ABSTRACT In the context of security protection against false data injection attacks (FDIAs) in power grids, traditional federated learning effectively utilizes decentralized data resources for distributed training and achieves global collaboration. However, during the model aggregation process, it often overlooks or drowns out local sparse key features, significantly increasing the risk of missed detection of specific attack patterns. To address this issue, this paper proposes a personalized detection framework based on federated learning. Initially, the bidirectional transformer detection (BTD) model detection algorithm is deployed on the client side and trained on local data. Subsequently, through personalized federated learning, the client dynamically combines the weights of the global and local models to generate a personalized detection model. The framework employs a collaborative optimization mechanism of “global knowledge sharing and local feature adaptation” to effectively mitigate the feature drowning problem while strictly safeguarding data privacy. Compared to existing methods, this approach significantly enhances detection accuracy and robustness against differentiated attack patterns, thereby establishing a more reliable security defense system for smart grids.

Multimodal sarcasm detection relies on identifying semantic incongruity between visual and textual modalities. However, existing methods typically model incongruity through monolithic feature fusion or shallow interactions, neglecting the hierarchical structure of sarcasm, which manifests distinctively at entity, attribute, and scene levels. Consequently, these models often rely on spurious correlations rather than genuine causal dependencies, resulting in limited robustness against modality imbalance and distribution shifts. To address these challenges, we propose the Hierarchical Causal Chain-of-Thought (HC-CoT) framework, where Chain-of-Thought refers to a structured inference trace over the hierarchical latent variables of our H-SCM (rather than LLM-style natural-language rationales), a hierarchical causal reasoning framework that models sarcasm with a three-level Hierarchical Structural Causal Model (H-SCM) whose bottom-up causal structure (Entity → Attribute → Scene) is learned under explicit sparsity and acyclicity constraints. Over this SCM, HC-CoT performs bidirectional inference: bottom-up evidence aggregation forms scene hypotheses, while top-down contextual refinement re-evaluates lower-level states without introducing reverse causal edges. Training combines supervised learning with (i) missing-modality consistency regularization and (ii) counterfactual augmentation with explicit label policies, improving robustness without relying on heuristic shortcut cues. Extensive experiments on the MMSD and MMSD2.0 benchmarks demonstrate that HC-CoT achieves new state-of-the-art performance, exhibiting significant gains in accuracy, robustness, and interpretability.

Abstract The deployment of artificial intelligence (AI) technology in various emerging network applications has spawned a large number of computing tasks, which require dynamic collaboration of multi-dimensional resources from the perspective of communication and computing to meet service requirements such as ultra-low latency, ultra-high reliability, and ultra-fast response. In this paper, we introduce a Computing Integration Networking (CIN) architecture, which built to integrate ubiquitous heterogeneous resources in the Internet into a distributed pool, enabling CIN paradigm to fully utilize all available heterogeneous resources and provide efficient customized services evolving both communication and computation. Owing to network function virtualization (NFV) technology, computing services can be implemented as service chains, which are composed of ordered virtual network functions. We then design a service chain-driven CIN architecture with “three layers and three domains” characteristics, which consists of generalized service layer, mapping adaption layer and converged network layer. The designed architecture enables the adaptability for users, flexibility for CIN, and profitability for providers. Furthermore, we outline key technologies such as measurement and modeling of multi-dimensional heterogeneous resources, fine-grained multi-dimensional awareness, multi-dimensional identification networking and heterogeneous resources allocation. In addition, we formulate the heterogeneous resource joint optimization problem and verify the effectiveness of the designed scheme. Furthermore, we explore open issues based on our review and indicate potential research trends of the new computing paradigm.

