Dynamic resource partitioning in multi-UAV (unmanned aerial vehicle) networks enables flexible, task-oriented allocation of resources. Existing approaches pose scalability challenges in jointly optimizing multi-type tasks while satisfying operational constraints. Traditional heuristic and integer programming-based solutions fail to coordinate heterogeneous task allocation, UAV deadline constraints, and hierarchical bandwidth quotas, leading to suboptimal welfare in dynamic environments. In this work, we investigate the problem of quality of service (QoS)-aware resource allocation in UAV networks, with the objective of maximizing social welfare. We design an online primal-dual algorithm to dynamically assign UAVs to tasks based on marginal prices. Our reformulated integer program captures UAV travel constraints, bidding prices, and task-specific bandwidth quotas alongside shared resource capacities. A complementary slackness-based dual framework guides real-time decision making on task acceptance and resource usage, with duality analysis establishing a 1-to-1 correspondence between dual solutions and primal allocations. Experimental results demonstrate 15 % welfare improvement compared to static allocation baselines.

In maritime navigation, vessels require internet access for communication and entertainment, typically provided by terrestrial-based stations via relay. For routes that are difficult to cover from the shore, long-endurance Unmanned Aerial Vehicles (UAVs) can be deployed to accompany ships and offer Internet connection. However, existing systems consider Internet Service Provider (ISP) competition only, and do not incorporate it into the users' resource selection process. Furthermore, user competition is limited due to the time slot allocation method. To address these limitations, we propose an effective Online Maritime Double Auction Mechanism (OMDAM) aimed at maximizing social welfare of the maritime network. We introduce an online algorithm,$A_{online}$, to solve the online social welfare maximization problem, with an inner algorithm,$A_{core}$, handling the selection of Internet accessing devices and task allocation between users and ISPs. Theoretical analysis demonstrates that our mechanism ensures budget balance, individual rationality, and economic efficiency. Simulation results show a performance improvement of up to 17% in social welfare compared to prior art.

We study resource allocation and task scheduling for mobile artificial intelligence generated content (AIGC) in a three-layer cloud-edge-device network. Escalating industry demand for computational resources presents significant challenges in resource allocation and optimization, particularly for edge-side AIGC, which faces high computational costs and requires advanced techniques for efficient model deployment on mobile devices. Optimal resource allocation in mobile AIGC networks is naturally formulated into a 0-1 ILP, which is proven NP-hard. We reformulate the problem into both its Comp-Exp and dual forms. Then, we design an online auction framework online AIGC task scheduling (OATS) to optimize decisions on instances and time schedules, maximizing social welfare for the AIGC ecosystem. Our analysis demonstrates that OATS achieves high social welfare through appropriate bid acceptance and resource allocation. Simulation results corroborate the theoretical analysis, showcasing the efficacy of our online algorithms.

Federated learning (FL) is increasingly adopted to combine knowledge from clients in training without revealing their private data. In order to improve the performance of different participants, personalized FL has recently been proposed. However, considering the non-independent and identically distributed (non-IID) data and limited bandwidth at clients, the model performance could be compromised. In reality, clients near each other often tend to have similar data distributions. In this work, we train the personalized edge-based model in the client-edge-server FL. While considering the differences in data distribution, we fully utilize the limited bandwidth resources. To make training efficient and accurate at the same time, An intuitive idea is to learn as much useful knowledge as possible from other edges and reduce the accuracy loss incurred by non-IID data. Therefore, we devise Grouping Hierarchical Personalized Federated Learning (GHPFL). In this framework, each edge establishes physical connections with multiple clients, while the server physically connects with edges. It clusters edges into groups and establishes client-edge logical connections for synchronization. This is based on data similarities that the nodes actively identify, as well as the underlying physical topology. We perform a large-scale evaluation to demonstrate GHPFL’s benefits over other schemes.

In times when ground infrastructure can be disrupted by conflicts or natural events, the use of Unmanned Aerial Vehicle (UAV) trajectories for network services has become a crucial backup plan. Yet, many current methods don't fully optimize how UAVs are scheduled or allocate resources, resulting in less effective service. Our research aims to enhance social welfare by optimizing UAV scheduling and trajectory planning. To tackle this challenging problem, we first set up a non-convex linear programming issue and then restructure it into both its exponential and dual forms. We introduce a two-part solution. The$A_{OST}$algorithm manages task bids and UAV resource allocation, considering factors like bid values, resources, and task needs. It ranks tasks based on the value they bring. Next, the$A_{dual}$algorithm refines decisions on tasks and UAV planning by weighing task costs against benefits. Our analysis shows our method reaches a balance that boosts social welfare, ensuring the best task and resource decisions. Tests back up these claims, showing improvement in network service, and proving our method's practical value in maximizing social welfare during disruptions.

This work explores a Personalized Federated Learning (PFL) system with a central server, multiple edges and clients. Edges contain significant data distribution diversity and data scarcity. We design an auction-based online algorithm for dynamically selecting clients, who are motivated to connect with edges for model serving and participate in training for rewards. Our auction focuses on bids for model service usage, yet incorporates a reward mechanism to compute social welfare. This approach fosters enhanced collaboration between clients and edges. Through extensive simulation, we demonstrate that our algorithm substantially improves the usage of model serving requests from clients, showing an increase in social welfare with a low competitive ratio. The personalized models benefit from clients’ computational contributions and diverse datasets, while outperforming conventional FL frameworks in model accuracy. Our contributions offer a scalable, practical solution to PFL challenges, ensuring improved model performance and client engagement in a data-sensitive, reward-driven context.

In federated learning a global model is trained with training data geographically distributed over a number of clients. To reduce the communication cost over the expensive wide area network, clients complete multiple local iterations before synchronization. However, since the training data are non-iid, such infrequent synchronization would compromise the accuracy after model convergence. In order to tackle this problem, we propose Two-Dimensional Learning Rate Decay (2D-LRD) in this paper, which aims to improve the model performance by adaptively tuning the learning rate on two dimensions: round-dimension and iteration-dimension during the model training. That is, we gradually decrease the learning rate and decrease the learning rates of local iterations in a synchronization round with different speeds. Based on our experiments and analysis, we find that the sum of the inner product of round updates is a valuable signal for learning rate tuning. We perform evaluation and demonstrate that 2D-LRD can make great progress compared to the baseline scheme.