While service-oriented low Earth orbit (LEO) satellite edge computing (LSEC) frameworks enable diverse edge services for user tasks, the heterogeneous distribution of terrestrial users causes substantial imbalance in computational task loads across satellites. This asymmetric workload leads to inefficient resource utilization and degraded edge computing performance. To address these challenges, we propose a joint optimization framework that integrates edge service deployment and task offloading, supported by service popularity analysis and spatiotemporal user-task modeling. The framework employs a two-timescale design: at the large timescale, an improved atomic orbital search (iAOS) heuristic dynamically optimizes service placement, configuration, and resource allocation; at the small timescale, a direction-selective multi-agent double deep Q-network (DS-MDDQN) leverages deep reinforcement learning to route tasks to the most suitable processing nodes. Extensive simulations show that our approach significantly outperforms six representative baselines in both user-perceived performance and system-level efficiency. Replacement studies further verify the effectiveness of each component: iAOS enhances resource utilization and reduces task failure through optimized service deployment, while DS-MDDQN mitigates network dynamics and lowers task completion latency via adaptive task offloading.

In this paper, we propose a lightweight privacy-preserving convolutional neural network framework for military vehicle images classification (LPP-CNN). Existing target classification methods primarily focus on improving accuracy and recognition efficiency but often overlook security threats during data transmission and processing, making them unsuitable for high-risk battlefield scenarios. Although some approaches incorporate image security through homomorphic encryption, the low efficiency and operational complexity of such techniques hinder their applicability in battlefield environments. To this end, we design a solution that integrates additive secret sharing with edge computing by encrypting images into two ciphertexts, which are then processed independently by two edge servers to prevent data leakage during upload. The encrypted data is subsequently processed by the LPP-CNN embedded with secure computation protocols. Finally, results are combined and decrypted to achieve target classification. This method ensures efficient and accurate military vehicle classification while significantly enhancing data privacy and security. Theoretical analysis demonstrates improved response speed and reduced communication overhead. Experimental evaluations show that under a classification accuracy not less than 94%, the computational efficiency of SComp, SReLU, and SMaxPool protocols increase by 32, 4, and 0.5 times, respectively, while communication costs decrease by 2, 2, and 13 times. The overall framework achieves a 33% improvement in computational efficiency compared to existing solutions. Compared to state-of-the-art methods, our framework not only meets battlefield requirements for high-precision object classification but also substantially strengthens data privacy and security, aligning more effectively with real-world military application demands.

Videos inherently contain complex temporal dynamics across various spatial directions, often entangled in ways that obscure effective dynamic extraction. Previous studies typically process video spatiotemporal features without disentangling, which hampers their ability to extract dynamic information. Additionally, the extraction of dynamics is disrupted by transient high-dynamic information in video sequences, e.g., noise or flicker, which has received limited attention in the literature. To tackle those problems, this paper proposes the Disentangling and Filtering Dynamics Network (DFDNet). Firstly, to disentangle the interwoven dynamics, DFDNet decomposes the spatially encoded video sequences into lower dimensional sequences. Secondly, a learnable threshold filter is proposed to eliminate the transient high-dynamic information. Thirdly, the model incorporates an MLP to extract the temporal dependencies from the disentangled and filtered sequences. DFDNet demonstrates competitive performance across four chosen datasets, including both low and high-resolution videos. Specifically, on the low-resolution Moving MNIST dataset, DFDNet achieves a 19% improvement on MSE over the previous state-of-the-art model. On the high-resolution SJTU4K dataset, it outperforms the previous state-of-the-art model by 10% on the LPIPS metric under similar inference time.

The pretrain+fine-tune paradigm is foundational for deploying large language models (LLMs) across various downstream applications. Within this framework, Low-Rank Adaptation (LoRA) stands out for its parameter-efficient fine-tuning (PEFT), producing numerous reusable task-specific LoRA adapters. However, this approach requires explicit task intention selection, posing challenges for autonomous task sensing and switching during inference with multiple existing LoRA adapters embedded in a single LLM. In this work, we introduce MeteoRA (Multiple-Tasks embedded LoRA), a scalable and efficient framework that reuses multiple task-specific LoRA adapters into the base LLM via a full-mode Mixture-of-Experts (MoE) architecture. This framework also includes novel MoE forward acceleration strategies to address the efficiency challenges of traditional MoE implementations. Our evaluation, using the LlaMA2-13B and LlaMA3-8B base models equipped with 28 existing LoRA adapters through MeteoRA, demonstrates equivalent performance with the traditional PEFT method. Moreover, the LLM equipped with MeteoRA achieves superior performance in handling composite tasks, effectively solving ten sequential problems in a single inference pass, thereby demonstrating the framework's enhanced capability for timely adapter switching.

