Images are typically sampled on a uniform grid,despite their non-uniform information distribution—some regions are rich in content while others are not. The mismatch leads to inefficient computation allocation in deep learning models. To address this, recent studies have proposed predictive downsampling methodsthat adaptively downsample images based on predicted per-pixel importance, allocating more pixels to informative areas. However,these methods require high-resolution processing to accurately estimate importance, which undermines their efficiency:the prediction itself must process the full-resolution image,consuming most of the computational budget. This high-resolution importance prediction is necessary because each input may differ significantly in structure and content. In this paper, we take a different approach and introduce a learn-to-downsample paradigmtailored for aligned vision recognition tasks, such as face recognition and palmprint recognition, where input alignment ensures consistent spatial structure across images. This alignment ensures structural consistency across images, allowing a shared, input-agnostic downsampling template applicable to all inputs. Furthermore, instead of relying on implicit importance maps, we introduce a flow-based representation that explicitly models the spatial warping from the original image to the downsampled version. The flow representation is not only more efficient but also more controllable: we regularize the flow using its Jacobian determinant to precisely control the sampling density and coverage,enabling interpretable and tunable sampling patterns. Extensive experiments on two aligned recognition tasks, face and palmprint recognition, demonstrate that our method substantially reduces computational cost with minimal accuracy degradation, achieving a significantly better performance-efficiency trade-off than existing predictive downsampling methods.

Temporal knowledge graph (TKG) reasoning involves inferring future unknown facts based on historical data. Current approaches to temporal reasoning can be broadly categorized into two main paradigms: embedding-based methods and symbolic methods. While embedding-based methods excel at capturing time by representing temporal facts as vector, symbolic methods exploit temporal dependencies using techniques such as random walks for inference purposes. However, existing methods often fail to fully exploit both the inherent time and intricate temporal relationship patterns simultaneously. To address this limitation, we propose Temporal neural probabilistic logic learning (TNPLL), an innovative framework that seamlessly integrates symbolic logic with neural embeddings for robust temporal reasoning. Our approach incorporates two key components, a set of temporal logic rules equipped with explicit temporal relationships and a scoring module implemented through a novel temporal memory network architecture. The proposed method effectively combines time and temporal relationship patterns to predict future facts. We conducted experiments on several benchmark datasets, demonstrating that TNPLL achieves improved performance while fully leveraging time information. Specifically, our framework excels in scenarios where prior knowledge is available, but data samples are sparse. The experimental outcomes show that TNPLL outperforms state-of-the-art models in such cases.

The vision of 6G networks aims to enable edge inference by leveraging ubiquitously deployed artificial intelligence (AI) models, facilitating intelligent environmental perception for a wide range of applications. A critical operation in edge inference is for an edge node (EN) to aggregate multi-view sensory features extracted by distributed agents, thereby boosting perception accuracy. Over-the-air computing (AirComp) emerges as a promising technique for rapid feature aggregation by exploiting the waveform superposition property of analog-modulated signals, which is, however, incompatible with existing digital communication systems. Meanwhile, hybrid reconfigurable intelligent surface (RIS), a novel RIS architecture capable of simultaneous signal amplification and reflection, exhibits potential for enhancing AirComp. Therefore, this paper proposes a Hybrid RIS-aided Digital AirComp (HRD-AirComp) scheme, which employs vector quantization to map high-dimensional features into discrete codewords that are digitally modulated into symbols for wireless transmission. By judiciously adjusting the AirComp transceivers and hybrid RIS reflection to control signal superposition across agents, the EN can estimate the aggregated features from the received signals. To endow HRD-AirComp with a task-oriented design principle, we derive a surrogate function for inference accuracy that characterizes the impact of feature quantization and over-the-air aggregation. Based on this surrogate, we formulate an optimization problem targeting inference accuracy maximization, and develop an efficient algorithm to jointly optimize the quantization bit allocation, agent transmission coefficients, EN receiving beamforming, and hybrid RIS reflection beamforming. Experimental results demonstrate that the proposed HRD-AirComp outperforms state-of-the-art digital AirComp baselines in terms of both inference accuracy and uncertainty, achieving performance close to the idealized case with perfect feature aggregation.

Unmanned aerial vehicle (UAV) communications are susceptible to malicious compound interference signals due to the complexity and variability of the electromagnetic environment and the openness of the air-to-ground wireless channels, leading to degradation of communication quality. Therefore, effective detection and accurate recognition of compound interference are the key to ensuring secure UAV communication in complex environments. However, existing deep learning-based interference recognition algorithms suffer from fewer recognizable compound interference types, a large number of model parameters, and lower interference recognition accuracy under low interference-to-noise power ratio (INR) conditions. This paper proposes a malicious compound interference recognition method for UAV communication based on multi-modal multi-label learning and designs a lightweight multi-modal interference recognition network. By introducing a multi-label learning mechanism and making full use of the complementary information between different modalities of the signal, the method can achieve more flexible, accurate and stable recognition of compound interference signals under low INR. We construct both simulation and real measured datasets containing 31 classes of compound interference signals, and conduct simulation experiments with sufficient samples, insufficient samples, and different training strategies. The results demonstrate that the proposed method enhances the recognition accuracy of UAV communication compound interference under low INR and across different training datasets, all while maintaining a small number of model parameters.