How to recommend an effective prescription with intelligent assistant treatment remains a key issue in Traditional Chinese Medicine (TCM). To address this issue, various intelligent assistant treatment methods have been developed in recent years. However, existing methods barely integrate together knowledge graphs of TCM and the characteristics of individual differences between patients, which play very important roles in the prescription recommendation of TCM. For this reason, this work proposes a novel framework of TCM prescription recommendation, referred to as CETCMKG , which integrates the TCM knowledge graph to improve the accuracy and interpretability of prescription recommendation. More specifically, features of herb and symptom nodes are learned by combining the graph embedding models and graph convolutional neural networks. During the prediction and recommendation phase, multi-head attention mechanisms and multi-layer perceptrons jointly analyze symptom patterns to guide herb selection. Following this analysis, the performance of CETCMKG is rigorously evaluated through comparative experiments, demonstrating its superiority over existing state-of-the-art prescription recommendation methods in TCM.

This article revisits the federated policy gradient algorithm with environment heterogeneity for finding the optimal policy in multiagent reinforcement learning (RL). Toward this direction, personalized federated RL methods have been presented recently. However, existing personalized federated policy gradient methods may confine the personalized capacity of local policy models. In order to tackle this challenge, this article develops a provably convergent personalized federated first-order policy gradient algorithm, referred to as PfoPG, which learns a personalized policy model by adaptively mixing optimal global and local policies. Moreover, the momentum-based importance sampling is also introduced into PfoPG to improve its convergence speed. Meanwhile, this article rigorously analyzes the nonasymptotic convergence behavior of PfoPG. More specifically, PfoPG converges to a stationary policy with rateO(1/K), whereKdenotes the number of iterations. Compared to the state-of-the-art federated policy gradient methods, PfoPG can improve the convergence rate fromO(1/K2/3) toO(1/K). Finally, we verify the effectiveness of PfoPG by various experiments based on the multiagent particle environment.

Exogenous variables provide complementary information that enhances endogenous representations, thereby facilitating more accurate multivariate time series forecasting (MTSF). However, existing methods typically overlook the synergistic interplay between exogenous and endogenous variables by adopting shallow fusion strategies such as simple concatenation or separate encoding, which fail to capture the dynamic dependencies essential for modeling complex temporal patterns. To address this issue, we propose Endexformer, a novel hierarchical Endogenous-Exogenous modeling framework built upon the Transformer architecture. Specifically, Endexformer adopts a hierarchical architecture to jointly model temporal embeddings of endogenous variables and structural embeddings of exogenous variables, enabling a unified representation of cross-variable dependencies. To capture the fine-grained temporal patterns of endogenous variables, we present a multilevel temporal attention mechanism that leverages variable-level embeddings to adaptively incorporate exogenous information. Furthermore, we design a dynamic interactive attention mechanism that selectively emphasizes informative endogenous and exogenous patterns, mitigating redundancy and preserving semantic integrity in variable representations. Extensive experiments on eight real-world datasets show that Endexformer achieves outstanding performance against competing benchmark approaches in MTSF tasks across various temporal scenarios.

Abstract The detection of pests and diseases in crops is currently a hot topic. The complexity of pest and disease object in the field, combined with inconsistent features across different levels, poses challenges for network detection. Additionally, the complex agricultural production environment tends to generate many interfering negative samples, which significantly complicates pest and disease differentiation. To address these two issues, the YOLOv7‐PSAFP network structure was first proposed. Based on YOLOV7, the progressive Spatial Adaptive Feature Pyramid (PSAFP) was introduced. Second, a combination of the Varifocal Loss and Loss Rank Mining loss functions was used for calculating the object loss, which reduces the interference of useless negative examples during training. On the filtered‐plant‐village‐dataset and rice‐corn pest dataset, the mAP results of YOLOv7‐PSAFP were 84.7 and 93.3, which are 2.9 and 2.1 higher than the baseline model (YOLOv7), respectively. The code for this paper is located at https://github.com/DuLJ72/PSAFP .

