This article addresses a non-cooperative game problem subject to privacy leakage, in which the players exchange their messages under limited transmission bandwidth. Based on the vector quantizer adopted to improve the communication efficiency, we propose a novel differentially-private distributed Nash equilibrium seeking algorithm, where a probabilistic mapping mechanism employing truncated discrete noise is designed to mask the quantization levels of exchanged messages. The theoretical analysis demonstrates that our algorithm achieves both exact and linear convergence in the mean-square sense, and also guarantees (& varepsilon;, S)-differential privacy for the players at each iteration. Additionally, we analyze the impact of privacy parameters on the truncated bound of the discrete noise, and further present a trade-off between the privacy level and convergence rate of our algorithm. (c) 2026 Elsevier Ltd. All rights are reserved, including those for text and data mining, AI training, and similar technologies.

With rapid development of industrial internet, industrial cyber-physical systems (ICPSs) have been widely deployed to perform and supervise industrial applications. However, ICPSs still face significant cybersecurity challenges. Traditional defense mechanisms are mostly static and passive, which may fail to provide real-time protection. To solve the aforementioned problem, moving target defense (MTD) technique has been proposed as a proactive solution. However, due to the increasing sophistication and persistence of cyberattacks, it is difficult for single-phase MTD approaches to provide effective defense by only mitigating individual phases of the attack process. Therefore, we present Deep-Shield, a novel multiphase MTD approach based on hierarchical deep reinforcement learning (HDRL) to improve the defense performance of single-phase MTD approaches when facing advanced persistent threat (APT) in ICPSs. We consider three representative MTD countermeasures, i.e., IP address shuffling, software diversity, and components redundancy, which are able to mitigate attacks at different phases of cyber kill chain and protect assets contained in critical infrastructures. First, we formulate the dynamic implementation of multiphase MTD countermeasures as a semi-Markov decision process. Second, we detect current attack patterns by a neural network called PerNet, which is derived from Anomaly Transformer. Then, we design a HDRL-based multiphase MTD algorithm for defense decision-making. Finally, through extensive experiments on a platform of software defined networks, we show that our proposed approach can achieve better defense performance compared with state-of-the-art solutions when dealing with APT.

This article addresses the bandit game problem subject to privacy leakage, where the cooperative players aim to learn the optimal action profile that minimizes the global cost. The players do not have closed-form expressions for their payoff functions and can only receive the feedback of their local costs. We propose a privacy-preserving distributed bandit learning algorithm based on the residual gradient estimator, which adopts the stochastic quantization with a binary randomized response scheme to mask action profile estimates before communication. The theoretical analysis demonstrates that our algorithm can achieve an expected regret order of O(T- 3/4) and preserve Edp-differential privacy for the players.

Deep reinforcement learning (DRL)-based moving target defense (MTD) emerges as an outstanding method to enhance the security of highly hostile Internet of Things (IoT) environments. However, due to the gap between certain stationary learning environment and real-world, even a well-trained DRL model may not adapt to unknown attacks in the real-world network environments. Therefore, we present a DRL-based self-evolving MTD approach against unknown attacks. First, we formulate the defense in a dynamic network environment as a Markov decision process (MDP), and utilize a DRL model based on actor-critic framework to obtain the optimal sequential defense strategies. Second, we deploy honeypots within the network environments to capture the traffic features of unknown attacks. These features are then specifically labeled to enable the DRL model to learn the characteristics of unknown attacks and identify them. Third, we design an actor network based on ResNet architecture to perceive the network states and make defense decisions in response to these unknown attacks. Finally, through extensive experiments on a platform based on software defined networks, we show that our proposed approach can enhance the defense performance compared with state-of-the-art solutions when dealing with unknown attacks.

This paper considers a distributed stochastic non-convex optimization problem, where the nodes in a network cooperatively minimize a sum of L-smooth local cost functions with sparse gradients. By adaptively adjusting the stepsizes according to the historical (possibly sparse) gradients, a distributed adaptive gradient algorithm is proposed, in which a gradient tracking estimator is used to handle the heterogeneity between different local cost functions. We establish an upper bound on the optimality gap, which indicates that our proposed algorithm can reach a first-order stationary solution dependent on the upper bound on the variance of the stochastic gradients. Finally, numerical examples are presented to illustrate the effectiveness of the algorithm.

This article considers an online aggregative game equilibrium problem subject to privacy preservation, where all players aim at tracking the time-varying Nash equilibrium, while some players are corrupted by an adversary. We propose a distributed online Nash equilibrium tracking algorithm, where a correlated perturbation mechanism is employed to mask the local information of the players. Our theoretical analysis shows that the proposed algorithm can achieve a sublinear expected regret bound while preserving the privacy of uncorrupted players. We use the Kullback–Leibler divergence to analyze the privacy bound in a statistical sense. Furthermore, we present a tradeoff between the expected regret and the statistical privacy, to obtain a constant privacy bound when the regret bound is sublinear.

