[Paper Review] Federated Learning in Intelligent Transportation Systems: Recent Applications and Open Problems

Shiying Zhang, Jun Li+6 more|arXiv (Cornell University)|Sep 20, 2023

Privacy-Preserving Technologies in Data4 citations

TL;DR

This paper proposes federated learning (FL) as a privacy-preserving, scalable solution for intelligent transportation systems (ITS), addressing challenges like data heterogeneity, limited resources, and communication latency. It surveys FL applications in object detection, traffic management, and service provisioning, identifies open problems such as non-IID data and edge trust, and advocates for deeper FL-ITS integration with optimized architectures and edge verification mechanisms.

ABSTRACT

Intelligent transportation systems (ITSs) have been fueled by the rapid development of communication technologies, sensor technologies, and the Internet of Things (IoT). Nonetheless, due to the dynamic characteristics of the vehicle networks, it is rather challenging to make timely and accurate decisions of vehicle behaviors. Moreover, in the presence of mobile wireless communications, the privacy and security of vehicle information are at constant risk. In this context, a new paradigm is urgently needed for various applications in dynamic vehicle environments. As a distributed machine learning technology, federated learning (FL) has received extensive attention due to its outstanding privacy protection properties and easy scalability. We conduct a comprehensive survey of the latest developments in FL for ITS. Specifically, we initially research the prevalent challenges in ITS and elucidate the motivations for applying FL from various perspectives. Subsequently, we review existing deployments of FL in ITS across various scenarios, and discuss specific potential issues in object recognition, traffic management, and service providing scenarios. Furthermore, we conduct a further analysis of the new challenges introduced by FL deployment and the inherent limitations that FL alone cannot fully address, including uneven data distribution, limited storage and computing power, and potential privacy and security concerns. We then examine the existing collaborative technologies that can help mitigate these challenges. Lastly, we discuss the open challenges that remain to be addressed in applying FL in ITS and propose several future research directions.

Motivation & Objective

• Address the urgent need for real-time, privacy-preserving decision-making in dynamic, data-intensive intelligent transportation systems (ITS).
• Identify key challenges in ITS, including high latency, data heterogeneity, limited vehicle resources, and privacy risks from centralized AI models.
• Survey recent FL deployments across ITS scenarios such as object recognition, traffic flow prediction, and service provisioning.
• Analyze inherent limitations of FL in ITS, including non-IID data, edge trust, and communication overhead, and propose mitigation strategies.
• Outline future research directions for deep integration of FL with ITS, including custom FL architectures and secure edge computation.

Proposed method

• Conduct a comprehensive survey of FL applications in ITS across four main scenarios: object recognition, traffic management, service provisioning, and route planning.
• Analyze system-level challenges in ITS, including unstable communication links, sensor data heterogeneity, and resource-constrained vehicles.
• Evaluate FL’s ability to mitigate privacy and security risks by keeping data local and enabling decentralized model training.
• Propose integration of blockchain and UAV-aided resource scheduling to enhance FL security and reliability in dynamic environments.
• Introduce lightweight verification mechanisms for edge nodes (e.g., RSUs) to ensure trustworthiness of offloaded computations.
• Explore model compression techniques like pruning and quantization to reduce transmission overhead and improve communication efficiency.

Experimental results

Research questions

• RQ1 How can federated learning effectively address privacy and security concerns in data-intensive ITS applications?
• RQ2 What are the key system-level challenges in deploying FL within dynamic, high-mobility vehicular networks?
• RQ3 How do data heterogeneity (non-IID distribution) and limited device resources impact FL performance in ITS?
• RQ4 In what ways can edge nodes like RSUs be secured and made trustworthy in FL-based ITS architectures?
• RQ5 What architectural and protocol-level innovations are needed to enable deep integration of FL with ITS for real-time, scalable operation?

Key findings

• FL significantly enhances data privacy in ITS by enabling decentralized model training without sharing raw vehicle data.
• Non-IID data distribution across vehicles remains a major challenge, degrading FL convergence and model accuracy.
• Limited storage and computing capabilities in vehicles necessitate model compression and edge offloading strategies.
• Unstable communication links and frequent topology changes in vehicular networks increase latency and disrupt FL training cycles.
• Edge nodes such as RSUs pose security risks due to potential data leakage and untrusted model updates, requiring lightweight authentication mechanisms.
• Future integration of FL with ITS should focus on custom architectures tailored to sensor modalities and task-specific workloads to maximize performance and efficiency.