Decentralized actor-critic (AC) is one of the most dominant algorithms for dealing with multiagent reinforcement learning (MARL) problems. However, exploration-efficient, sample-efficient, and communication-efficient are difficult to achieve simultaneously by existing decentralized AC methods. For this reason, this article develops a decentralized multiagent AC algorithm by incorporating entropy regularization to improve exploration with theoretical guarantees, referred to as multi-agent AC algorithm with entropy regularization (MACE). Moreover, we rigorously prove that MACE can achieve sample complexity $\mathcal {O}(\epsilon ^{-2}\ln \epsilon ^{-1})$ and communication complexity of $\mathcal {O}(\epsilon ^{-1}\ln \epsilon ^{-1})$ , which match the best complexities at present. Finally, the performance of MACE is also evaluated on reinforcement learning (RL) tasks. The experimental results show that the proposed algorithm achieves better exploration efficiency than state-of-the-art decentralized AC-type algorithms.

Abstract Weed recognition is an inevitable problem in smart agriculture, and to realise efficient weed recognition, complex background, insufficient feature information, varying target sizes and overlapping crops and weeds are the main problems to be solved. To address these problems, the authors propose a real‐time semantic segmentation network based on a multi‐branch structure for recognising crops and weeds. First, a new backbone network for capturing feature information between crops and weeds of different sizes is constructed. Second, the authors propose a weight refinement fusion (WRF) module to enhance the feature extraction ability of crops and weeds and reduce the interference caused by the complex background. Finally, a Semantic Guided Fusion is devised to enhance the interaction of information between crops and weeds and reduce the interference caused by overlapping goals. The experimental results demonstrate that the proposed network can balance speed and accuracy. Specifically, the 0.713 Mean IoU (MIoU), 0.802 MIoU, 0.746 MIoU and 0.906 MIoU can be achieved on the sugar beet (BoniRob) dataset, synthetic BoniRob dataset, CWFID dataset and self‐labelled wheat dataset, respectively.

Federated meta-learning (ML) is a promising optimization framework for the intelligent Internet of Things (IoT). However, the generalization ability of existing federated ML is limited because it is a bilayer structure, which has a more complex loss landscape. Moreover, the loss landscape of bilevel optimization has more saddle points and sharp points, which may lead to different generalization performances. Therefore, how to choose an optimal point is crucial for improving the generalization ability of federated ML. For this reason, this article proposes a provable federated ML algorithm by using the sharpness-aware minimization technique, referred to as FedAvg-sharp-MAML (FSM). Furthermore, we rigorously analyse the convergence and generalization bound of FSM. Specifically, when local iteration rounds$T=1$, the rate of$O(1/K)$can be achieved, where K is the number of global iterations. Furthermore, this rate can match the Per-Fedavg method. Meanwhile, we achieve a better generalization bound than the state of the art federated ML, where PAC-Bayesian generalization bounds are introduced in our analysis. Finally, we conduct some experiments to verify the performance of FSM. The experimental results show that the FSM has good generalization performance compared to the existing federated ML algorithms.

In multiagent reinforcement learning, policy evaluation is a central problem. To solve this problem, decentralized temporal-difference (TD) learning is one of the most popular methods, which has been investigated in recent years. However, existing decentralized variants of TD learning often suffer from slow convergence due to the sensitive selection of learning rates. Inspired by the great success of adaptive gradient methods in the training of deep neural networks, this article proposes a decentralized adaptive TD$(\lambda )$learning algorithm for general$\lambda $with linear function approximation, referred to asD-AMSTD$(\boldsymbol {\lambda })$, which can mitigate the selective sensitivity of learning rates. Furthermore, we establish the finite-time performance bounds ofD-AMSTD$(\boldsymbol {\lambda })$under the Markovian observation model. The theoretical results show thatD-AMSTD$(\boldsymbol {\lambda })$can linearly converge to an arbitrarily small size of neighborhood of the optimal weight. Finally, we verify the efficacy ofD-AMSTD$(\boldsymbol {\lambda })$through a variety of experiments. The results show thatD-AMSTD$(\boldsymbol {\lambda })$outperforms existing decentralized TD learning methods.