This article investigates a distributed online learning problem with privacy preservation, in which the learning nodes in a distributed network aims to minimize the sum of local loss functions over time horizon T. Based on the push-sum protocol and the Laplace mechanism, we propose a differentially private distributed dual averaging algorithm for constrained distributed online learning problem over time-varying digraphs. It is shown that the expectation of the regret of our algorithm achieves a sublinear rate of O(T). Furthermore, we provide an analysis of differential privacy, which reveals a tradeoff between the accuracy and the privacy level of our algorithm. Finally, numerical examples are presented to validate the effectiveness of the algorithm.

Virtual network embedding (VNE) that instantiates virtualized networks on a substrate infrastructure, is one of the key research problems for network virtualization. Most existing VNE approaches, however, focus on the current virtual network request (VNR) and treat all VNRs equally, which disregard the long-term impact and waste many resources on the process of embedding infeasible VNRs (i.e., VNRs that cannot be embedded completely). To address these problems, a proactive virtual network embedding algorithm based on hierarchical reinforcement learning, VNE-HRL, is proposed in this paper. Within our framework, the VNE task is performed by a two-level agent that considers both the long-term impact of a VNR and the short-term effect of an embedding action. For each processing, a high-level agent aims to select a currently feasible VNR with the maximum long-term reward from a window-based batch, and a low-level agent is assigned to embed the selected VNR on a substrate infrastructure by performing a series of embedding actions. Extensive simulation results indicate that our algorithm best performance on most metrics compared with existing state-of-the-art solutions, with up to 9.92% and 33.03% improvement on acceptance ratio and average revenue.

本文设计了基于线性二次型微分博弈的多个攻击者、多个防御者和单个目标的追逃问题最优策略.首先,针对攻防双方保持聚合状态的情形,基于攻击方内部、防御方内部以及双方之间的通信拓扑,分别给出了目标沿固定轨迹运动和目标采取逃跑时攻防双方的最优策略.其次,针对攻防双方保持分散状态的情形,利用二分图最大匹配算法分配相应的防御者与攻击者,将多攻击者、多防御者追逃问题转化为多组两人零和微分博弈,并求解出了攻防双方的最优策略.最后,数值仿真验证了所提策略的有效性.

This paper is concerned with a trajectory optimization problem for multiple unmanned aerial vehicles (multi-UAV) systems, where the optimization model is constructed based on quadrotor UAV dynamics. The problem is decomposed into a two-layer structure consists of a master problem and n subproblems, with n the amount of UAVs in the system. We propose a model-based distributed algorithm to minimize the energy consumption of the multi-UAV system, in which each UAV obtains its own trajectory by solving the corresponding subproblem, and the UAVs coordinate their trajectories according to master problem. It is shown that the proposed algorithm achieves exact convergence over directed networks, and an upper bound on the residual of cost function is provided. A numerical simulation is presented to demonstrate the validity of our algorithm.

共识算法是区块链系统维护数据一致性的核心机制.本文深入调研并分析了具有代表性的共识算法及其演化历程;基于共识过程提出共识算法的分类模型,并对各类型中代表性的共识算法进行详细分析;最后从去中心化、可扩展性、安全性、一致性、可用性、分区容忍性六个方面建立了一套共识算法的评价指标体系,并对代表性的共识算法进行对比分析,给出各类算法综合性的性能评价,希望为共识算法的应用与创新提供参考.

Blockchain has recently attracted great interest from both academia and industry. Ethereum introduces programmability into blockchain through smart contracts and provides an open-source computing platform for blockchain-based decentralized applications (DApps). There are currently thousands of DApps pertaining to different application domains, including games, gambling and finance. In order to better comprehend blockchain application scenarios and help developers understand DApps better, clear DApps classification criteria are needed. However, many DApps that can be found through collection websites (commonly known as DApp Stores) are not classified properly, making these datasets imprecise. This issue has motivated the present empirical study of DApp categories, as a part of which over 2,500 DApps in three DApp Stores are investigated, allowing us to produce and publicly release a high-quality dataset in which misclassified DApps are relabeled manually, facilitating their more precise classification. We also propose DAppClassifier, a novel technique for classifying DApps based on their actual functionalities. When developing the new classifier, we extracted features from source code, bytecode and historical transactions, and trained neural networks to classify DApps. Our approach was evaluated on the released dataset and achieved good precision.

This paper addresses a linear quadratic differential game approach for leaderless formation control of multi-agent system with agents suffering from the model uncertainties and the unknown external disturbances. The external disturbance is viewed as a fictitious player attempting to maximize the cost functions of agents. Due to the bounded uncertainties of the model, the guaranteeing cost strategies for agents are given on the basis of the worst-case disturbance. A set of coupled Riccati differential equations is solved to achieve the desired formation. Finally, a simulation example of triangle formation is provided to verify the feasibility of the proposed scheme.

Single image haze removal is a challenging problem to address, and various constraints/priors have been previously considered to obtain acceptable dehazing solutions. In this paper, we propose a trainable end-to-end system for single image dehazing called ReDehazeNet based on the residual and dilation convolutional neural networks. The first part of the networks incorporated into the system is used for recovering a coarse clear image, which is predicted by adopting a context aggregation sub-network that can capture the global structure information. The second part of the network adopts a novel hierarchical convolutional neural network to further refine the details of the clean image by integrating the local context information. Experiments on benchmark images show that ReDehazeNet outperforms several existing state-of-the-art methods while being highly efficient and easy to use.