Currently, Large Language Model (LLM), as a typical Generative AI (GAI) implementation, has been deployed to lead to artificial general intelligence for numerous applications. Numbers of LLM releases, like ChatGPT, Llama, Gemini, ChatGLM, can online interact with users to give instructions against different kinds of queries. Moreover, LLMs can process multi-model data, like text, image, voice, video, to support the workings of robots, expert systems, and operation control systems. Due to its merits, LLM in theory can help the operation of Unmanned Aerial Vehicle (UAV) communication networks, where each UAV is constrained by its size, weight and power (SWAP) to provide limited data collection functions in the air. In this paper, we apply LLM to online control the operation of UAVs by optimizing the trajectory of UAV , to lead to UAV energy efficiency. We firstly provide well established air-ground wireless communication model within UAV and ground terminals (GTs) to provide API library to LLM to help its closed-loop reasoning. Further, we establish Chain-of-Thought (CoT) and generate optimizing instructions to support UAV communication via prompt engineering. In the end, such LLM en-powered UAV communication network will be validated via experiments. The numerical results validate that LLM can enable UAV communication networks to more adaptive to complicated environments to work with less energy consumption and higher energy efficiency.

Video prediction is an important yet challenging task that generates future frames based on previous observations. Despite recent progress, existing methods still suffer from motion blur, due to weak motion perception capabilities leading to uncertainty in motion direction. To address this, we propose a Motion Direction Awareness (MDA) mechanism inspired by the direction-selective mechanism in animal visual systems. Specifically, MDA can decompose complex motions into horizontal and vertical components, allowing dimension reduction and independent processing, thereby effectively enhancing motion perception and reducing uncertainty in predicted motion directions. Based on MDA, we design a multi-scale feature fusion network named MDANet for video prediction, which incorporates different scales of spatially encoded features in conjunction with MDA mechanism to extract the temporal evolution information of global and local spatial features. Extensive experiments on representative datasets demonstrate that MDANet can alleviate motion blurring, improving prediction accuracy and temporal consistency over state-of-the-art models. Furthermore, we validate the generalizability and effectiveness of our MDA mechanism by integrating it into other advanced models. The code is available at supplementary.

Routing optimization within a low Earth orbit (LEO) satellite broadband network (LSBN) has seen advancements through deep reinforcement learning (DRL) approaches in academia. Nonetheless, a crucial aspect often overlooked in these approaches pertains to the inference time of deep neural network (DNN) models during the routing of packets. Our investigation reveals that this oversight can significantly impair routing throughput in LSBN. In response, this paper innovatively proposes a decentralized flow-centric DRL approach, shifting the focus from routing individual packets to entire traffic flows. To align with the large-scale feature of LSBN, we embrace a fully-distributed architecture for flow-centric routing, which is modeled as a partially observable Markov decision process. In this construct, each satellite operates as an independent agent, locally classifying flows following a tailor-designed definition, and is responsible for forwarding a flow to an adjacent satellite based on its internal policy. Notably, the DNN inference is conducted only once on each agent to determine the route for the initial packet of a specific flow; subsequent packets are directed along the same route. Recognizing the potential impact of dynamic LSBN topologies on routing performance, we also introduce an adaptive flow routing update scheme. This scheme is completely free from LSBN environment modelling and aims to bolster the efficacy of the flow-centric approach. Comparative experiments showcase the superiority of the proposed approach over baseline algorithms across various metrics. Consequently, the flow-centric DRL approach can enable high-throughput traffic transmission for LSBN.

Video prediction is a challenging spatiotemporal prediction task that generates future frames based on historical observations. Although recently proposed deep learning-based methods significantly outperform legacy approaches, there still exist gaps between prediction and ground truth, primarily rooted in edge and motion blurring. On the one hand, since conventional performance metrics like Mean Square Error (MSE) and Structure Similarity Index Measure (SSIM) cannot decently evaluate this deficiency, we design a 3D Frequency Loss (3DFL) metric to better assess the similarity of predicted video frames. On the other hand, edge and motion blurring is mainly attributed to the predictive model's insufficient attention to high spatial frequency arising from rapid pixel value variations at object edges, and it is observed that shallow networks are more adept at capturing high spatial frequency information. Therefore, aiming to alleviate edge and motion blurring, we propose a novel video prediction model termed SDFNet that can extract and integrate both spatially encoded shallow and deep-level features. To accommodate SDFNet's multi-branch input structure, a frequency adaptive translator (FATranslator) is derived, which leverages involution operators to adaptively extract inter-frame temporal dependencies from different spatial encoding layers, and further mitigates motion blurring. Extensive experiments demonstrate that our proposed model achieves significant improvements in prediction accuracy and temporal consistency over the current state-of-the-art models on various benchmarks. The results highlight the importance of spatial frequency modeling for enhancing video prediction performance, contributing to the advancement of multimedia technologies.

