Despite the growing interest in developing decentralized federated learning (FL) in wireless networks, few works have explored the impact of network topology. I.e., due to devices sharing model parameters (or logits) by following a consensus protocol, the convergence performance of decentralized FL algorithms is affected by the network topology. In case the network topology dynamically changes, the existing solutions may fail. In this work, we investigate a topology control technique in the design of the consensus protocol. Using the strategy as a building block, we develop upon knowledge distillation a decentralized framework for heterogeneous FL, dubbed as DistHFL. By conducting the consensus process on a connected dominating set associated subnetwork rather than the whole network, and updating the subnetwork towards topology change, the DistHFL significantly accelerates the convergence of the model learning algorithm and enhances its robustness against dynamic topology. We develop mathematical models to demonstrate the convergence of the proposed consensus protocol, as well as the convergence of the DistHFL. We derive an upper bound for the generalization performance of the ensemble models trained with DistHFL by applying the domain measurement tools. Through conducting extensive numerical experiments, we verify that DistHFL outperforms the state-of-the-art algorithms under all setups in all cases.

In this work, we propose a generative model for spatial networks. The central idea of the model is to use empirical data from real-world spatial networks as a foundation for the network structure that is constructed starting from a set of disconnected nodes and assigning the node positions and the links between them based on key topological and spatial features. In particular, for each node, the number of links to add is determined by sampling values from the degree distribution of real-world networks, while the nodes to connect with are randomly selected based on a probability that depends on the distance between the node positions. For the node positions, we propose either leaving them unconstrained or applying one of four physical constraints with varying levels of strictness. The model can be applied to both two- and three-dimensional spaces. As an example of the former, we consider power grids, while for the latter, we study brain networks. In both cases, we show that the model can be effectively applied to generate surrogate networks, providing a useful tool when spatial network data is limited.

Quantum adiabatic optimization has long been expected to outperform classical methods in solving NP -type problems. While this has been proven in certain experiments, its main applications still reside in academic problems where the size of the system to be solved would not represent an obstacle to any modern desktop computer. Here, we develop a systematic procedure to find the global optima of the modularity function to discover community structure in complex networks solely relying on pure annealers rather than hybrid solutions. We bypass the one-hot encoding constraints by hierarchically and recursively encoding binary instances of the problem that can be solved without the need to guess the exact penalties for the Lagrange multipliers. We study the variability and robustness of the annealing process as a function of network size, directness of connections, topology, and the resolution of the communities. We show how our approach produces meaningful and at least equally optimal solutions to state-of-the-art community detection algorithms while maintaining tractable and potentially scalable computing times. Lastly, due to its recursive nature, the annealing process returns intermediate subdivisions, thus offering interpretable rather than black-box solutions. These <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">dendrograms</i> can be used to unveil normal and pathological hidden hierarchies in brain networks, hence opening the door to clinical workflows. Overall, this represents a first step towards an applicable practice-oriented usage of pure quantum annealing, potentially bridging two segregated communities in modern science and engineering: that of network science and quantum computing.

Multi-user semantic communication enables multiple users (such as mobile devices) to collaborate on intelligent tasks, reducing communication load while enhancing efficiency. However, it is vulnerable to backdoor attacks due to the absence of auditing mechanisms for mobile users. In such attacks, malicious attackers can inject carefully crafted samples to implant backdoor neurons into the target model, which can be triggered to mispredict the target class based on the attacker's intent. Most existing defense methods for backdoor attacks are inadaptable to semantic communication systems, post-analysis, and are unable to respond timely. To address these challenges, we propose Reject-BD, a contrastive training-based defense mechanism tailored for multi-user semantic communication systems. Firstly, Reject-BD can actively apply test gradients to local semantic encoders. Then, contrastive training is used to amplify the differences in the updated test gradients of different clients to predict the credence degree of each client. Finally, the global semantic decoder is dynamically updated according to each client's credence degree. Experimental results demonstrate that Reject-BD effectively reduces the success rate of backdoor attacks while maintaining relatively high classification accuracy of the semantic decoder compared to five baseline methods.

Proof of Work (PoW) is a fundamental and widely adopted consensus mechanism in blockchain systems that enables effective consensus in permissionless environments, solely through block propagation without additional message exchange between nodes. However, PoW can result in excessive energy consumption and network centralization. This paper proposes VRF-PoW, a novel consensus protocol that reduces energy consumption while preserving the decentralized and permissionless nature of PoW. The VRF-PoW method integrates Verifiable Random Functions (VRFs) into the PoW process, reducing network energy consumption. To prevent Sybil attacks, this work introduces a pre-PoW phase. Further, this work validates VRF-PoW via a theoretical analysis and numerical evaluation. The results demonstrate that VRF-PoW significantly reduces energy consumption compared to the traditional PoW, while maintaining decentralization, stable block generation, and network liveness.

