The data scaling law has significantly enhanced large multi-modal models (LMMs) performance across various downstream tasks. However, in the domain of perceptual video quality assessment (VQA), the potential of data scaling remains unprecedented due to the scarcity of labeled resources and the insufficient scale of datasets. To address this, we propose OmniVQA, a framework designed to efficiently build high-quality, machine-dominated synthetic multi-modal instruction databases (MIDBs) for VQA. We then scale up to create OmniVQA-Chat-400K, the largest dataset in the VQA field concurrently. Our focus is on the technical and aesthetic quality dimensions, with abundant in-context instruction data to provide fine-grained VQA knowledge. Additionally, we build the OmniVQA-MOS-20K dataset to enhance the model's quantitative quality rating capabilities. We then introduce a complementary training strategy that effectively leverages the knowledge from datasets for different tasks. Furthermore, we propose the OmniVQA-FG (fine-grain)-Benchmark to evaluate the fine-grained performance of models. Our results demonstrate that our models achieve state-of-the-art performance in both tasks.

In 6G-enabled Industrial Internet of Things (IIoT) systems, edge intelligence faces heterogeneous resource availability and highly uncertain operating conditions, which undermine semantic fault tolerance and low-latency inference. To this end, we propose a resilient edge intelligence framework with training–inference co-design, based on a system-level Non-Parametric Mixture of Experts (NP-MoE) in which each expert is instantiated as an execution-capable inference pipeline deployed on an edge node, and a latency-aware gating policy activates only a subset of experts per request for conditional computation. To avoid gradient-heavy parametric MoE training and gating updates, the framework uses lightweight modality-specific backbones together with a non-parametric semantic-evidence aggregation mechanism based on feature memory banks and statistical semantic-score alignment, where “non-parametric” refers to the absence of additional trainable gating or fusion parameters rather than parameter-free feature backbones, enabling retraining-free and robust multimodal inference under modality missingness. Across the lifecycle, split learning serves as a resource mapper to enable feasible model partitioning in resource-constrained edge environments, while the deployed experts form a distributed expert pool for elastic inference scheduling. Then the joint cooptimization of model partitioning and multi-domain resource allocation is formulated as a MINLP and solved via a hybrid graph-theoretic and convex optimization method to minimize lifecycle latency. Experiments on industrial datasets show that the proposed framework achieves a favorable accuracy–latency trade-off and remains robust under extreme modality missingness, outperforming representative multimodal fusion baselines.

With the rapid advancement of generative models, the realism of AI-generated images has significantly improved, posing critical challenges for verifying digital content authenticity. Current deepfake detection methods often depend on datasets with limited generation models and content diversity that fail to keep pace with the evolving complexity and increasing realism of the AI-generated content. Large multimodal models (LMMs), widely adopted in various vision tasks, have demonstrated strong zero-shot capabilities, yet their potential in deepfake detection remains largely unexplored. To bridge this gap, we present DFBench, a large-scale DeepFake Benchmark featuring (i) broad diversity, including 540,000 images across real, AI-edited, and AI-generated content, (ii) latest model, the fake images are generated by 12 state-of-the-art generation models, and (iii) bidirectional benchmarking and evaluating for both the detection accuracy of deepfake detectors and the evasion capability of generative models. Based on DFBench, we propose MoA-DF, Mixture of Agents for DeepFake detection, leveraging a combined probability strategy from multiple LMMs. MoA-DF achieves state-of-the-art performance, further proving the effectiveness of leveraging LMMs for deepfake detection. Database and codes are publicly available at https://github.com/IntMeGroup/DFBench.

The rapid evolution of Multi-modality Large Language Models (MLLMs) is driving significant advancements in visual understanding and generation. Nevertheless, a comprehensive assessment of their capabilities, concerning the fine-grained physical principles especially in geometric optics, remains underexplored. To address this gap, we introduce GOBench, the first benchmark to systematically evaluate MLLMs' ability across two tasks: 1) Generating Optically Authentic Imagery and 2) Understanding Underlying Optical Phenomena. We curate high-quality prompts of geometric optical scenarios and use MLLMs to construct the GOBench-Gen-1k dataset. We then organize subjective experiments to assess the generated imagery based on Optical Authenticity, Aesthetic Quality, and Instruction Fidelity, revealing MLLMs' generation flaws that violate optical principles. For the understanding task, we apply crafted evaluation instructions to test the optical understanding ability of eleven prominent MLLMs. The experimental results demonstrate that current models face significant challenges in both optical generation and understanding. The top-performing generative model, GPT-4o-Image, cannot perfectly complete all generation tasks, and the best-performing MLLM model, Gemini-2.5Pro, attains a mere 37.35% accuracy in optical understanding. Database and codes are publicly available at: https://github.com/aiben-ch/GOBench.