The increasing demand for autonomous and open peer-to-peer (P2P) data sharing has driven the widespread adoption of decentralized file systems, such as the InterPlanetary File System (IPFS). However, decentralized data sharing inherently requires distributed access control mechanisms due to the absence of centralized authorities. Although blockchain-based access control has become a primary solution, the public nature of blockchain can unintentionally reveal user attributes, posing significant privacy risks. To address the leakage of attribute sets in blockchain, we propose zk-Guard, a decentralized access control framework integrating blockchain and zero-knowledge Succinct Non-interactive Arguments of Knowledge (zk-SNARKs) tailored for IPFS. To further improve the efficiency of zero-knowledge policy checking and reduce the delay of policy updating, we employ a universal constraint circuit and encode policies into sparse configuration matrices, achieving fine-grained, rapid policy updates without regenerating proving keys while guaranteeing constant-time verification regardless of policy complexity. Additionally, to prevent repeated permission checks for large f iles and improve system responsiveness, zk-Guard integrates Merkle Tree Proof (MTP) mechanisms to securely link sub-data blocks to their root block. Comprehensive theoretical complexity analysis and extensive experiments demonstrate that zk-Guard achieves substantial performance improvements over existing schemes, with constant-time proof verification under 2.5 ms enabling efficient data retrieval, and policy deployment and updates completed within 0.2 seconds even for 1,000 attributes. The source code is available at https://github.com/ningboliucug/zk-Guard.

Motion cues play a vital role in multi-frame infrared small target detection (MISTD). However, most targets in existing datasets exhibit regular and slow motion, which cannot reflect the complex and diverse motion patterns in real-world scenarios. This biased data distribution makes recent data-driven methods highly rely on simplified motion assumptions that tend to fail in irregular or fast motion, resulting in noisy feature representations cluttered with target-irrelevant factors. Hence, we stress that methods for MISTD should also work when targets are in complex motion. To enable this research, we propose a large-scale dataset called MIST for airborne infrared detection scenarios. The dataset is built on a synthetic data engine that models variations in pose, size, and intensity of moving targets while seamlessly blending them into real backgrounds for physical, geometric, and visual realism. Targets in MIST exhibit low signal-to-clutter ratios and complex motion, making it a promising yet challenging benchmark for developing algorithms focused on motion analysis. To tackle the challenges of MIST, we develop MISTNet, a robust baseline based on the Information Bottleneck theory. To handle irregular and fast motion, we propose a shifted neighborhood compensation block to efficiently model multi-scale correspondences for implicit motion compensation. To distill compact representations free from irrelevant cues, we design a progressive distillation decoder to hierarchically filter out redundancy while preserving target-relevant information. We benchmark 31 state-of-the-art methods and find that their performance on MIST drops significantly compared with that on the widely used NUDT-MIRSDT dataset. Our MISTNet outperforms all other methods by a large margin, with an over 6% gain in the IoU metric, demonstrating its superiority. The dataset, code, and model weights are available at https://github.com/GR-ray/MIST.

Accurate sleep staging is vital for evaluating sleep quality and diagnosing sleep disorders. Yet most automated sleep staging methods rely on large datasets labeled by experts. However, clinical annotation is both time-consuming and subjective, making it difficult to obtain sufficient high-quality data for automated sleep staging research. To address this bottleneck, we propose a few-shot, dual-branch contrastive pre-training framework for single-channel electroencephalogram (EEG)-based sleep staging. The framework first conducts fully self-supervised pre-training on unlabeled data, then performs fine-tuning that requires only a small set of labeled samples. We developed and evaluated our solution with the public Sleep-EDF-v2 EEG dataset, achieving state-of-the-art results despite using limited labeled data. Specifically, with only 1% labeled data, our method delivers an accuracy of 76.10% and Macro F1-score of 61.34%, comparable to supervised models trained on 100% labeled data. We further validated our approach on the ISRUC-1 and ISRUC-3 datasets, where similar robust results were consistently observed. The ability to effectively develop sleep classification models using minimal labeled data demonstrates the potential value of our framework across diverse clinical settings.

The sixth-generation (6G) mobile communication systems are expected to provide seamless connectivity for unmanned aerial vehicles (UAVs) to support them in fulfilling various tasks. However, the line-of-sight (LoS)-dominated channels of cellular-connected UAVs expose them to severe co-channel interference from nearby base stations (BSs), which significantly degrades communication reliability. To address this challenge, this paper investigates a movable antenna (MA)-enhanced cellular-connected UAV communication system, where the additional spatial degrees of freedom (DoFs) offered by MAs are exploited for the interference-aware UAV trajectory planning. Specifically, we formulate an optimization problem to minimize the UAV mission completion time by jointly optimizing the UAV beamforming matrix, antenna position vector (APV), UAV trajectory, and UAV–BS association, subject to constraints on signal-to-interference-plus-noise ratio (SINR) requirements, UAV mobility, and MA mobility. To overcome the inherent challenges of the continuous-time formulation, we discretize both the flight region and trajectory of the UAV, thereby reformulating the problem into a tractable discrete optimization problem. A selective uniform cost search (SUCS) algorithm is then developed for UAV trajectory planning, where the feasibility of candidate grid points is evaluated by jointly optimizing beamforming, APV, and UAV–BS association to maximize the expected SINR. Simulation results show that, compared with benchmark schemes, the proposed MA-enhanced design significantly improves the expected SINR of cellular-connected UAVs along the optimized trajectory, thereby reducing UAV mission completion time while ensuring reliable communication links.

