must be processed in real time to avoid major disruptions. Traditional centralized approaches are increasingly unsuitable to this task, as they struggle to scale with expanding sensor networks, and reliability issues in central components can easily affect the whole deployment. To address these challenges, we explore and adapt semi-decentralized training techniques for Spatio-Temporal Graph Neural Networks (ST-GNNs) in smart mobility domain. We implement a simulation framework where sensors are grouped by proximity into multiple cloudlets, each handling a subgraph of the traffic graph, fetching node features from other cloudlets to train its own local ST-GNN model, and exchanging model updates with other cloudlets to ensure consistency, enhancing scalability and removing reliance on a centralized aggregator. We perform extensive comparative evaluation of four different ST-GNN training setups -- centralized, traditional FL, server-free FL, and Gossip Learning -- on large-scale traffic datasets, the METR-LA and PeMS-BAY datasets, for short-, mid-, and long-term vehicle speed predictions. Experimental results show that semi-decentralized setups are comparable to centralized approaches in performance metrics, while offering advantages in terms of scalability and fault tolerance. In addition, we highlight often overlooked issues in existing literature for distributed ST-GNNs, such as the variation in model performance across different geographical areas due to region-specific traffic patterns, and the significant communication overhead and computational costs that arise from the large receptive field of GNNs, leading to substantial data transfers and increased computation of partial embeddings.

Semi-decentralized Training of Spatio-Temporal Graph Neural Networks for Traffic Prediction

TL;DR

The paper tackles real-time traffic prediction over distributed sensor networks by introducing semi-decentralized training of Spatio-Temporal Graph Neural Networks across edge cloudlets. It develops a simulation framework and compares four training schemes—centralized, traditional FL, server-free FL, and Gossip Learning—on METR-LA and PeMS-BAY, demonstrating that semi-decentralized approaches achieve competitive accuracy with enhanced scalability and fault tolerance. Key findings note significant geographic variation in model performance and nontrivial communication costs due to the large receptive fields of ST-GNNs, yet the planar graph structure helps keep cloudlet costs bounded as the network scales. The work contributes a practical assessment of distributed ST-GNN training for smart mobility and highlights avenues for reducing overhead, personalizing cloudlets, and optimizing cloudlet deployment.

Abstract

In smart mobility, large networks of geographically distributed sensors produce vast amounts of high-frequency spatio-temporal data that must be processed in real time to avoid major disruptions. Traditional centralized approaches are increasingly unsuitable to this task, as they struggle to scale with expanding sensor networks, and reliability issues in central components can easily affect the whole deployment. To address these challenges, we explore and adapt semi-decentralized training techniques for Spatio-Temporal Graph Neural Networks (ST-GNNs) in smart mobility domain. We implement a simulation framework where sensors are grouped by proximity into multiple cloudlets, each handling a subgraph of the traffic graph, fetching node features from other cloudlets to train its own local ST-GNN model, and exchanging model updates with other cloudlets to ensure consistency, enhancing scalability and removing reliance on a centralized aggregator. We perform extensive comparative evaluation of four different ST-GNN training setups -- centralized, traditional FL, server-free FL, and Gossip Learning -- on large-scale traffic datasets, the METR-LA and PeMS-BAY datasets, for short-, mid-, and long-term vehicle speed predictions. Experimental results show that semi-decentralized setups are comparable to centralized approaches in performance metrics, while offering advantages in terms of scalability and fault tolerance. In addition, we highlight often overlooked issues in existing literature for distributed ST-GNNs, such as the variation in model performance across different geographical areas due to region-specific traffic patterns, and the significant communication overhead and computational costs that arise from the large receptive field of GNNs, leading to substantial data transfers and increased computation of partial embeddings.

Paper Structure

This paper contains 29 sections, 1 equation, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Traffic prediction
  4. Spatio-Temporal Graph Neural Networks
  5. Traditional Federated Learning
  6. Server-free Federated Learning
  7. Gossip Learning
  8. Distributed training
  9. Methodology
  10. Scenario
  11. Distributed Training Setups
  12. GNN Partitioning in Distributed Training
  13. Experimental Setup
  14. Datasets
  15. Evaluation Metrics
  16. ...and 14 more sections

Figures (4)

  • Figure 1: Graph partitioning and communication. a) Geographically distributed sensor network and base stations b) Graph partitioning of the sensors into cloudlets based on geographical proximity. c) Cloudlet-to-cloudlet communication network for exchanging node features and model updates. d) After communicating with neighbouring cloudlets, each cloudlet can construct the ST-GNN subgraph required for training on its local nodes
  • Figure 2: Sensor assignment to cloudlets based on communication range
  • Figure 3: WMAPE for individual cloudlets
  • Figure 4: Validation loss over FLOPs and epochs for short-term prediction