Date of Defense
21-5-2026 11:00 AM
Location
H1-0057
Document Type
Dissertation Defense
Degree Name
Doctor of Philosophy in Informatics and Computing
College
CIT
Department
Computer Science and Software Engineering
First Advisor
Dr. Moulay Elarbi Badidi
Keywords
Adaptive rewards, client selection, incentive mechanism, fairness, VFL.
Abstract
Vehicular Federated Learning (VFL) is becoming a crucial enabler for the implementation and optimization of automated transport systems. This technology enables intelligent transportation systems to function by enabling networked vehicles to develop perception and control models through joint training while preserving their original data. Nevertheless, the effective deployment of VFL is dependent on the sustained participation of trustworthy vehicles. However, the dynamic nature of vehicular environments poses critical challenges, including unstable participation, data heterogeneity, and resource constraints. The sustained collaboration of smart vehicles necessitates reliable client selection and equitable incentive schemes with verifiable transparency. The distributed nature of VFL makes it vulnerable to malicious client updates that can compromise model integrity. Moreover, conventional FL frameworks lack effective client selection strategies and fair and adaptive incentive mechanisms that are capable of maintaining fairness and reliability under these fluctuating conditions. Additionally, most existing incentive schemes adopt basic contribution measures or the Shapley value method to determine reward distribution. These incur high computational cost and provide inadequate definitions of fairness. Moreover, these do not guarantee that vehicles with higher contributions receive higher rewards, which leads to fairness disparities that could undermine long-term incentives for participation and hinder successful VFL deployment. To address these limitations, we propose a trustworthy VFL framework that integrates three core innovations: (i) a multi‑dimensional reputation evaluation system that assesses each vehicle’s data quality, stability, and behavioral consistency to mitigate erratic or malicious participation. (ii) a dynamic incentive control module that self‑tunes reward weights in real time to maintain equitable participation and stable convergence. (iii) a lightweight fairness‑verification engine that ensures reward allocations are transparent, auditable, and aligned with actual contributions. The results confirm the potential of our reputation-regulated fair VFL architecture for smart transportation systems.
Included in
AI-Driven Vehicular Federated Learning: Fairness-Aware Adaptive Incentives with Blockchain Verifiability for Smart Transportation
H1-0057
Vehicular Federated Learning (VFL) is becoming a crucial enabler for the implementation and optimization of automated transport systems. This technology enables intelligent transportation systems to function by enabling networked vehicles to develop perception and control models through joint training while preserving their original data. Nevertheless, the effective deployment of VFL is dependent on the sustained participation of trustworthy vehicles. However, the dynamic nature of vehicular environments poses critical challenges, including unstable participation, data heterogeneity, and resource constraints. The sustained collaboration of smart vehicles necessitates reliable client selection and equitable incentive schemes with verifiable transparency. The distributed nature of VFL makes it vulnerable to malicious client updates that can compromise model integrity. Moreover, conventional FL frameworks lack effective client selection strategies and fair and adaptive incentive mechanisms that are capable of maintaining fairness and reliability under these fluctuating conditions. Additionally, most existing incentive schemes adopt basic contribution measures or the Shapley value method to determine reward distribution. These incur high computational cost and provide inadequate definitions of fairness. Moreover, these do not guarantee that vehicles with higher contributions receive higher rewards, which leads to fairness disparities that could undermine long-term incentives for participation and hinder successful VFL deployment. To address these limitations, we propose a trustworthy VFL framework that integrates three core innovations: (i) a multi‑dimensional reputation evaluation system that assesses each vehicle’s data quality, stability, and behavioral consistency to mitigate erratic or malicious participation. (ii) a dynamic incentive control module that self‑tunes reward weights in real time to maintain equitable participation and stable convergence. (iii) a lightweight fairness‑verification engine that ensures reward allocations are transparent, auditable, and aligned with actual contributions. The results confirm the potential of our reputation-regulated fair VFL architecture for smart transportation systems.