The bistatic Integrated Sensing and Communication (ISAC) is poised to become a key application for next generation communication networks (e.g., B5G/6G), providing simultaneous sensing and communication services with minimal changes to existing network infrastructure and hardware. However, a significant challenge in bistatic cooperative sensing is clock asynchronism, arising from the use of different clocks at far separated transmitters and receivers. This asynchrony leads to Timing Offsets (TOs) and Carrier Frequency Offsets (CFOs), potentially causing sensing ambiguity. Traditional synchronization methods typically rely on static reference links or GNSS-based timing sources, both of which are often unreliable or unavailable in UAVbased bistatic ISAC scenarios. To overcome these limitations, we propose a Time-Varying Offset Estimation (TVOE) framework tailored for clock-asynchronous bistatic ISAC systems, which leverages the geometrically predictable characteristics of the Line-of-Sight (LoS) path to enable robust, infrastructure-free synchronization. The framework treats the LoS delay and the Doppler shift as dynamic observations and models their evolution as a hidden stochastic process. A state-space formulation is developed to jointly estimate TO and CFO via an Extended Kalman Filter (EKF), enabling real-time tracking of clock offsets across successive frames. Furthermore, the estimated offsets are subsequently applied to correct the timing misalignment of all Non-Line-of-Sight (NLoS) components, thereby enhancing the high-resolution target sensing performance. Extensive simulation results demonstrate that the proposed TVOE method improves the estimation accuracy by 60%.

Image deblurring is the process of recovering a high-quality image from a degraded image, which can be lost sharpness by blur filters, noise, compression, or other degradation factors. Image deblurring is a challenging task, as it requires dealing with complex and ill-posed inverse problems, and balancing the trade-off between performance and efficiency. This paper introduces a multi-stage knowledge distillation approach using teacher-student network interactions. Our method employs a three-stage teacher network combining Window-based Transformer and Unet models for the first and second stage, followed by a Channel-level Transformer as the third stage to enhance detail extraction. The student network, streamlined for speed, mirrors the teacher's structure with lower complexity, incorporating a standard Unet model and channel attention mechanism. A new Supervised Attention Module is introduced for effective knowledge transfer and feature enhancement. We evaluate our method on various datasets and show significant enhancements in image deblurring quality, while balancing performance and model efficiency. Our findings suggest that our method has great potential for real-world applications and opens up new possibilities for future research in image deblurring techniques.

Getting dense, uniform, time-series point cloud data is critical for effective rendering. However, due to the limited computational power of edge devices, existing methods cannot achieve real-time results, which affects the visual quality of the consumer experience. To effectively address this issue, this paper presents a self-supervised adversarial upsampling method for point cloud video streams called SPR-GAN. In the generator, we design the Temporal Iterative Graph module to learn local features for each frame and captures long-range spatial information using three iterations of graph convolution operations. Then the Contextual Temporal Fusion module is developed to merge information between different frames, synthesizing temporal information and enriching the dynamic feature representation of the point cloud. Meanwhile, in the discriminator, we introduce the Efficient Shape module. Through dynamic graph convolution operations and stacked learning, it significantly improves the resolution efficiency of global shape information in point clouds. The final experiments show that the proposed method exhibits high practicality and superiority. The model achieves a good result on both the D-FAUST and DeformingThing4D-Animals datasets.

Compared with vision-based methods, millimeter-wave (mmWave) radar offers a non-invasive and privacy-preserving approach to human activity recognition (HAR), making it a robust solution for tasks requiring reliable performance under challenging conditions. Its ability to operate effectively in low-light or partially occluded environments renders it particularly suitable for applications such as smart healthcare and human-computer interaction. This paper introduces STFNet, an end-to-end spatiotemporal feature fusion network for human activity recognition using mmWave radar point clouds. The proposed method first employs a multi-scale spatial feature encoding module, which integrates multiple customized adaptive spatial feature extraction blocks at different scales to capture multi-level spatial information. A subsequent Transformer-based network then fuses these multi-scale features to enrich spatial representations. Next, a Bidirectional Gated Recurrent Unit (BiGRU) module is introduced to model the temporal dynamics present in the point cloud data. Finally, a Squeeze-and-Excitation Residual Network (SE-ResNet) is utilized to adaptively calibrate channel importance in the concatenated spatiotemporal features, significantly boosting recognition accuracy for complex activities. Experimental evaluation on the MMActivity dataset demonstrates that STFNet achieves a classification accuracy of 98.65%, outperforming state-of-the-art methods and confirming its effectiveness and superiority.

The simultaneous transmission of numerous users presents substantial challenges due to the inherent trade-off between channel estimation and information transmission in multi-user multiple-input multiple-output (MIMO) system. In this paper, we explore the use of the superimposed pilot (SP) scheme to tackle the large transmitting users, where the number of users may exceed the coherent time. SP scheme incorporates both transmitted data and noise in the channel estimation process, which is significant different from the counterpart of RP scheme. We provide an in-depth analysis of the interaction between interference caused by channel estimation errors and noise. We then derive the explicit expression for the scaling law of the mutual information lower bound (MILB) in relation to the number of users and the levels of transmitted power. Besides, the optimal power allocation between pilots and data transmission is also derived analytically. The analytical results demonstrate that the SP scheme significantly improves performance compared to traditional RP scheme in our consider case. Numerical results are also presented to validate our theoretical derivations.

The brain-inspired Simultaneous Localization And Mapping technology (SLAM) in Internet of Underwater Things enables real-time location of multi-Autonomous Underwater Vehicle (AUV) with low computational overhead. However, in actual scenarios, the movement of AUV swarm interferes with visual images, which affects the performance of SLAM seriously. To deal with the above problem, a Brain-inspired Navigation Method based on Composite Spatial Navigation Cell model is proposed to achieve the autonomous navigation of AUVs, which is inspired from the speed cell and boundary cell of biological brain. Firstly, this method establishes a brain-inspired navigation scene of multi-AUV, where the stereo camera of AUVs is used to collect environmental information. Next, this method establishes speed cells based on Spiking Neural Network (SNN) to obtain semantic information, the location, score, and descriptor of feature points, which helps to estimate the motion information of AUVs such as displacement changes and heading angle changes) accurately. In addition, based on the distance information and angle information from stereo camera of AUVs to static obstacles, the proposed method calculates the activity value of boundary cells to assist in the loop-closure detection of local view cells. Finally, this method uses pose cells to represent the motion posture of AUVs, and applies experience map to record the movement trajectory of AUVs. To measure the performance of proposed method, this paper establishes an underwater SLAM dataset and a land SLAM dataset respectively, which are affected by dynamic entities. Extensive experimental results show that this method has good adaptability in different SLAM datasets, and is better than other methods in terms of trajectory error.