Large Multi-modality Models (LMMs) have made significant progress in visual understanding and generation, but they still face challenges in General Visual Editing, particularly in following complex instructions, preserving appearance consistency, and supporting flexible input formats. To study this gap, we introduce RISEBench, the first benchmark for evaluating Reasoning-Informed viSual Editing (RISE). RISEBench focuses on four key reasoning categories: Temporal, Causal, Spatial, and Logical Reasoning. We curate high-quality test cases for each category and propose an robust evaluation framework that assesses Instruction Reasoning, Appearance Consistency, and Visual Plausibility with both human judges and the LMM-as-a-judge approach. We conducted experiments evaluating nine prominent visual editing models, comprising both open-source and proprietary models. The evaluation results demonstrate that current models face significant challenges in reasoning-based editing tasks. Even the most powerful model evaluated, GPT-image-1, achieves an accuracy of merely 28.8%. RISEBench effectively highlights the limitations of contemporary editing models, provides valuable insights, and indicates potential future directions for the field of reasoning-aware visual editing. Our code and data have been released at https://github.com/PhoenixZ810/RISEBench.

6G networks will serve as the key infrastructure for the converged world of human-machine-object-intelligence to support large-scale multiple information interactions. In order to meet the future society’s demand for large-scale, intelligent, and low-latency communications, 6G network architecture must be highly resilient and adaptive. However, with the dramatic increase in the number of nodes and the risk of various interferences, attacks, and failures, 6G networks are facing increasing challenges, especially how to maintain the reliability and security of the network under extreme conditions. Focusing on the resilience optimization of future 6G distributed network architecture, this article proposes an architectural entropy-based network resilience characterization and assessment model to address the multidimensional challenges encountered by 6G networks in complex environments. The model combines metrics such as eigenvectors, K-shells, and closeness centrality to quantify network destructive power and resilience. On this basis, the Effective edge addition method based on particle swarm optimization and genetic algorithm (EA-PSOGA) is proposed, aiming at malicious attacks or random fault conditions, to improve the resilience and resilience of the network by optimizing the topology. Simulation experimental results show that EA-PSOGA outperforms other algorithms in enhancing the resilience and security of 6G networks in response to sudden attacks, which provides a solid theoretical support and technical foundation for the application of 6G networks in future communications and distributed computing.

The emergence of 6G applications such as artificial intelligence, augmented reality, and digital twins has imposed stringent requirements on the high reliability, low latency, and energy efficiency of computing networks. Therefore, in this paper, we propose a novel high-reliability resource scheduling optimization method for heterogeneous computing networks, leveraging hypergraph neural networks (HGNNs) and coded distributed computing (CDC). We first construct a multidimensional resource representation model for computing networks based on hyper-networks, effectively illustrating heterogeneous nodes and their interactions within computing networks. Then, targeting the need for collaborative optimization of task offloading, as well as computing, communication, and caching resources in cloud-edge-end computing networks, we propose the collaborative task offloading and resource allocation (CTOHRA) problem, which minimizes the total task processing delay. By incorporating CDC, we enhance resilience against edge node failures and unstable network links. To solve this problem, we utilize hypergraph neural networks to capture high-order correlations and improve the accuracy of dynamic resource scheduling, and combine particle swarm optimization (PSO) to handle discrete variables and find the global optimal solution. Extensive simulations show that the proposed method can significantly improve the task success rate, reduce the average system latency, and minimize energy consumption, especially under unfavorable network conditions.

Low Earth Orbit (LEO) satellite networks, with their advantages of low latency, wide coverage, and high data rates, have become a core component of 6G networks. However, due to the complexity and interactivity of satellite networks, they face higher uncertainty and vulnerability, and there is currently a lack of effective frameworks for survivability and recovery in harsh environments. This paper proposes a resilience evaluation and optimization model for large-scale LEO satellite networks, based on entropy theory and the minimum-cut theorem. Considering the time-varying characteristics of satellite networks, a spatiotemporal extended graph is used to model the network topology, along with a network communication model that incorporates node state transitions. By analyzing various factors such as topological structure, node functionality, and link quality in real-time, entropy theory is leveraged to provide a dynamic and comprehensive evaluation of LEO satellite network resilience. In response to network failures, multiple attributes, including node load and computational capacity, link signal-to-noise ratio (SNR), and link duration, are integrated into an optimization decision-making framework, formulating a resilient recovery optimization problem. To address this, a network resilience bottleneck identification and global load optimization-based edge augmentation strategy is proposed to enhance the network’s adaptive and rapid recovery capabilities. Simulation results demonstrate that the proposed resilience evaluation model effectively reflects the resilience performance of LEO satellite networks in complex and dynamic environments, and the proposed optimization method significantly enhances network resilience.