Semantic communication (SC) is one of the promising solutions for future wireless communications due to its superior transmission efficiency. However, the semantic information cannot be transmitted accurately under noisy channels by the existing methods. For this reason, we propose a fairness-based transmission strategy for STAR-RIS assisted SC systems. On this basis, we investigate two operating protocols of STAR-RIS, energy splitting (ES) and mode switching (MS). More specifically, we maximize the minimum signal-to-noise ratio (SNR) of the users by jointly optimizing the active beamforming and the passive beamforming under the constraint of limited power at the base station (BS). To tackle this max-min optimization problem, for ES, we develop a double-loop iterative algorithm by using the successive convex approximation (SCA) and penalty function methods. For MS protocol, we further add an additional penalty in the objective function to address the optimization problem. Moreover, we rigorously prove that the proposed algorithm can converge to a locally optimal solution. At last, we conduct various experiments to verify the performance of the proposed algorithms. Simulation experiments demonstrate that our algorithm outperforms other benchmark methods in fairness and semantic similarity.

Adaptive gradient algorithms have recently become extremely popular because they have been applied successfully in training deep neural networks, such as Adam, AMSGrad, and AdaBound. Despite their success, however, the distributed variant of the adaptive method, which is expected to possess a rapid training speed at the beginning and a good generalization capacity at the end, is rarely studied. To fill the gap, a distributed adaptive subgradient algorithm is presented, called D-AdaBound, where the learning rates are dynamically bounded by clipping the learning rates. Moreover, we obtain the regret bound of D-AdaBound, in which the objective functions are convex. Finally, we confirm the effectiveness of D-AdaBound by simulation experiments on different datasets. The results show the performance improvement of D-AdaBound relative to existing distributed online learning algorithms.

This paper considers a decentralized online submodular maximization problem over time-varying networks, where each agent only utilizes its own information and the received information from its neighbors. To address the problem, we propose a decentralized Meta-Frank-Wolfe online learning method in the adversarial online setting by using local communication and local computation. Moreover, we show that an expected regret bound of $O(\sqrt{T})$ is achieved with $(1-1/e)$ approximation guarantee, where $T$ is a time horizon. In addition, we also propose a decentralized one-shot Frank-Wolfe online learning method in the stochastic online setting. Furthermore, we also show that an expected regret bound $O(T^{2/3})$ is obtained with $(1-1/e)$ approximation guarantee. Finally, we confirm the theoretical results via various experiments on different datasets.

Named Data Networking (NDN) can effectively deal with the rapid development of mobile video services. For NDN, selecting a suitable forwarding interface according to the current network status can improve the efficiency of mobile video communication and can also avoid attacks to improve communication security. For this reason, we propose a stochastic adaptive forwarding strategy based on deep reinforcement learning (SAF-DRL) for secure mobile video communications in NDN. For each available forwarding interface, we introduce the twin delayed deep deterministic policy gradient algorithm to obtain a more robust forwarding strategy. Moreover, we conduct various numerical experiments to validate the performance of SAF-DRL. Compared with BR, RFA, SAF, and AFSndn forwarding strategies, the results show that SAF-DRL can reduce the delivery time and the average number of lost packets to improve the performance of NDN.

Cooperative caching among nodes is a hot topic in Content Centric Networking (CCN). However, the cooperative caching mechanisms are performed in an arbitrary graph topology, leading to the complex cooperative operation. For this reason, hierarchical CCN has received widespread attention, which provides simple cooperative operation due to the explicit affiliation between nodes. In this study, the authors propose a heuristic cooperative caching algorithm for maximising the average provider earned profit under the two‐level CCN topology. This algorithm divides the cache space of control nodes into two fractions for caching contents which are downloaded from different sources. One fraction caches duplicated contents and the other caches unique contents. The optimal value of the split factor can be obtained by maximising the earned profit. Furthermore, they also propose a replacement policy to support the proposed caching algorithm. Finally, simulation results show that the proposed caching algorithm can perform better than some traditional caching strategies.