Automatic modulation recognition plays a crucial role in the domain of electromagnetic situational awareness. Early recognition methods predominantly relied on expert experience and prior knowledge, demanding a high level of professional background and experience from practitioners, and usually underperformed in complex signal environments. In recent years, the continual development of deep learning (DL) technologies has introduced solution ideas to address the challenge of modulation recognition in complex electromagnetic environments. However, DL methods heavily depend on large volumes of high-quality labeled data and face challenges in real electromagnetic environments with limited samples. To fully leverage the respective strengths of expert knowledge in the radio domain and data-driven approaches, this article proposes a hybrid-driven neural network (HDNet) framework for radio signal recognition. HDNet integrates deep features extracted through data-driven methods with manual features extracted based on expert knowledge, aiming to enhance recognition performance in few-shot scenarios. Experimental results on both simulated and real measured datasets demonstrate that HDNet achieves high-recognition accuracy and robustness.

Terahertz (THz) and Non-Orthogonal Multiple Access (NOMA) technologies have illustrated a great potential in the use of large-scale Internet of Things (IoT) applications. However, in most applications, ground sensors (GSs) have poor energy capacity, while THz communication suffers from significant path loss, which leads to a low data transfer rate in air-ground communication system. Fortunately, with the development of unmanned aerial vehicle (UAV) assisted THz-NOMA communication technology, it becomes one of the promising solutions to deal with above challenges. To conserve GSs’ limited energy while enhancing the throughput of THz-NOMA communication system, we utilize a UAV to collect and transmit data from GSs to an aerostat. Then, energy efficiency (EE) of GSs network is maximized by leveraging GSs’ communication resource allocation strategy and UAV trajectory design, while taking into account each GS’s minimum throughput and each GS’s transmission power constraints. The proposed design is a mixed integer non-convex problem, which is generally intractable. To promptly solve the original problem, we divide it into three subproblems: GSs’ communication resource allocation, UAV’s altitude optimization, and horizontal trajectory design. An iterative algorithm is proposed to deal with these subproblems, based on Dinkelbach method, dual decomposition, and convex relaxation techniques. Simulation results show that the proposed iterative algorithm can achieve a higher EE with a shorter convergence time compared with baseline algorithms.

With the rapid development of 5G and big data technologies, the Internet of Things is increasingly evolving toward applications characterized by massive connectivity and high data throughput. Mobile edge computing (MEC) leverages resource integration to form clusters with collaborative capabilities, which is crucial for addressing the issues of functional disparities and resource limitations in network nodes. This article considers the MEC scenario of network clustering to provide collaborative services through clusters. To achieve effective and balanced network clustering, we formulate a clustering problem that maximizes modularity under a scale constraint. By analyzing the service capacity and content of nodes and the connections between them, we design an attraction function to compute modularity and set upper and lower bounds for cluster scales to ensure controllable cluster scales. We propose a cluster scale-constrained modularity-based Louvain (CSM-Louvain) algorithm, which incorporates splitting and merging mechanisms into the traditional Louvain algorithm to meet the scale constraints. Simulation results demonstrate that the proposed method effectively balances network resources and load distribution, thereby enhancing the quality of MEC collaborative services.

The global aging process accelerates, making geriatric disease prevention and management urgent. Combining IoMT with wearable devices for health monitoring is effective, but IoMT task offloading has challenges like weak terminal processing, poor battery life, low - latency needs, and privacy protection difficulties. To address these challenges, this paper proposes a Cloud-edge Collaborative Air-ground Integrated IoMT Network (C2AI2N) that combines ground Base Stations, air-based Autonomous Aerial Vehicles (AAVs) and cloud servers to adapt to future IoMT scenarios and enhance the system capabilities. Edge servers are able to use Wireless Power Transfer (WPT) to provide energy to wearable devices. We then consider the Ultra-Reliable Low Latency Communication (URLLC) task constraints as well as power limitations, and formulate the problem of minimizing the overall system energy consumption. Next, we employ the Lyapunov optimization method to break down the problem into three sub-problems: 1) device-side task splitting; 2) task offloading as well as charging strategy; and 3) server-side resource allocation, which are addressed via Lagrange multiplier method, Clustered Asynchronous Federated Learning (CAFL) based rainbow Deep Q-Network (DQN) approach, and the coati optimization algorithm, respectively. Extensive simulations demonstrate that the proposed method outperforms other baselines. Specifically, the energy consumption of the proposed method can be reduced by over 30.37% compared to the baselines. In addition, the backlog of the URLLC virtual queue can be decreased by more than 59.27%, achieving superior URLLC satisfaction rates.

Automatic modulation classification (AMC) plays an increasingly crucial role in intelligent spectrum management and dynamic spectrum access, which can effectively support the reallocation of low-utilization spectrum resources in wireless communication systems. While deep learning approaches have been widely employed in AMC, most deep learning-based AMC methods focus on signal classification as a singular task. Therefore, this paper proposes a multi-task learning-based method for radio signal recognition aimed at enhancing AMC performance. This method utilizes the designed multi-task collaborative learning network (MCLNet) model to achieve complementary gains across different tasks. By sharing parameters, it enhances the learning capability of crucial signal features, thereby acquiring more discriminative signal features and improving classification accuracy. Experimental results demonstrate that the proposed method outperforms other benchmark models on two benchmark datasets and exhibits greater performance gains in few-shot scenarios.