With the widespread adoption of cryptocurrencies, exemplified by Bitcoin, mixing services have been increasingly misused for illicit activities such as fraud, extortion, and money laundering. These services cut off the direct link between input and output addresses to conceal publicly available transaction traces on the blockchain, thereby evading regulatory investigations. In recent years, significant efforts have been made to deanonymize Bitcoin mixing services (BMS). While these efforts have yielded some achievements in service detection, they still face several challenges, such as vague theoretical definitions, coarse-grained identification capabilities, and insufficient labeling information. To address these challenges, this paper formalizes the notion of BMS with a syntax, security model, and goals, providing a universal definition for evaluating service anonymity and detection feasibility. We also propose a fine-grained detection framework for BMS, called BMS-FDET, which leverages heterogeneous Bitcoin transactional networks and multiple attention mechanisms to identify mixing transactions across different services. To improve the model's detection capabilities and compare performance, a sufficient ground-truth dataset with over 596,200 entries covering more than 15 mixing services is built from extensive data sources. Comprehensive experiments demonstrate the superiority of BMS-FDET over existing methods.

This paper investigates the challenge of radar echo loss caused by urban occlusions in low-altitude aerial drone surveillance. We propose a deep learning framework, built on an enhanced U-Net, to restore radar signals and improve joint velocity and range estimation. A Slow-Fast Time Echo Matrix (SFTEM) is constructed to encode temporal and phase features, serving as the input to the network. Matched filtering is first applied to form the SFTEM as the network input, followed by Doppler FFT on the restored SFTEM for range–velocity estimation. To address severe pulse loss, the model incorporates axial attention and residual connections, while a Partial-Edge Preserved Pulse Loss (PEPPL) model is introduced to simulate realistic urban blockages. Experimental results demonstrate a mean range error of about <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$ 2.08m$</tex-math></inline-formula>, which is well below the range resolution (<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$ 5--15m$</tex-math></inline-formula>), and a velocity error of <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$0.749\,\mathrm{m/s}$</tex-math></inline-formula>, even under conditions with over 90% central pulse loss, highlighting the robustness and effectiveness of the proposed approach for aerial drone sensing in complex urban environments.

This paper addresses the time-varying formation tracking (TVFT) for second-order delayed multi-UAV systems (SODMUSs) subject to denial-of-service (DoS) attacks utilizing an event-triggered secure strategy. To begin with, a novel event-triggered secure strategy is designed, accompanied by a corresponding delayed piecewise differential inequality to address challenges introduced by DoS attacks and state delay. Then, a sufficient criterion is obtained to guarantee the TVFT for SODMUSs subject to asynchronous DoS attacks by proposed event-triggered secure strategy. Moreover, it is proved that the Zeno behavior does not exist in the closed-loop system. Finally, the event-triggered secure strategy is utilized to quadrotor UAV TVFT experiments where the numerical simulation and the software-in-the-loop simulation under hping3-based DoS attacks are presented to verify the validity of the theoretical results.

Large language models (LLMs) have made significant advances in the field of natural language processing (NLP). However, their centralized training approach faces challenges related to data privacy, communication efficiency, and robustness against adversarial attacks, particularly in wireless environments. With the gradual depletion of high-quality public data, there is an urgent need to leverage private data distributed across various parties. Although federated learning (FL) offers a privacy-preserving collaborative training paradigm, it struggles to meet the high computational demands of edge devices and remains vulnerable to adversarial attacks. This paper introduces <sc xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">ROFED</small>-LLM, a novel framework for robust, privacy-preserving training of LLMs on decentralized private data over wireless networks. By integrating split federated learning, which partitions the model across devices to enhance privacy, with adaptive jamming defense mechanisms, <sc xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">ROFED</small>-LLM enables collaborative LLM training without raw data sharing while ensuring resilience against wireless adversarial attacks. Our multi-modal defense strategy combines model-level protections, such as differential privacy and dynamic pruning, with communication-level safeguards, including adaptive beamforming which optimizes wireless signal transmission to mitigate interference, and resource allocation optimization. Extensive experiments across diverse NLP tasks demonstrate <sc xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">ROFED</small>-LLM's superiority, achieving a 12.87% improvement in privacy preservation and 18.26% enhancement in jamming resilience compared to existing methods such as FedAvg and SCAFFOLD, with only a marginal 3.94% trade-off in model accuracy. Our code repository has been open sourced at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://anonymous.4open.science/r/RoFed-LLM-54E1</uri>.