The vehicular environment presents unique challenges, including massive data generation, stringent latency requirements for safety-critical applications, bandwidth limitations, and intermittent connectivity, which make centralized learning approaches impractical. Vehicular Federated Learning (VFL) enables distributed model training by leveraging local data from connected vehicles, while preserving data privacy and reducing network overhead. However, the dynamic nature of VFL presents several additional challenges. High vehicle mobility and unstable channels lead to inconsistent client participation, while heterogeneous vehicle capabilities result in unbalanced training workloads and competitive resource allocation. These challenges significantly degrade VFL model performance and prolong training periods. In this paper, we propose a Mobility Resilient Vehicular Federated Learning (MR-VFL) scheme, which comprises two key components: an amplification-based adaptive vehicular FL (AVFL) training scheme and a dual-timescale FL scheduler. Specifically, AVFL adapts local training epochs to vehicle capabilities to improve scheduling flexibility and alleviate the impact of insufficient local epochs on model updates, which enhances training efficiency and reduces communication competition. The dual-timescale FL scheduler includes a macro scheduling strategy that optimizes long-term VFL performance based on the correlation between convergence speed and model accuracy, and a Mamba-based real-time scheduler that enhances training efficiency and reduces decision latency in massive vehicles scenarios. Extensive simulations show that MR-VFL effectively mitigates performance degradation due to complex vehicle mobility and heterogeneity, and improves training efficiency.

Large language models (LLMs) are revolutionizing various fields due to their powerful generation capabilities. However, their immense computational complexity poses significant challenges in resource consumption, inference latency, and data privacy for traditional cloud-centric deployments. Edge artificial intelligence (Edge-AI) offers promising LLMs deployment solutions by leveraging distributed resources at the network edge. However, existing approaches struggle to adapt to dynamic workloads and efficiently utilize heterogeneous resources in Mobile Edge Computing (MEC) environments. This paper proposes aDynamicBatching andAdaptivePartitioning (DyBAP) scheme for LLMs deployment, which utilizes ubiquitous geo-distributed resources via end-edge-cloud collaboration. Firstly, we formulate a collaboration deployment optimization problem to minimize inference latency and resource usage under heterogeneous resource and user requirements for latency and accuracy constraints, which is NP-hard. Secondly, to solve this, we develop a dynamic batch fusion optimization algorithm that optimizes the batch size of inference by utilizing the parallel processing power of computing units to balance the latency and resource usage. A block-aware partition optimization algorithm based on multi-agent reinforcement learning (MARL) is proposed for efficient transformer block allocation, integrating mobility awareness for optimal partitioning across dynamic network environments. Simulation results demonstrate the superiority of DyBAP over other benchmarks, reducing inference latency by 17.94% and saving 11.12% in memory resource consumption compared to the end-edge-cloud collaboration approaches.

Asynchronous Federated Learning (AFL) can mitigate the straggler problem due to unbalanced training time of clients in Synchronous Federated Learning (SFL), thereby reducing the aggregation time and improving the training efficiency. However, AFL introduces training bias since different updating frequencies of heterogeneous clients can cause unequal knowledge contributions to the global model. Meanwhile, if the client data are heterogeneous, the local optimum may be drifted from the global one, which exacerbates the training bias problem. In order to solve the above issues, we propose a Knowledge Alignment framework for AFL with Heterogeneous Data, termed as KAFL-HD. Firstly, considering data heterogeneity, a data quality-aware aggregation method is proposed to estimate client contributions precisely, where both model staleness and data quality are utilized in aggregation weights. Secondly, a knowledge distillation method with staleness is designed to supplement more knowledge from slow clients to the global model. Thirdly, an adaptive learning rate adjustment method is proposed to customize the local learning rate based on the aggregation frequency and weight, which aligns the knowledge contributions of clients in the local training process. Furthermore, we provide theoretical analysis under a non-convex setting to show the convergence speed of KAFL-HD. Finally, comprehensive experiments are conducted, and the results show that KAFL-HD achieves the highest accuracy and fairness performance compared to the state-of-the-art baselines.