The lack of privacy-preserving capabilities hinders the further development of blockchains and smart contracts. While numerous privacy solutions have been proposed, limitations persist. First, most existing solutions focus on specific privacy protections such as anonymous payments, private data, or multi-party computation tasks. However, these solutions lack a general privacy ability, allowing users to deploy applications with diverse privacy requirements. Second, existing solutions have limited customizability, which means users cannot easily customize and adapt the privacy policies according to their specific demands or preferences. In this article, we present EtherCloak, which adopts trusted execution environments (TEEs) to achieve a general and customizable privacy policy on account model blockchains, enabling users to conceal any on-chain information. To address the security issues caused by the unreliability of the host the TEE runs on, we design the enclave state check and crash recovery mechanisms and employ them in the block generation process. In addition, we propose an access control mechanism for privacy policy management and data query. We prove that EtherCloak offers general and customizable privacy protection with a minimal increase in transaction size (less than triple) and communication overhead (approximately 10%) compared to Ethereum.

Crowdsourcing has emerged as an effective paradigm for large-scale data collection and statistical analysis. However, the paramount concern about worker privacy has driven the development of privacy-preserving statistical analysis methods. We propose PSAC, a novel framework that leverages histograms to facilitate privacy-preserving statistical analysis in crowdsourcing. PSAC integrates secure statistical analysis protocols based on homomorphic encryption and secure two-party computation, addressing the limitations of a single cryptographic technique. It introduces innovative algorithms using histograms for statistical operations, including functions such as quantile estimation, outlier elimination, contingency table construction for$\chi ^{2}$test, and the Mann-Whitney$U$test. These algorithms exhibit minimal overhead growth with respect to data volume, demonstrating exceptional scalability for large numbers of data. Moreover, through a key-separation design, PSAC ensures that only the requester can decrypt the final results independently, even if the ciphertexts of data are exposed. Comprehensive evaluations validate the security, efficiency, and scalability of the PSAC framework.

Service Function Chaining (SFC) is widely deployed by telecom operators and cloud service providers, offering traffic QoS guarantees and other additional functions for various applications. The network state at the time of SFC deployment can differ significantly from the runtime conditions, leading to excessive resource allocation and consequent energy waste. The existing SFC reconfiguration methods face the challenge of meeting the latency requirements of delay-sensitive applications while achieving significant energy savings. This paper proposes FR-SFCO, a flow rate-aware SFC offloading framework on programmable data planes for delay-sensitive flows. Specifically, we designed a TCAM-friendly table matching method for FR-SFCO to reduce the flow entries needed for SFC offloading in programmable switches and support larger numbers of offloaded SFC. Then, we proposed a dual-threshold-based offloading trigger mechanism that, according to the real-time traffic arrival rate, can fast offload SFC flows before they default to servers. Building on this, we propose DQN-AOTA, an adaptive offloading thresholds adjustment algorithm based on Deep Q-Learning, which can wisely change the offloading thresholds by interacting with a dynamic network traffic environment to minimize the packet loss and long-term energy consumption. Finally, we build a testbed using BMv2 software switches and Docker containers for extensive evaluation. The experimental results demonstrate the effectiveness of our solution which not only meets the latency constraints for delay-sensitive SFC flows but also reduces energy expenditure by at least 14.6%.

Natural disasters pose a significant threat to human life. In these extreme conditions, terrestrial networks frequently become incapacitated, hindering the provision of essential communication and computing services required for emergency response efforts. In recent years, the rapid advancement of drone technology, coupled with the maturation of lightweight communication and computing equipment, has led to the emergence of unmanned aerial vehicle (UAV) networks as a crucial asset in disaster rescue missions. These networks provide significant advantages, including rapid response times, flexible deployment capabilities, and heightened resilience to complex terrains, showcasing considerable potential for further development. UAV networks exemplify resource-constrained systems where efficient scheduling of computing resources is vital. Especially in emergency rescue scenarios, this complexity is exacerbated by the diverse range of tasks, varying demands, and stringent real-time requirements. Effectively managing and allocating the computational resources of drones is essential for maximizing their operational efficiency in response to the intricate dynamics of disaster situations. To improve the computational service efficiency of the network, this paper proposes an emergency rescue UAVs network architecture. Additionally, we investigate a joint optimization approach for task planning and SFC scheduling. Current research on SFC scheduling primarily focuses on ground data center networks, with comparatively limited investigation into UAV networks. Fully considering the mobility of computing nodes, as well as the wireless transmission modes within the aerial environment, we establish a joint optimization model for task planning and SFC scheduling aiming at minimizing the total weighted end-to-end delay. Then we design the A3C based algorithm to learn the optimization strategy. Simulation results are presented to demonstrate the superiority of the proposed approach in the aspect of total weighted end to end delay and training time against other benchmark algorithms.

The recommendation algorithm may feed to end users information by analyzing the potential interests of the users in last interaction period, and finally the information obtained by the users may be constrained within a limited range. This problem is so-called information cocoon and has be attracted more and more attentions. In this paper, we study this problem in the perspective from both service providers and end users. An analysis model in terms of probabilistic transfer graph is proposed, and corresponding matrix is analyzed. We also propose a method to compute the power of matrix by approximating any transfer graph into a symmetric matrix, and the stability on the probability can be easily computed by matrix division and eigenvalues.