Legacy computing offloading approaches are originally designed for the terrestrial networks with rather static topologies, and may not be appropriate for the next-generation LEO satellite broadband networks (LSBNs) featured with high dynamicity. This paper presents a multi-agent deep reinforcement learning (MADRL) algorithm for making edge computing multi-level offloading decisions in the LSBNs. Particularly, computing offloading is formulated as a partially observable Markov decision process (POMDP) based multi-agent decision problem. Each LEO satellite is an intelligent agent, either conducting a received edge computing task or forwarding it to its four neighboring satellite or the nearest cloud node on the ground. These agents are fully cooperative and their deep neural network models used to make offloading decisions share the same parameter values and are trained by the same replay buffer. A centralized training and distributed execution framework is utilized to ensure that globally optimized offloading decisions can be achieved based on local observations. Comparative simulation experiments for six representative offloading approaches show that the proposed MADRL aided approach outperforms the others regarding to decreasing edge computing task processing delay and increasing onboard compute resource utilization ratio. In addition, the convergence of this MADRL aided approach is also the best among the three DRL-based approaches.

Network traffic prediction can serve as a proactive approach for network resource planning, allocation, and management. Besides, it can also be applied for load generation in digital twin networks (DTNs). This paper focuses on background traffic prediction of typical local area networks (LANs), which is vital for synchronous traffic generation in DTN. Conventional traffic prediction models are firstly reviewed. The challenges of DTN traffic prediction are analyzed. On that basis, a spatiotemporal-enhanced recurrent neural network (RNN) based approach is elaborated to accurately predict the background traffic matrices of target LANs. Experiments compare this proposed model with four baseline models, including LSTM, CNN-LSTM, ConvLSTM, and PredRNN. The results turn out that the spatiotemporal-enhanced RNN model outperforms the baselines on accuracy. In particular, it can decrease the MSE of PredRNN more than 18%, with acceptable efficiency degradation.

Conventional computation offloading approaches are originally designed for ground networks, and are not effective for low earth orbit (LEO) satellite networks. This paper proposes a multi-agent deep reinforcement learning (MADRL) algorithm for making multi-level offloading decisions in LEO satellite networks. Offloading is formulated as a partially observable Markov decision process based multi-agent decision problem. Each satellite as an agent either conducts a received task, forwards it to neighbors, or sends it to ground clouds based on its own policy. These agents are independent and their deep neural networks to make offloading decisions share identical parameter values and are trained by using the same replay buffer. A centralized training and distributed executing mechanism is adopted to ensure that agents can make globally optimized offloading decisions. Comparative experiments demonstrate that the proposed MADRL algorithm outperforms the five baselines in terms of task processing delay and bandwidth consumption with acceptable computational complexity.

Cyber-Physical Systems (CPS) provide the foundation of our critical infrastructures, which form the basis of emerging and future smart services and improve our quality of life in many areas. In such CPS, sensor data is transmitted over the network to the controller, which will make real-time control decisions according to the received sensor data. Due to the existence of spoofing attacks (more specifically to CPS, false data injection attacks), one has to protect the authenticity and integrity of the transmitted data. For example, a digital signature can be used to solve this issue. However, the resource-constrained field devices like sensors cannot afford conventional signature computation. Thus, we have to seek for an efficient signature mechanism that can support the fast and continuous message authentication in CPS, while being easy to compute on the devices.

In order to meet the needs of real-time services in the low-orbit communication network, the paper draws on the related research of edge computing technology in the ground broadband network and applies it to the LEO satellite constellation communication network, sinking the processing power of the backend cloud center (CC) to the nodes close to the user terminals, thereby reducing the response delay of the task and the bandwidth consumption of the backhaul network in the LEO constellation networks. This paper proposes a sever selection strategy based on queuing theory and weighting method. For the computing offload requests submitted by frontend users, this strategy first regards the nearest accessing satellite edge computing node as the first choice to offload computing tasks; if it is not qualified, then synthetically consider various factors, such as the total energy consumption of the data transmission and calculation for the request, the load balancing among the computing nodes, and the response delay to the user, etc. These factors for candidate computing nodes are scored separately, and the all-around score is finally used to select the most suitable computation offloading node. Comprehensive simulation experiments show that, compared with another two approaches, the proposed strategy can ensure that user requests are well satisfied, and can reduce the average response delay in the range from 17% up to 34%.

The combination of edge computing and LEO satellite broadband network has the potential to provide ground users enhanced services with significantly lower latencies. This paper derives a offloading model and a user model of LEO satellite constellation networks and proposes a multi-level computation offloading scheme. This scheme designs a multilevel edge-projection pursuit (MLEPP) model and uses an optimization method to find the optimized projection direction in order to obtain the most suitable offloading choices. On the other hand, a discrete event-based Monte Carlo simulator is developed to evaluate the proposed computation offloading scheme. Comparative evaluation experiments have been conducted, and the results indicate that, in resource-restricted scenarios, using the derived multi-level offloading scheme can effectively and efficiently decrease the average request blocking probability, the average request corresponding delay, and keep the average request cost at a restricted level.