The rapid expansion of Low Earth Orbit (LEO) satellite networks and has heightened the need for preliminary evaluations of network capacity. However, accurately modeling Multi-Layer Satellite Networks (MLSN) capacity poses significant challenges due to dynamic topologies, complex traffic flows, and multi-layer configurations, etc. Therefore, in this paper we propose a capacity analysis model for large-scale, two-layered satellite networks. Initially, based on the current neighborbased inter-satellite link (ISL) establishment rules, we establish a two-layered collaborative satellite network transmission model with delay constraints, where higher-layer satellites assist with communication to mitigate delays caused by excessive relays in lower-layer satellites, thereby reducing the number of hops for long-distance transmissions. Subsequently, we derive a capacity analysis model by comparing traffic demand with network transmission capabilities. By employing a region-based Binary Point Process (BPP) to model gateway satellites, we provide analytical expressions for both service and network transmission capacity. Simulation and analysis results indicate that both the average service transmission capacity and the network transmission capacity exhibit an asymptotic relationship with the scale of LEO satellites, represented as Θ(1/Na). Furthermore, simulation results reveal optimal configurations for the number of LEO satellites, gateways, LEO/Geostationary Earth Orbit (GEO) switching points, and the distribution density of gateway satellites that maximize transmission capacity. The results also indicate that current neighbor-based link establishment rules are inadequate for future large-scale satellite networks, necessitating the development of new ISL mechanisms.

New services such as Extended Reality (XR) and holographic communication in the future 6G era will require a lot of computing power. With the increasing trend of terminal mobility, research on traditional task offloading problems in the Mobile Edge Computing (MEC) paradigm needs to be extended from quasi-static scenarios to mobile scenarios. In this article, we focus on the problem of mobility-aware task offloading with dependency guarantee. We have established a nonlinear integer programming task offloading problem that maximizes task utility by optimizing the task offloading decision while taking into account user mobility and subtask dependencies. Confronted with a dynamic user-centric service network, we use the temporal graph to characterize the dynamically changing set of offloading nodes including base stations surrounding the user and the terminal device. Then, we abstract the task offloading problem as a graph homomorphism problem, considering the task offloading decision is a one-to-many mapping relationship. To solve this problem, we have developed the Optimized A*(star) Algorithm (OASA) that incorporates the delayed offloading strategy and the serial offloading strategy, specifically designed for user mobility and task dependency. Simulation results show that the proposed algorithm outperforms four benchmark algorithms in terms of three metrics.

Reconfigurable intelligent surface (RIS) has the characteristics of programmable wireless environments, and device-to-device (D2D) communication can fully utilize network resources. Therefore, in this article, RIS and D2D technologies are cleverly integrated into multicast multiple-input-multiple-output (MIMO) communication system. The goal is to maximize the multigroup multicast sum-rate (SR) by jointly optimizing the orthogonal resource block (ORB) reusing coefficients, beamforming vector of the access point (AP), power assignment of D2D pairs, and reflection coefficients of RIS while meeting the transmission power of the AP, rate requirements of user equipments (UEs), ORBs reusing, and reflecting elements modulus-one constraints. This is a mixed-integer nonlinear programming (MINLP) problem. To solve this problem, we propose a new block coordinate descent (BCD)-based algorithm to decouple the original optimization problem into four subproblems. First, a new directional search algorithm is proposed to effectively optimize ORB reuse coefficients. Second, based on Cauchy-Schwarz inequality, the value of the beamforming vector is calculated. Next, successive convex approximation (SCA)-based method is adopted to optimize the power assignment of D2D pairs on each ORB. Finally, we develop the auxiliary relaxation approach (ARA) to optimize the reflection coefficients of RIS. The simulation results confirm that compared to existing designs and algorithms, the proposed design can greatly improve the multicast SR, the proposed algorithm is reasonable and efficient.

Reconfigurable Intelligent Surface (RIS) is a technology that can intelligently control the signal transmission link to enhance the signal transmission quality. Rate Splitting Multiple Access (RSMA) technology can effectively restrain the same channel interference between users and achieve better system performance by flexibly controlling the rate allocation strategy and beamforming strategy. In this paper, we combine RSMA with RIS and propose an energy efficiency optimization algorithm for 6G multicell communication system. We firstly formulate an optimization problem which is to maximize energy efficiency of the system by jointly optimizing beamforming vectors of base stations (BSs), phase shifts matrix of RISs and rate allocation matrix of users. Then, in order to solve this nonconvex optimization problem, we decouple the original problem into two subproblems: beamforming optimization and phase shifts optimization. For the former, we use the successive convex approximation (SCA) method to get the solution, and for the latter, we adopt semidefinite relaxation and exterior point methods to solve. In addition, we analyze the complexity of the proposed algorithm. Simulation result shows that compared with space division multiple access (SDMA) and non-orthogonal multiple access (NOMA), the proposed mechanism in this paper can significantly improve system energy efficiency.