Continuous blood pressure (BP) monitoring is crucial for cardiovascular care but is often constrained by financial concerns. Continuous BP estimation based on ballistocardiogram (BCG) signals offers a non-invasive and cost-effective alternative. Traditional approaches to BP estimation from BCG signals have been hampered by single-channel noise interference, high costs of multi-signal acquisition, and insufficient feature capture. This paper introduces BP-STFNet, a novel deep learning framework that transcends these limitations by leveraging a unique fusion of time-frequency domain information. BP-STFNet enhances input signal quality and ensures robustness against physical movement artifacts. Our custom-designed parallel gated dilated convolution architecture, along with Squeeze-and-Excitation ResNet and Bidirectional Gated Recurrent Unit (Bi-GRU) module, extract detailed features and capture dynamic temporal patterns. The framework achieves a Mean Absolute Error (MAE) of 4.08 mmHg for systolic and 2.12 mmHg for diastolic pressure measurements, demonstrating highly competitive results for systolic pressure estimation and state-of-the-art performance for diastolic pressure prediction. The promising results achieved with BP-STFNet suggest its potential as a viable solution for both continuous and real-time BP monitoring, paving the way for improved cardiovascular health management.

Blockchain can be envisioned as an enabling framework in terms of cryptography and distributed computing to construct trusted data records among un-trusted users. It can be applied in two typical scenarios - historical process provenance and transaction tracing, by immutably sequentialized blocks. In the former there exists no tokens; but in the latter token handover among accounts should be cryptographically guaranteed. In some situations, above two applications may be both required. E.g., in the manufacture of a jewel, processing flow by different operators should be recorded for quality guaranteeing, and in its trading stage, the ownership handover should be recorded for value confirmation. To tackle above two requirements, two chains are usually thus demanded. In this paper, we propose a single chain solution to avoid cross-chain burden, by a hybrid block structure. The process data can be imported from the legacy (non-anonymous) information processing system, and transaction data stems from anonymous payment demand for ownership handover. We formally describe and prove token handover model with cryptographical authenticity. We also propose several security enhancements in the design for shortening the length of the chain, and protecting the data confidentiality in the chain. Our design is general and not limited to single application. The experimental results and extensive analysis justified that the performance and security of our proposed scheme is efficient.

BACKGROUND: An inherent difference exists between male and female bodies, the historical under-representation of females in clinical trials widened this gap in existing healthcare data. The fairness of clinical decision-support tools is at risk when developed based on biased data. This paper aims to quantitatively assess the gender bias in risk prediction models. We aim to generalize our findings by performing this investigation on multiple use cases at different hospitals. METHODS: First, we conduct a thorough analysis of the source data to find gender-based disparities. Secondly, we assess the model performance on different gender groups at different hospitals and on different use cases. Performance evaluation is quantified using the area under the receiver-operating characteristic curve (AUROC). Lastly, we investigate the clinical implications of these biases by analyzing the underdiagnosis and overdiagnosis rate, and the decision curve analysis (DCA). We also investigate the influence of model calibration on mitigating gender-related disparities in decision-making processes. RESULTS: Our data analysis reveals notable variations in incidence rates, AUROC, and over-diagnosis rates across different genders, hospitals and clinical use cases. However, it is also observed the underdiagnosis rate is consistently higher in the female population. In general, the female population exhibits lower incidence rates and the models perform worse when applied to this group. Furthermore, the decision curve analysis demonstrates there is no statistically significant difference between the model's clinical utility across gender groups within the interested range of thresholds. CONCLUSION: The presence of gender bias within risk prediction models varies across different clinical use cases and healthcare institutions. Although inherent difference is observed between male and female populations at the data source level, this variance does not affect the parity of clinical utility. In conclusion, the evaluations conducted in this study highlight the significance of continuous monitoring of gender-based disparities in various perspectives for clinical risk prediction models.

In this paper, we investigate the mmWave beam tracking for vehicle-to-infrastructure (V2I) communications to find the optimal beam via sub-6 GHz channel state information (CSI). We consider two scenarios: 1) sub-6 GHz and mmWave transceivers are co-located on the same base station (BS), and 2) sub-6 GHz and mmWave BSs are separated in different places constituting heterogeneous networks (HetNets) where one sub-6 GHz BS controls multiple mmWave BSs. Considering the mobility of the vehicle and time-varying channels, we propose a predictive beam tracking method based on long short-term memory (LSTM) to construct the maps from historical sequential sub-6 GHz CSI to the future optimal mmWave beam. A single LSTM model can handle the beam tracking in the co-located scenario, since there is a one-to-one correspondence between the sub-6 GHz and mmWave transceivers, and the propagation of sub-6 GHz and mmWave signals is similar. However, in the HetNet scenario, it is difficult to select the best one among the beams of multiple mmWave BSs only via the CSI of one sub-6 GHz BS. To address this challenge, we design an LSTM fusion model, which exploits not only the historical sequential sub-6 GHz CSI but also a number of mmWave wide beam measurements, to obtain the optimal mmWave BS and beam in the HetNet. In this case, the collected sub-6 GHz CSI and mmWave wide beam measurements are analyzed by the LSTM and fully connected network (FCN) modules, respectively, providing two beam prediction results. Then the results are fused by an attention-based FCN module to accomplish the final prediction. Simulation results verify the effectiveness and superiority of our LSTM-based beam tracking models compared with other state-of-the-art deep learning beam tracking models that also leverage sub-6 GHz channels. Besides, the robustness and generalization of our proposed LSTM models are illustrated through simulations.