Abstract In this paper, we propose and evaluate fast, scalable approaches for solving the linear complementarity problems (LCP) arising from the fluid pressure equations with separating solid boundary conditions. Specifically, we present a policy iteration method, a penalty method, and a modified multigrid method, and demonstrate that each is able to properly handle the desired boundary conditions. Moreover, we compare our proposed methods against existing approaches and show that our solvers are more efficient and exhibit better scaling behavior; that is, the number of iterations required for convergence is essentially independent of grid resolution, and thus they are faster at larger grid resolutions. For example, on a 2563 grid our multigrid method was 30 times faster than the prior multigrid method in the literature.

Public key cryptographic primitive (e.g., the famous Diffie-Hellman key agreement, or public key encryption) has recently been used as a standard building block in authenticated key agreement (AKA) constructions for wireless sensor networks (WSNs) to provide perfect forward secrecy (PFS), where the expensive cryptographic operation (i.e., exponentiation calculation) is involved. However, realizing such complex computation on resource-constrained wireless sensors is inefficient and even impossible on some devices. In this work, we introduce a new AKA scheme with PFS for WSNs without using any public key cryptographic primitive. To achieve PFS, we rely on a new dynamic one-time authentication credential that is regularly updated in each session. In particular, each value of the authentication credential is wisely associated with at most one session key that enables us to fulfill the security goal of PFS. Furthermore, the proposed scheme enables the principals to identify whether they have been impersonated previously. We highlight that our scheme can be very efficiently implemented on sensors since only hash function and XOR operation are required.

In Public‐Key Cryptography (PKC) 2015, Bergsma et al . introduced an interesting one‐round key exchange protocol (which will be referred to as BJS scheme) with strong security in particular for perfect forward secrecy (PFS). In this study, the authors unveil a PFS attack against the BJS scheme. This would simply invalidate its security proof. An improvement is proposed to fix the problem of the BJS scheme with minimum changes.

In PKC 2015, Bergsma et al. introduced a generic one-round key exchange (ORKE) protocol (which is referred to as BJS) from digital signature (SIG) and simplified non-interactive key exchange (NIKE) without involving any identity. The BJS scheme is shown to satisfy extended Canetti and Krawczyk-PFS security if the NIKE is adaptive-CKS-light secure and the SIG is strongly secure against existential unforgeability under chosen message attacks. However, the BJS scheme cannot be instantiated with NIKE scheme with identity (e.g. the one proposed by Boneh and Zhandry in Crypto 2014). In this paper, we propose a much simpler generic construction for ORKE from NIKE and SIG. In particular, our scheme only makes weaker security assumptions on the underlying building blocks. Namely, we first show that the static-CKS-light security of NIKE, where the target identities are chosen by the adversary before seeing the system parameters, is sufficient for our construction. On the second, we observe that the SIG only needs to provide strong existential unforgeability under weak chosen message attacks for our construction. These results enable our proposal to have more concrete instantiations which might be easier to build and realize. At the same time, our new protocol is much more computationally efficient than the BJS protocol.

The study presents a new efficient way to construct the one‐round key exchange (ORKE) without random oracles based on standard hard complexity assumptions. The authors propose a (PKI‐based) ORKE protocol which is more computational efficient than existing pairing‐based ORKE protocols without random oracles in the post‐specified peer setting. The core idea of this construction is to integrate the consistency check of the ephemeral public key and the verification of the signature into the session key generation. This enables us to roughly save two pairing operations. The authors just call this kind of scheme that is deeply composed by signature and one‐round key exchange as SignORKE. The authors’ protocol is shown to be secure in a variant of the Canetti–Krawczyk security model which covers the majority of state‐of‐the‐art active attacks.

Vehicular Networks represent one of the main topics in the communication systems area. Increased deployment of vehicular networks together with new possibilities for Internet Protocol Television (IPTV) services has spurred the interest in providing mobile IPTV via vehicular networks. Provisioning of IPTV services over vehicular networks can be expected to play a significant role in the next generation of TV systems. Users will choose IPTV based on the Quality of Experience (QoE). Among QoE measures, TV channel availability is one of the most significant. In this paper, we investigate the channel availability in IPTV services with different traffic intensity and a varying number of TV channels offered to find out the acceptable availability of TV channels or the Call Blocking Probability (CBP) in our study. Comprehensive experiments are carried out by means of our own simulation tool, i.e., a Monte Carlo simulator to conduct the evaluations for numerous scenarios taking into account realistic IPTV user behavior.