[Paper Review] Secure and Efficient Blockchain based Knowledge Sharing for Intelligent Connected Vehicles

Haoye Chai, Supeng Leng+2 more|arXiv (Cornell University)|Aug 3, 2021

Blockchain Technology Applications and Security4 citations

TL;DR

This paper proposes a secure, efficient DAG blockchain-based knowledge sharing framework for Intelligent Connected Vehicles (ICVs), enabling cross-regional, micro-transaction-driven sharing of machine learning models for autonomous driving. It introduces a lightweight RTH-Tip Selection Algorithm (RTH-TSA) for consensus and identity verification, and an adaptive asynchronous distributed learning (ADL) scheme that reduces communication overhead while improving model accuracy, achieving a 20% reduction in driving violation rates and 30% lower loss under malicious attacks compared to baseline methods.

ABSTRACT

The emergence of Intelligent Connected Vehicles (ICVs) shows great potential for future intelligent traffic systems, enhancing both traffic safety and road efficiency. However, the ICVs relying on data driven perception and driving models face many challenges, including the lack of comprehensive knowledge to deal with complicated driving context. In this paper, we are motivated to investigate cooperative knowledge sharing for ICVs. We propose a secure and efficient directed acyclic graph (DAG) blockchain based knowledge sharing framework, aiming to cater for the micro-transaction based vehicular networks. The framework can realize both local and cross-regional knowledge sharing. Then, the framework is applied to autonomous driving applications, wherein machine learning based models for autonomous driving control can be shared. A lightweight tip selection algorithm (TSA) is proposed for the DAG based knowledge sharing framework to achieve consensus and identity verification for cross-regional vehicles. To enhance model accuracy as well as minimizing bandwidth consumption, an adaptive asynchronous distributed learning (ADL) based scheme is proposed for model uploading and downloading. Experiment results show that the blockchain based knowledge sharing is secure, and it can resist attacks from malicious users. In addition, the proposed adaptive ADL scheme can enhance driving safety related performance compared to several existing algorithms.

Motivation & Objective

• To address the lack of comprehensive knowledge and security vulnerabilities in data-driven ICVs facing complex driving scenarios.
• To overcome the limitations of centralized management and synchronous learning in dynamic vehicular networks.
• To enable secure, decentralized, and efficient knowledge sharing among ICVs using blockchain and asynchronous learning.
• To reduce communication cost and improve model accuracy through adaptive model aggregation based on driving behavior indicators.
• To ensure resilience against malicious users by integrating blockchain-based verification and consensus mechanisms.

Proposed method

• Design a DAG blockchain framework where each transaction (site) represents a micro-transaction of knowledge, eliminating mining overhead and enabling lightweight consensus.
• Propose a RTH-Tip Selection Algorithm (RTH-TSA) that uses a reputation and trust-based heuristic to select tips for verification, enabling fast cross-regional identity authentication.
• Integrate a lightweight consensus mechanism based on the DAG structure, where new sites verify previous ones, enhancing scalability and real-time performance.
• Develop an adaptive asynchronous distributed learning (ADL) scheme that dynamically weights local models based on freshness and driving behavior indicators to improve global model accuracy.
• Apply a verification rule $ e \leq \epsilon $ to filter out malicious knowledge, ensuring only legitimate models are appended to the ledger.
• Use a driving style indicator to identify and selectively include or exclude ICVs with deviant behaviors, minimizing communication cost while preserving model quality.

Experimental results

Research questions

• RQ1 How can a secure and efficient knowledge sharing framework be designed for ICVs in highly dynamic, decentralized vehicular networks?
• RQ2 What mechanisms can ensure consensus and identity verification in cross-regional, micro-transaction-based knowledge sharing without relying on mining?
• RQ3 How can asynchronous distributed learning be adapted to reduce communication overhead while maintaining or improving model accuracy in ICV environments?
• RQ4 To what extent can model freshness and driving behavior indicators improve the quality of aggregated knowledge in ICV learning systems?
• RQ5 How effective is the proposed framework in resisting malicious attacks, such as injection of biased or false knowledge?

Key findings

• The proposed ADL scheme achieved a driving violation rate reduction of over 20% compared to FedAve and cook-AD algorithms, demonstrating improved safety performance.
• The model loss of the proposed ADL was very close to the centralized training baseline, indicating high global model accuracy despite asynchronous updates.
• The framework reduced the loss function value by an average of 30% under malicious attacks compared to a baseline system, proving strong resilience.
• Excluding ICVs with highly deviant driving behavior indicators (e.g., $ m_3 $-type) reduced communication cost without significantly degrading model quality, as long as some such models were retained.
• The RTH-TSA enabled fast consensus and identity verification across regions, supporting real-time knowledge sharing in dynamic vehicular networks.
• The adaptive ADL scheme converged faster than synchronous methods due to on-the-fly model aggregation, even with heterogeneous and asynchronous ICV contributions.