In response to critical security threats such as data tampering, identity impersonation, and channel eavesdropping in Vehicle-to-Grid (V2G) networks, numerous charging authentication schemes have been proposed. However, these schemes either lack sufficient anonymity, physical security, or electricity consumption data aggregation for electricity dispatch. In light of these considerations, we propose a comprehensive solution—a physically secure and privacy-preserving charging authentication framework with data aggregation, comprising two foundational schemes. The first scheme introduces a fully anonymous authentication system. In this approach, an Electric Vehicle (EV) seeking charging generates a random signature for its charging request. Subsequently, a Charging Station (CS) verifies the signature, granting charging services upon successful validation. Notably, this process guarantees the EV’s real identity remains undisclosed, even to the Control Center (CC). Moreover, this scheme also addresses potential physical attacks through the incorporation of a physical unclonable function. The second scheme involves a privacy-preserving data aggregation scheme, aggregating total electricity consumption of CSs in a given area while simultaneously preserving individual CSs’ electricity consumption data from potential leakage. Subsequently, the aggregated electricity consumption data is transmitted back to the CC, enabling efficient electricity coordination. A detailed security and privacy analysis demonstrates that our proposed framework meets intended security and privacy objectives. The final performance evaluation underscores the advantages of our proposed framework in comparison with related work.

Remote surgery is an emerging medical business derived from information networking technology and plays an increasingly essential role in the medical system. In remote surgery, it is imperative to facilitate cross-regional information transmission and processing by leveraging medical information networks to establish a collaborative service model served by multiple data centers in different regions, enabling collaboration and support for surgery operations. Additionally, the implementation of service function chain scheduling technology is crucial for the efficient allocation of computing resources of data centers. In this paper, we design a novel multi-cloud collaborative medical information network framework. Based on this framework, the service function chain (SFC) scheduling problem is investigated to minimize the total weighted end-to-end delay. To solve the scheduling problem, the original problem is reformulated as a Multiple Markov Decision Process (MMDP). Then, a multiple-state-action deep reinforcement learning (MSA-DRL) algorithm is developed to learn the best scheduling policy. Simulation results are presented to demonstrate the superiority of the proposed approach in the aspect of total weighted end to end delay against other benchmark algorithms.

To improve the security of crowdsourcing, existing studies introduce blockchain to ensure reliability and utilize cryptography (e.g., encryption and zero-knowledge proof) to protect data privacy. Nevertheless, the crowdsourcing process may involve sensitive identity information, and identity protection remains unresolved during stages such as data submission and correct payment. Especially, when workers invoke a smart contract to submit data, it inevitably exposes their identities. Identity disclosure significantly impacts the credibility of crowdsourcing platforms. Existing solutions suggest solving the problem through anonymous token contracts such as Zether. However, tokens can easily result in fund freezing or extra information leakage. Moreover, invoking the contract to submit data will still disclose workers' blockchain accounts. To tackle the identity protection issue, in this paper, we propose AcCrowd which achieves worker anonymity and payment correctness in crowdsourcing systems atop blockchain. We first design a verifiable proxy submission mechanism for data submission, enabling workers to invoke contracts without disclosing their accounts. Then, we introduce and improve the BlockMaze architecture to replace previous anonymous token-based methods, enhancing privacy and flexibility. Besides, we designed a revealed payment mechanism that utilizes an adaptor signature to bind data reveal and reward payment together, simultaneously protecting the requester and worker. Our security and performance evaluations demonstrate the security and practicability of AcCrowd.

Signal-channel electroencephalogram (EEG) based automatic sleep stage classification with machine learning is widely used in the study of sleep quality and analysis of sleep disorders. While due to the inevitable class imbalance problem, relatively poor accuracy in the detection of the first stage of the Non-Rapid-Eye-Movement (N1) is found. In this work, we propose to use a generative adversarial network (GAN) based on transformer encoder to reduce the class imbalance problem. In order to ensure the quality of the synthetic signal, the percentage form of the Kullback-Leibler divergence (KLP) index is designed to measure the similarity of synthetic signals generated by GAN and real ones. Meanwhile, we design a Residual Shrinkage Sequence Network for Sleep Staging (RsSleepNet) as the baseline to compare other resolutions of class imbalance with ours. The performance of the new method with the combination of GAN and RsSleepNet is effectively verified on two public datasets from PhysioNet, in which the accuracy of the N1 stage can be improved by more than 10% as compared to the current state-of-the-art approaches, largely alleviating the class imbalance problem in automatic sleep stage classification.

Some multimedia data from real life can be collected as multivariate time series data, such as community-contributed social data or sensor data. Many methods have been proposed for multivariate time series forecasting. In light of its importance in wide applications including traffic or electric power forecasting, appearance of the Transformer model has rapidly revolutionized various architectural design efforts. In Transformer, self-attention is used to achieve state-of-the-art prediction, and further studied for time series modeling in the frequency recently. These related works prove that self-attention mechanisms can reach a satisfied performance whether in time or frequency domain, but we used recurrent neural network (RNN) to verify that these are not critical and necessary. The correlation structure of RNN has time series specific inductive bias, but there are still some shortcomings in long multivariate time series forecasting. To break the forecasting bottleneck of traditional RNN architectures, we introduced RNNGAN, a novel and competitive RNN-based architecture combining the generation capability of Generative Adversarial Network (GAN) with the forecasting power of RNN. Differentiated from the Transformer, RNNGAN uses long short-term memory (LSTM) instead of the self-attention layers to model long-range dependencies. The experiment shows that, compared with the state-of-the-art models, RNNGAN can obtain competitive scores in many benchmark tests when training on multivariate time series datasets in many different fields.