In 6G, the communication cost of large-scale distributed machine learning (DML) will be much higher than the computing cost, which will become a bottleneck restricting the development of DML. To solve this problem, a wireless large-scale DML architecture combined with blockchain (WLDMLB) for 6G networks is proposed, where the distributed nodes involved in DML are divided into shards and a layered adaptive cascaded architecture is used in each shard to reduce the communication overhead. To reduce the system energy, improve training efficiency and achieve on-demand networking, a joint optimization model of the number of shards, network topology, and allocation of computing resources is established to ensure that the model can run efficiently on different devices. Then, a closed-form expression of one-round training delay and energy is derived. The optimal number of shards, the optimal network topology and the optimal computing resource allocation are further analysed. In addition, a main-shards blockchain architecture with the directed acyclic graph (DAG) and practical Byzantine fault tolerance (PBFT) consensus is proposed to ensure the trusted sharing of model and ensure system scalability. Simulation results show that the algorithm can greatly reduce the communication overhead, one round-training delay and energy of DML.

Blockchain is an effective technology that enables secure data sharing and trusted energy trading in smart grids (SGs). The consensus algorithm plays a key role in the blockchain’s security and consistency. However, due to nonconcurrency, long consensus time, low throughput, and high transaction fees, current consensus algorithms are not suitable for SGs, where there is a large-scale distributed wireless field area network (FAN) with limited resources and frequent small transactions. In this article, we propose a performance analysis model and a new consensus algorithm for the tangle in the FAN. First, we propose an analytical analysis model for the consensus time and probability of a successful parasitic chain attack on the FAN by comprehensively considering the number of data transmission hops, communication protocols, computing resources, and checked tips. Furthermore, based on the analysis, a new consensus algorithm is proposed by clustering nodes cooperating with each other to complete the Proof of Work (PoW). Finally, we show the accuracy of the analysis results by comparing them with the simulation results. Additionally, we show that the proposed consensus algorithm can reduce the required consensus time by approximately 50%.

Protein classification is a crucial field in bioinformatics. The development of a comprehensive tool that can perform feature evaluation, visualization, automated machine learning, and model interpretation would significantly advance research in protein classification. However, there is a significant gap in the literature regarding tools that integrate all these essential functionalities. This paper presents iProps, a novel Python-based software package, meticulously crafted to fulfill these multifaceted requirements. iProps is distinguished by its proficiency in feature extraction, evaluation, automated machine learning, and interpretation of classification models. Firstly, iProps fully leverages evolutionary information and amino acid reduction information to propose or extend several numerical protein features that are independent of sequence length, including SC-PSSM, ORDip, TRC, CTDC-E, CKSAAGP-E, and so forth; at the same time, it also implements the calculation of 17 other numerical features within the software. iProps also provides feature combination operations for the aforementioned features to generate more hybrid features, and has added data balancing sampling processing as well as built-in classifier settings, among other functionalities. Thus, It can discern the most effective protein class recognition feature from a multitude of candidates, utilizing three automated machine learning algorithms to identify the most optimal classifiers and parameter settings. Furthermore, iProps generates a detailed explanatory report that includes 23 informative graphs derived from three interpretable models. To assess the performance of iProps, a series of numerical experiments were conducted using two well-established datasets. The results demonstrated that our software achieved superior recognition performance in every case. Beyond its contributions to bioinformatics, iProps broadens its applicability by offering robust data analysis tools that are beneficial across various disciplines, capitalizing on its automated machine learning and model interpretation capabilities. As an open-source platform, iProps is readily accessible and features an intuitive user interface, ensuring ease of use for individuals, even those without a background in programming.