Personalized Federated Learning (pFL) can improve the accuracy of local models and provide enhanced edge intelligence without exposing the raw data in Mobile Edge Computing (MEC). However, in the MEC environment with constrained communication resources, transmitting the entire model between the server and the clients in traditional pFL methods imposes substantial communication overhead, which can lead to inaccurate personalization and degraded performance of mobile clients. In response, we propose a Communication-Efficient pFL architecture to enhance the performance of personalized models while minimizing communication overhead in MEC. First, a Knowledge-Aware Parameter Coaching method (KAPC) is presented to produce a more accurate personalized model by utilizing the layer-wise parameters of other clients with adaptive aggregation weights. Then, convergence analysis of the proposed KAPC is developed in both the convex and non-convex settings. Second, a Bidirectional Layer Selection algorithm (BLS) based on self-relationship and generalization error is proposed to select the most informative layers for transmission, which reduces communication costs. Extensive experiments are conducted, and the results demonstrate that the proposed KAPC achieves superior accuracy compared to the state-of-the-art baselines, while the proposed BLS substantially improves resource utilization without sacrificing performance.

Federated Learning (FL) enables efficient and privacy-preserving Edge Intelligence (EI) in Mobile Edge Computing (MEC). However, implementing FL-enabled EI services faces critical challenges, including data and device heterogeneity, limited network resources, uneven distribution of network infrastructure, etc., which may intensify with increasing system scale. These challenges are particularly acute in multi-provider environments where edge servers are suboptimally allocated across federations, leading to degraded convergence and increased training costs. In this paper, we present a novel Multiple Hierarchical Federated Learning (MHFL) architecture for large-scale FL and design an Edge Server Reallocation scheme (ESR-MHFL) to enhance training efficiency by optimally redistributing edge servers among federations based on their contribution to model convergence. We first develop a closed-form analysis model for MHFL to quantify training time, computation, and communication costs. To improve training efficiency, we analyze the impacts of edge server allocation on convergence and formulate server reallocation as a multi-item auction problem with theoretical guarantees. We then propose ESR-MHFL, which leverages Coalition Structure Generation (CSG) and greedy matching methods to simplify the reallocation problem and enhance efficiency. Extensive numerical simulations demonstrate that ESR-MHFL not only improves model accuracy while reducing training cost but also exhibits strong compatibility with existing client selection methods, achieving improved training efficiency. The total economic expenditure combining all components

Personalized Federated Learning (pFL) can effectively exploit the non-IID data from distributed clients by customizing personalized models. Existing pFL methods either simply take the local model as a whole for aggregation or require significant training overhead to induce the inter-client personalized weights, and thus clients cannot efficiently exploit the mutually relevant knowledge from each other. In this paper, we propose a knowledge-aware parameter coaching scheme where each client can swiftly and granularly refer to parameters of other clients to guide the local training, whereby accurate personalized client models can be efficiently produced without contradictory knowledge. Specifically, a novel regularizer is designed to conduct layer-wise parameters coaching via a relation cube, which is constructed based on the knowledge represented by the layered parameters among all clients. Then, we develop an optimization method to update the relation cube and the parameters of each client. It is theoretically demonstrated that the convergence of the proposed method can be guaranteed under both convex and non-convex settings. Extensive experiments are conducted over various datasets, which show that the proposed method can achieve better performance compared with the state-of-the-art baselines in terms of accuracy and convergence speed.

As a distributed learning paradigm, federated learning (FL) can be applied in mobile edge computing (MEC) to support real-time artificial intelligence by leveraging edge computation resources while preserving data privacy in the end devices. However, the unpredictable wireless connections between end devices and edge servers in MEC (e.g., frequent handovers and unstable wireless channels) may result in the loss of important model parameters, which slows down the FL training process and degrades the quality of the global model. In this paper, we propose an adaptive collaborative federated learning (ACFL) scheme to accelerate the convergence and improve model reliability by mitigating communication-based parameter loss under a three-layer MEC architecture. First, a dynamic epoch adjustment method is proposed to reduce communication rounds by dynamically adjusting the training epochs in end devices. In addition, to accelerate the FL convergence, we present an edge server collaborative training scheme by leveraging a multi-layer computing architecture, where edge servers utilize their maintained data to collaboratively train models with end devices. Finally, extensive simulations are conducted and show that ACFL can efficiently improve model reliability and accelerate the convergence of the FL process in MEC.