UAV-based air-ground integrated computing networks (AGIN) have gained significant traction in remote areas for the Power Internet of Things (PIoT). This paper considers an AGIN-PIoT, where computing tasks generated by ground PIoT devices are offloaded to aerial UAVs that perform edge computing. Jointly optimizing task offloading and UAV trajectory poses challenges such as many decision variables, information uncertainty, and long-term queue delay constraints. Due to the limited battery capacity of PIoT devices and UAVs, our objective is to minimize system energy consumption under long-term queue delay constraints by jointly optimizing task offloading, trajectory planning, and computing resource assignment. In light of Lyapunov optimization, we decompose the original challenging optimization problem into two sub-problems: (1) task offloading and UAV trajectory planning and (2) aerial edge resource allocation. Accordingly, we develop a multi-agent deep reinforcement learning-based algorithm called AGIN-MADDPG for the former to achieve the maximum accumulative reward and propose a greedy solution for the latter. Extensive experiments and numerical results demonstrate that our approach can avoid the problem of gradient vanishing and outperforms other benchmark methods in terms of power consumption, task backlog, queue delay, and system throughput.

Remote first-aid treatment on ambulances is a promising application of 5G. However, there still exist gaps between the capabilities of current 5G networks and the stringent requirements of remote emergency on ambulances. Dual connectivity (DC) is an efficient technology to fill these gaps by integrating 5G millimeter wave (mmWave) with Sub-6GHz networks. In this paper, we investigate a dual-connectivity handover scheme to enhance the transmission rate of the wireless links for a 5G-enabled ambulance. Due to the long delay caused by signal transmission and processing, the conventional handover schemes based on reference signal received power (RSRP) measured by users are not sufficiently sensitive to the rapidly changing propagation environments surrounding the 5G-enabled ambulance. Instead, considering the randomness of environments and the delay caused by the handover process, we employ a deep Q network (DQN)-based algorithm to find a far-sighted policy for solving the handover problem. However, due to the drawbacks of single-step bootstrapping, value overestimation, and low-efficiency exploration, the vanilla DQN is performance-limited. To this end, we adopt effective techniques including multi-step learning, double DQN, and NoisyNet to improve learning performances, and propose a noisy double DQN (NDDQN)-based dual-connectivity handover scheme. Simulation results verify the effectiveness and superiority of our NDDQN-based handover scheme compared with the vanilla DQN and upper confidence bound (UCB)-based handover schemes, and then show that our handover scheme can adapt to various handover models.

Lung cancer is one of the leading causes of death worldwide. Early diagnosis through cancer screening can significantly improve lung cancer patients’ survival. Recently, deep learning based diagnostic systems for nodule detection have shown great potential in assisting radiologists to screen cancer more efficiently. However, studies have found that deep learning models lack robustness against imperceptible crafted adversarial attacks and few studied improving the robustness of pulmonary nodule detection. Therefore, making pulmonary nodule detection models robust remains challenges. Moreover, traditional adversarial training methods either hurt the natural generalization or need expensive computational cost. To address these challenges, here we propose a novel adversarial training method called, Adaptive Iteration Adversarial Training (AIAT). AIAT generates adversarial samples by adding adversarial noise with an adaptive iteration strategy, so that it can stably and fast train models with improving robustness. Extensive experiments on the LUNA 16 dataset show that AIAT improves robustness for pulmonary nodule detection without compromising the natural generalization, and largely reduces training time.

Truth discovery is an effective tool to infer true information from multi-source data and has been widely applied in mobile crowdsensing systems. In some specific scenarios, the sensory data are collected in a streaming fashion with time-varying information, and the server should update the truth in time. Under such circumstances, local differential privacy-based mechanism can satisfy the requirement of real-time processing properly while keeping the privacy of sensory data. However, directly applying local differential privacy to handle streaming data will disclose the long-term potential privacy and decrease the accuracy. To address these problems, we propose an incentive-based privacy-preserving truth discovery framework over streaming data. Firstly, we adopt the sequential composition theorem of w-event privacy to protect workers' long-term privacy. Second, we design an incentive mechanism to improve the submitted data utility and thus avoid the decrease in accuracy. In this way, our scheme ensures that workers submit more accurate data while their global privacy is still guaranteed. Finally, we prove our scheme satisfies w-event (∊, δ) differential privacy and theoretically analyze the result utility. Extensive experiments also demonstrate the effectiveness of our incentive mechanism.

Recently, there have been many discussions in mobile crowd-sensing about privacy-preserving truth discovery because of its ability to extract truthful information from noisy or biased sensory data without privacy breaches. However, in practical applications, users (referred to as workers) may report outliers due to device malfunction, malicious workers, etc. These outliers will dramatically impact the accuracy of the truth discovery result. Detecting outliers based on existing privacy preservation schemes will carry an intolerable overhead, dramatically reducing the system's availability. In this paper, we propose our privacy-preserving truth discovery scheme that can detect outliers. Specifically, we adopt an anonymous mechanism to achieve privacy preservation. Since the existing anonymous mechanisms require huge overhead and do not work correctly when some workers exit, they are difficult to be applied in mobile crowdsensing systems. We design a lightweight and robust anonymous mechanism based on the edge computing paradigm. In addition, we eliminate the impact of outliers through outlier detection to achieve robustness of truth discovery results. Finally, we demonstrate the security of our scheme through security analysis and the efficiency of our scheme in terms of computation and communication overhead through extensive experiments.

Large-scale self-supervised pre-training Transformer architecture have significantly boosted the performance for various tasks in natural language processing (NLP) and computer vision (CV). However, there is a lack of researches on processing multivariate time-series by pre-trained Transformer, and especially, current study on masking time-series for self-supervised learning is still a gap. Different from language and image processing, the information density of time-series increases the difficulty of research. The challenge goes further with the invalidity of the previous patch embedding and mask methods. In this paper, according to the data characteristics of multivariate time-series, a patch embedding method is proposed, and we present an self-supervised pre-training approach based on Masked Autoencoders (MAE), called MTSMAE, which can improve the performance significantly over supervised learning without pre-training. Evaluating our method on several common multivariate time-series datasets from different fields and with different characteristics, experiment results demonstrate that the performance of our method is significantly better than the best method currently available.