This paper presents FFS-ORAN-IDS, a federated few-shot intrusion-detection framework that secures streaming traffic in Open Radio Access Networks (O-RAN) while respecting their stringent latency and resource constraints. The framework addresses the twin challenges of scarce attack labels and heterogeneous data, where naïve pseudo-label injection without sufficient confidence propagates errors, large-scale labeling of streaming traffic is impractical, and inherent uncertainty often requires costly human intervention. FFS-ORAN-IDS combines three coordinated x-functional blocks: a confidence-adaptive curriculum that releases pseudo-labels only when local TabTransformers are reliable, a diversity filter that retains the most informative uncertain packets, and a token-budgeted large-language-model (LLM) oracle that verifies the remaining hard samples. A mixed-integer optimization jointly governs curriculum pacing, sampling size, and Oracle LLM calls so that each federated round minimizes detection loss, propagation error, and LLM token cost under per-round resource caps. We train FFS-ORAN-IDS in two stages: an initial few-shot phase that fits the TabTransformer on the scarce ground-truth packets, followed by iterative rounds that refine the model with oracle-verified pseudo-labels. Experimental evaluation on the CIC-IDS 2018 benchmark shows that the proposed framework improves detection accuracy by 6%, reduces label-error propagation by 20%, and lowers energy consumption by 40% in the most label-constrained scenarios.

Deploying transformer models in Personalized Federated Learning (PFL) over wireless networks is challenging due to their large size, which leads to high communication overhead, increased latency, and excessive energy consumption. Traditional pruning and sparsification methods, designed mainly for conventional deep learning architectures, are ineffective for transformers and can cause divergence or degrade performance—especially when applied to self-attention layers or through direct federated averaging. To address these challenges, we propose a novel dual approach called PFL-TPS (PFL with Transformer Pruning and Sparsification). Our approach efficiently reduces communication and computation costs while maintaining model performance, making it suitable for resource-constrained wireless networks. Specifically, we apply adaptive pruning with trainable thresholds to the transformer's Feed-Forward Layers (FFLs), and only these trainable thresholds are shared with the server, resulting in minimal uploaded data. For the Self-Attention Layers (SALs), instead of transmitting bandwidth-intensive model parameters, we employ a server-side hypernetwork that generates personalized parameters based on device-specific embedding vectors sent by the devices, significantly reducing communication overhead and maintaining personalization. Extensive experiments show that PFL-TPS reduces energy consumption by up to 50%, decreases training time by 60.44%, and improves model accuracy by 49.87% compared to baselines in wireless networks.

Deploying large language models (LLMs) at the wireless edge is a promising solution to meet the low-latency, high-computation demands of next-generation AI applications. Although the existing literature has introduced approaches to enable distributed LLM inference, these methods largely overlook the distinct computational and communication characteristics of the two-phase LLM inference process—the pre-fill and decode phases. This oversight leads to suboptimal performance and limits scalability in real-world deployments. To address these issues, we propose a novel collaborative inference framework that strategically minimizes inference latency by optimally distributing computational loads across edge devices, the edge server, and the cloud. Our approach introduces a hybrid framework that combines head-wise parallel processing with layer-wise partitioning of LLM models, supported by a dual-phase optimization strategy. In the pre-fill phase, we optimize assigning attention heads to selected edge devices for parallel computation and efficient resource use. We then optimize for minimal latency by selecting participants, determining head assignments per device, and allocating bandwidth while meeting all constraints. In the de-code phase, our framework adaptively decides whether to execute computations locally on the edge server, offload them to the cloud, or redistribute tasks among edge devices, optimizing this decision based on the remaining latency budget and the sequential nature of the decode phase. The simulation results demonstrate that the proposed framework significantly outperforms the baseline methods, achieving a 56% reduction in inference latency, 40% improvement in bandwidth efficiency and 35% improvement in resource utilization.

Clustered Federated Multi-task Learning (CFL) has emerged as a promising technique to address statistical challenges, particularly with non-independent and identically distributed (non-IID) data across users. However, existing CFL studies entirely rely on the impractical assumption that devices possess access to accurate ground-truth labels. This assumption becomes specifically problematic in hierarchical wireless networks (HWNs), with vast unlabeled data and dual-level model aggregation, not only leading to slowing down convergence speeds and extending processing times but also resulting in increased resource consumption. To this end, we propose Clustered Federated Semi-Supervised Learning (CFSL), a novel framework tailored for more realistic scenarios in HWNs. We leverage specialized models resulting from device clustering and present two prediction model schemes, the best-performing specialized model and the weighted-averaging ensemble model, to correctly label unlabeled, unseen data. For the best-performing specialized model scheme, a specialized model excelling in label prediction for a specific device is assigned to correctly label the unlabeled data, even when the data originates from other environments, while the weighted-averaging ensemble model combines all specialized models into a unified model, capturing more details from broader data distributions across edge networks. The CFSL also introduces two novel prediction time schemes, split-based and stopping-based, for accurately timing the labeling process, alongside two strategic device selection schemes, greedy and round-robin, upon reaching each cluster’s stopping point. Extensive testing validates CFSL’s superiority over existing models in labeling and testing accuracies and resource efficiency, achieving up to 51% energy savings.

Clustered Federated Multi-task Learning (CFL) has gained considerable attention as an effective strategy for over-coming statistical challenges, particularly when dealing with non-independent and identically distributed (non-IID) data across multiple users. However, much of the existing research on CFL operates under the unrealistic premise that devices have access to accurate ground-truth labels. This assumption becomes especially problematic in, especially hierarchical wireless networks (HWNs), where edge networks contain a large amount of unlabeled data, resulting in slower convergence rates and increased processing times-particularly when dealing with two layers of model aggregation. To address these issues, we introduce a novel frame-work-Clustered Federated Semi-Supervised Learning (CFSL), designed for more realistic HWN scenarios. Our approach leverages a best-performing specialized model algorithm, wherein each device is assigned a specialized model that is highly adept at generating accurate pseudo-labels for unlabeled data, even when the data stems from diverse environments. We validate the efficacy of CFSL through extensive experiments, comparing it with existing methods highlighted in recent literature. Our numerical results demonstrate that CFSL significantly improves upon key metrics such as testing accuracy, labeling accuracy, and labeling latency under varying proportions of labeled and unlabeled data while also accommodating the non-IID nature of the data and the unique characteristics of wireless edge networks.

Clustered federated multi task learning (CFL) is introduced as an effective and efficient approach for addressing statistical challenges such as non-independent and identically distributed (non-IID) data among workers. Workers in CFL are clustered in groups based on similarity (i.e., cosine similarity) in their data distributions, in which each cluster is equipped with an efficient specialized model. However, this approach can be costly and time-consuming when implemented in hierarchical wireless networks (HWNs) due to uploading several models at every round to enable the cloud server to capture the incongruent data distribution from different edge networks. This brings about the need for novel solutions to address these challenges. To this end, this paper introduces a framework with two cloud-based model aggregation approaches, round-based and split-based, so as to minimize latency and resource consumption while attaining satisfying personalized accuracy. In the round-based scheme, the cloud aggregates the models from the edge servers after a predetermined number of rounds. As for the split-based scheme, the models are collected by the cloud only when edge servers perform the split. Extensive experiments are conducted to evaluate and compare the proposed heuristics against approaches presented in the recent literature. The numerical results and findings demonstrate that the proposed heuristics significantly conserve resources by reducing energy consumption by 60% and saving time, all while accelerating the convergence rate for cluster workers across various edge networks.