Predicting future connections among Unmanned Aerial Vehicles (UAVs) is one of the fundamental tasks in UAV Ad Hoc Networks (UANETs). In adversarial environments where UAV operational information is unavailable, future link prediction must rely solely on the observed historical topo logical data. However, the highly dynamic and sparse nature of UANET topologies poses substantial challenges in capturing structural and temporal link formation patterns. Most existing link prediction methods focus only on single-scale structural features while neglecting the effects of network sparsity, thus limiting their performance when applied to UANETs. In this paper, we propose MUST, a Multi-scale Structural-Temporal link prediction model for UANETs. In our model, multi-scale structural representations are learned using a weighted graph attention network combined with multi-scale pooling, capturing features at the levels of individual UAVs, UAV communities, and the entire network, which are then fused via concatenation. Then, a stacked long short-term memory network is employed to learn the temporal dynamics of these multi-scale structural features. To address the impact of network sparsity, we develop a tailored loss function that emphasizes the contribution of existing links during training. We validate the performance of MUST using several UANET datasets generated through simulations. Extensive experimental results demonstrate that MUST achieves state-of-the-art link prediction performance in highly dynamic and sparse UANETs.

This paper presents a cooperative edge-caching framework for LEO-assisted vehicular networks, focusing on the joint optimization of content placement and delivery under dynamic conditions. The system is modeled as a dual time-scale Markov Decision Process (DTS-MDP), where slow-scale decisions capture cache placement and fast-scale decisions capture realtime content delivery. Unlike prior work, our formulation introduces a novel triplet-based state representation ($m, f, s$) that explicitly captures the mobility of requesting vehicles, the demand characteristics of content files, and the cache status of serving satellites. This modeling choice enables the effective application of deep reinforcement learning (DRL) to a problem space that involves highly dynamic mobility, intermittent satellite visibility, storage limitations, and latency constraints. To optimize within this framework, we apply a Deep Deterministic Policy Gradient (DDPG) algorithm, which jointly manages discrete caching actions and continuous resource allocation. Simulation results demonstrate that our approach achieves significant improvements in content hit ratio, delivery latency, and system cost compared to baseline strategies. While evaluated under a serving single satellite setup, the framework is inherently scalable to larger constellations and denser vehicular networks, providing a solid foundation for future multi-satellite extensions and real-world deployments.

Node-similarity distributions not only characterize different types of complex networks, but also offer insights in the structural predictability of complex networks, and even facilitate prediction tasks in complex networks. By means of the generating function, we propose a framework to calculate the common neighbor based similarity (CNS) distributions, offering theoretical results of similarity distributions of various complex networks. Furthermore, we apply node-similarity distributions to link prediction, a key task in network analysis. Specifically, by deriving analytical solutions for two metrics: i) precision and ii) area under the receiver operating characteristic curve (AUC), we give theoretical evaluation of link prediction. Also, by analyzing i) the expected prediction accuracy of similarity scores and ii) optimal prediction priority of unconnected node pairs, we optimize link prediction with similarity distributions. Simulation results confirm our findings and also validate the proposed methods for evaluating and optimizing link prediction.

Realistic epidemic spreading is usually driven by traffic flow in networks, which is not captured in classic diffusion models. Moreover, the progress of a node's infection from mild to severe phase has not been particularly addressed in previous epidemic modeling. To address these issues, we propose a novel traffic-driven epidemic spreading model by introducing a new epidemic state, that is, the severe state, which characterizes the serious infection of a node different from the initial mild infection. We derive the dynamic equations of our model with the tools of individual-based mean-field approximation and continuous-time Markov chain. We find that, besides infection and recovery rates, the epidemic threshold of our model is determined by the largest real eigenvalue of a communication frequency matrix we construct. Finally, we study how the epidemic spreading is influenced by representative distributions of infection control resources. In particular, we observe that the uniform and Weibull distributions of control resources, which have very close performance, are much better than the Pareto distribution in suppressing the epidemic spreading.

Over the years, quantifying the similarity of nodes has been a hot topic in network science, yet little has been known about the distribution of node-similarity. In this paper, we consider a typical measure of node-similarity called the common neighbor based similarity (CNS). By means of the generating function, we propose a general framework for calculating the CNS distributions of node sets in various networks. Particularly, we show that for the Erdös-Rényi random network, the CNS distribution of node sets of any size obeys the Poisson law. Furthermore, we connect the node-similarity distribution to the link prediction problem, and derive analytical solutions for two key evaluation metrics: i) precision and ii) area under the receiver operating characteristic curve (AUC). We also use the similarity distributions to optimize link prediction by i) deriving the expected prediction accuracy of similarity scores and ii) providing the optimal prediction priority of unconnected node pairs. Simulation results confirm our theoretical findings and also validate the proposed tools in evaluating and optimizing link prediction.