Abstract In a distributed wireless network with a large number of nodes, competitive access of nodes may result in the deterioration of throughput and energy. Therefore, in this paper we propose an intelligent access algorithm based on the mean field game (MFG). First, we formulate the competitive access process between nodes as a game Query ID="Q1" Text="Please check and confirm that the authors and their respective affiliations have been correctly identified and amend if necessary." process by a stochastic differential game model, which maximizes the energy efficiency of nodes and obtain the optimal behavior strategy while meeting the requirements of channel access. However, as the number of nodes increases, the dimension of the matrix used to characterize the interaction between nodes becomes too large, which increases the complexity of the solution procedure. Therefore, we introduce the MFG and the interaction between nodes can be approximately transformed into the interaction between nodes and the mean field, which not only reduces the complexity, but also reduces the computational overhead. In addition, the HJB-FPK equation is solved to obtain the Nash equilibrium of the MFG. Finally, a backoff strategy based on the Markov model is proposed, and the node obtains the corresponding backoff strategy according to the network situation and its own state. Simulation results show that the proposed algorithm has good performances on optimizing network throughput and energy efficiency for a large scale multi-hop wireless network.

Sharing is undoubtedly one of the most important features of 6G networks, and blockchain can provide an extended trust-as-a-service (TAAS) distributed sharing solution for 6G networks. However, as a special distributed technology system, blockchain naturally needs to face the “impossible triangle” problem: in the case of ensuring security and decentralization, scalability will inevitably become the Achilles heel of the blockchain system. This article will design a high-throughput blockchain system for 6G networks to achieve trusted sharing and efficient scheduling of network infrastructure, assist future networks in integrating an open Internet architecture, and realize the combination of openness and distributed control. We adopt a shard blockchain design and incorporate digital twins and federated learning; formulate an optimization model for maximizing the throughput of the blockchain network by getting the optimal number of shards, also analyzing other factors affecting the throughput such as service distribution; propose a security performance analysis model for the blockchain network to describe the measures taken against the Byzantine attacks on the network. Analysis and simulation results show that the transaction throughput of the proposed method can reach more than 30 times larger than that of a nonsharding scheme. They also show that when one-third of the nodes in the system are attacked, the consensus of the system is hardly affected; even if the number of nodes being attacked at the same time reaches half of the total number of nodes, the probability of the occurrence of failed shards is still less than$10^{-4}$, and the system still has good survivability.

Abstract In the large scale Internet of things, edge gateway (EG) deployment is used to find the minimal number of gateways required in the network and their optimal locations under the design constraints to meet different service requirements, which is one significant issue for improving network performances. We formulate the EG deployment problem as a k ‐median problem with some constraints, which is known as a NP‐hard problem. For that, we propose a new heuristic based on simulated annealing with external penalty function (SA‐AEP) for its solution. The external penalty function is used to transform the multi‐constrained optimization problem into a single constraint one. Also the complexity of proposed heuristic algorithm is analyzed. In addition, in order to evaluate the proposed algorithm, we compare it with an adaptive variable‐length particle swarm optimization algorithm with varying lengths. Simulation results show that the proposed SA‐AEP algorithm has much better performances on network cost and efficiency than other algorithms when the higher service rate is required.

Network virtualization has emerged as a significant trend to solve the issues caused by ossification of traditional network. Under the circumstance of network virtualization, substrate network and virtual network are inextricably interdepending each other. The substrate network serves many virtual networks. Substrate network faults may lead to different virtual network faults. A service''s failure may introduce additional influence on other services. Therefore, it has become a big challenge to predict when and where a fault happens in the network. In this paper, we propose a fault diagnosis method by deep learning to predict the failure of virtual network. Our deep learning model enables the earlier failure prediction by the Long Short-Term Memory (LSTM) network, which discovers the long-term features of network history data. Simulation results show that the proposed method performs well on faults prediction.

This paper presents a supplementary control of doubly-fed induction generator (DFIG) to damp the inter-area oscillation. Firstly, the small signal dynamic model of DFIG is established. Secondly, a supplementary damping controller of DFIG is investigated. Two input signals are selected, one is the tie-line power measured from the local area and the other is the voltage phase angles obtained from the Wide Area Measurement Signals. Considering the capability of decoupled active power and reactive power control of DFIG, the supplementary damping controller is added to the active power control loop and reactive power control loop separately, and the effects of the control are analyzed. Finally, the proposed controller's performance is accessed through eigenvalue analysis and time domain simulation.

We present a new method for data gathering that maximizes lifetime for wireless sensor networks. It involves three parts. First, nodes organize themselves into several static clusters by the Hausdorff clustering algorithm based on location, communication efficiency and network connectivity. Second, clusters are formed only once but the role of cluster-head is optimally scheduled among the cluster members. We formulate the cluster-head scheduling that maximizes the network lifetime as an integer programming problem and propose a greedy algorithm for its solution. Third, after cluster-heads are selected, they form a backbone network to periodically collect, aggregate, and forward data to the base station, where a minimum energy (cost) routing is used. Comparing with other known methods, significant lifetime extension is obtained with the use of this method.