Many problems in the Internet of Things (IoT) can be regarded as online optimization problems. For this reason, an online-constrained problem in IoT is considered in this article, where the cost functions change over time. To solve this problem, many projected online optimization algorithms have been widely used. However, the projections of these algorithms become prohibitive in problems involving high-dimensional parameters and massive data. To address this issue, we propose a Frank- Wolfe Adam online learning algorithm called Frank-Wolfe Adam (FWAdam), which uses a Frank-Wolfe method to eschew costly projection operations. Furthermore, we first give the convergence analysis of the FWAdam algorithm, and prove its regret bound to O(T3/4) when cost functions are convex, where T is a time horizon. Finally, we present simulated experiments on two data sets to validate our theoretical results.

The information name replaces traditional IP address as the identity of the network transmission is a typical feature of Information-Centric Networking (ICN). Therefore, designing efficient lookup algorithms of information names becomes a new challenge. For this reason, we propose an efficient name lookup structure for ICN, called Learned Bloom-Filter Lookup, which combines Recurrent Neural Networks (RNN) with standard Bloom filter to improve lookup efficiency. In our scheme, RNN trains the element set and non-element set, which are used to obtain the pre-filtering of names. Moreover, we look up the contents by using the backup Bloom filter, which can improve the accuracy of the search. In addition, we evaluate the performance of the proposed algorithm by experimental simulations. Compared with the Bloom-Hash method, our results show that our method can reduce the false positive rate. Furthermore, the memory required by our method is less than the Bloom-Hash method.

Mobile Microlearning, a novel fusion form of the mobile Internet, cloud computing, and microlearning, becomes more prevalent in recent years. However, its high deployment and operational costs make energy saving in cloud become a concerning issue. In this paper, to save energy consumption, a resource deployment approach to cloud service provision for Mobile Microlearning is proposed. Chinese Lexical Analysis System and Dynamic Term Frequency-Inverse Document Frequency (D-TF-IDF) are adopted to implement resource classification. Resources are deployed to the 2-tier cloud architecture according to the classification results. Grey Wolf Optimization (GWO) algorithm is used to forecast real-time energy consumption per byte. The simulation results show that, compared to traditional algorithm, the classification accuracy of small sample categories was significantly improved; the forecast energy consumption value and the standard values are 7.67% in private cloud and 2.93% in public cloud; the energy saving reaches 2.22% to 16.23% in 3G and 7.35% to 20.74% in Wi-Fi.

Computing offloading of mobile devices (MDs) through cloud is a greatly effective way to solve the problem of local resource constraints. However, cloud servers are usually located far away from MDs leading to a long response time. To this end, edge cloud servers (ECSs) provide a shorter response time due to being closer to MDs. In this paper, we propose a computing offloading game for MDs and ECSs. We prove the existence of a Stackelberg equilibrium in the game. In addition, we propose two algorithms, F‐SGA and C‐SGA, for delay‐sensitive and compute‐intensive applications, respectively. Moreover, the response time is reduced by F‐SGA, which makes decisions quickly. An optimal decision is obtained by C‐SGA, which achieves the equilibrium. Both algorithms above proposed can adjust the computing resource and utility of system users according to parameters control in computing offloading. The simulation results show that the game significantly saves the computing resources and response time of both the MD and the ECSs during the computing offloading process.

Information-Centric Networking (ICN) is a clean-slate networking architecture that puts information is focus instead of addressed hosts. Construction of content retrieval model to estimate the delivery performance is challenging in this ICN-based Mobile Ad hoc Networks (MANETs). In this paper, we propose a novel content retrieval model (PRCRM) for Information-Centric MANETs (ICMs) in 2-dimensional case. By investigating the distribution of content popularity, receiver-driven mechanism, content caching and replacement mechanism and generalized mobility model in 2-dimensional space, PRCRM constructs a novel content retrieval model based on the content hit/miss probability to estimate the content retrieval-related performance. We evaluate PRCRM by comparing its performance with another state of the art solution in terms of RTT and throughput. Simulation results demonstrate PRCRM's rationality and validity and it is shown that PRCRM is available to analyze content retrieval in ICMs.