Federated learning (FL) is a distributed training paradigm that enables collaborative learning across clients without sharing local data, thereby preserving privacy. However, the increasing scale and complexity of modern deep models often exceed the computational or memory capabilities of edge devices. Furthermore, clients may be constrained to use heterogeneous model architectures due to hardware variability (e.g., ASICs, FPGAs) or proprietary requirements that prevent the disclosure or modification of local model structures. These practical considerations motivate the need for model-heterogeneous FL, where clients participate using distinct model architectures. In this work, we propose Generative Model-Aided Federated Learning (GeFL), a framework that enables cross-client knowledge sharing via a generative model trained in a federated manner. This generative model captures global data semantics and facilitates local training without requiring model homogeneity across clients. WhileGeFLachieves strong performance, empirical analysis reveals limitations in scalability and potential privacy leakage due to generative sample memorization. To address these concerns, we proposeGeFL-F, which utilizes feature-level generative modeling. This approach enhances scalability to large client populations and mitigates privacy risks. Extensive experiments across image classification tasks demonstrate that bothGeFLandGeFL-Foffer competitive performance in heterogeneous settings. Code is available at [1].

Glaucoma is one of the leading causes of blindness worldwide, where early and precise diagnosis is critical for effective treatment. We explore the potential of deep learning to generate a Visual Field Grayscale Map (GM) directly from a Conventional Fundus Image (CFI) at the same time point, enabling an intuitive representation of visual field defects for the diagnosis of glaucoma. The Humphrey Visual Field (HVF) test is a standard method for assessing visual field defects, but its dependency on active participation and prolonged testing time limits its reliability, particularly in elderly patients. To address these challenges, we propose CFI2GM-GAN, a novel deep learning framework that directly generates GM from CFI, offering a faster and cost-effective alternative while complementing the HVF test. Our method incorporates dataset partitioning based on defect severity to mitigate class imbalance, attention-guided fusion techniques for improved prediction reliability, and a refined loss function customized to fundus image characteristics.

Federated learning (FL) enables distributed training while preserving data privacy, but stragglers—slow or incapable clients can significantly slow down the total training time and degrade performance. To mitigate the impact of stragglers, system heterogeneity, including heterogeneous computing and network bandwidth, has been addressed. While previous studies have addressed system heterogeneity by splitting models into submodels, they offer limited flexibility in model architecture design, without considering potential inconsistencies arising from training multiple submodel architectures. We proposenested federated learning (NeFL), a generalized framework that efficiently divides deep neural networks into submodels using both depthwise and widthwise scaling. To address theinconsistencyarising from training multiple submodel architectures, NeFL decouples a subset of parameters from those being trained for each submodel. An averaging method is proposed to handle these decoupled parameters during aggregation. NeFL enables resource-constrained devices to effectively participate in the FL pipeline, facilitating larger datasets for model training. Experiments demonstrate that NeFL achieves performance gain, especially for the worst-case submodel compared to baseline approaches (7.63% improvement on CIFAR-100). Furthermore, NeFL aligns with recent advances in FL, such as leveraging pre-trained models and accounting for statistical heterogeneity.

Recently, unmanned aerial vehicles (UAVs) have been the subject of much attention, and UAV swarms are a promising solution to resolve the limitations of a single UAV. In terms of aerodynamics, fixed-wing UAV swarms fly in a V-formation to reduce their energy usage via upwash. UAVs in the V-formation are typically in close proximity, reducing power consumption aerodynamically. However, these formations lack spatial separability, degrading the performance of communication employing non-orthogonal multiple access schemes. In this paper, to tackle this problem, the formation of UAV swarms considering both aerodynamics and communications employing the beam division multiple access (BDMA) scheme is examined. Problems related to the minimization of power consumption of UAVs and the ground control station (GCS) are formulated separately, after which we optimize the beamforming vector, power, and formation of the UAV swarm and propose a novel UAV swarm formation, termed the O-formation, that is power-efficient for communications. Numerical results show that formation of the UAV swarm affects the power consumption related to communications and that the proposed O-formation improves the power efficiency compared to the V-formation and a random formation.