At present, most edge servers adopt popularity-based caching strategies, prioritizing the caching of content with the highest overall popularity on user-side edge servers. However, in social network scenarios, user interests and preferences are highly personalized and dynamically changing. This results in existing caching strategies often failing to adjust the cache placement of content in real time according to individual user preferences, leading to suboptimal edge cache hit rates, increased user request response latency, and a decline in quality of service (QoS) for the user experience. To address this issue, we propose a new caching strategy tailored for large-scale social content based on knowledge graph neural network (KGNNC). First, an entity-relation KG is constructed from users' triple data$(\boldsymbol{h}, \boldsymbol{r}, \boldsymbol{t})$on social platforms. Next, a graph convolutional neural network is employed to iteratively aggregate feature information from neighboring nodes and learn vector representations of the nodes. Finally, a reinforcement learning-based algorithm is utilized to determine the optimal caching location for content. Experimental results on multiple public datasets demonstrate, compared with several existing baseline algorithms (least recently used, least frequently used, neural network-based collaborative filtering, KG-DQN, and CAFR), the algorithm proposed in this article achieves a reduction in the average response latency of requests by 38.26%, 34.46%, 13.56%, 6.49%, 4.19% on MovieLens 1M dataset and 30.31%, 27.59%, 14.11%, 5.83%, 4.78% on last FM dataset, respectively. Meanwhile, experiment results demonstrate that the caching hit rate is increased by 24.2%, 25.1%, 14.4%, 6.5%, 3.53% on MovieLens 1M and 39.51%, 39.05%, 23.40%, 10.66%, 8.67% on last FM compared with the four existing baseline algorithms, respectively. These results verify the effectiveness of our algorithm in reducing response latency of user requests and improving caching hit rate of edge servers in social networks.

The evolution of 6G networks introduces heterogeneous services with stringent computing and latency demands. However, constrained edge resources, intricate task dependencies, and dynamic network fluctuations intensify resource contention, increasing the risk of node faults and service interruption. Current fault-tolerant methodologies lack the necessary adaptability to handle the coupled complexity of task interdependencies and volatile resource states, leading to sub-optimal decisions or excessive system overhead. To address these challenges, this paper innovatively proposes TransDiffuse—an intelligent preemptive service migration framework for 6G edge networks. First, the framework employs a Transformer-GAT hybrid model to capture long-range temporal load dynamics and spatial topological constraints, enabling accurate failure prediction. Second, to navigate the trade-off between migration overhead and service robustness, we devise a diffusion-based decision module. This module efficiently explores the discrete combinatorial solution space to synthesize near-optimal service orchestration. Furthermore, a comprehensive evaluation system is constructed to validate the effectiveness of TransDiffuse. Experiments demonstrate that TransDiffuse reduces energy consumption by 32.4%, decreases task completion time by 25.6%, and improves resource balance by 18.7%, while keeping service violations below 5%. This work achieves joint optimization of energy, delay, and resource efficiency, offering a robust solution for resilient service orchestration in 6G edge networks.

Multiview photometric stereo (MVPS) seeks to recover high-fidelity surface shapes and reflectances from images captured under varying views and illuminations. However, existing MVPS methods often require controlled darkroom settings for varying illuminations or overlook the recovery of reflectances and illuminations properties, limiting their applicability in natural illumination scenarios and downstream inverse rendering tasks. In this paper, we propose RotatedMVPS to solve shape and reflectance recovery under rotated natural light, achievable with a practical rotation stage. By ensuring light consistency across different camera and object poses, our method reduces the unknowns associated with complex environment light. Furthermore, we integrate data priors from off-the-shelf learning-based single-view photometric stereo methods into our MVPS framework, significantly enhancing the accuracy of shape and reflectance recovery. Experimental results on both synthetic and real-world datasets demonstrate the effectiveness of our approach.

Medical Visual Question Answering (Med-VQA) based on medical image understanding is a current research hotspot, which has attracted widespread attention and exploration in the industry. Poor performance caused by scarcity of training data is the most challenging problem of the Med-VQA task. Although algorithms based on self-supervised contrastive learning framework have alleviated this problem to a certain extent, unreliable and poor-quality training samples still severely impact the results of contrastive learning. Besides existing methods have poor inference ability. This paper proposes a method based on single-modal supervised contrastive learning and inference enhancement, which mainly includes two stages: pre-training and fine-tuning. In the pre-training stage, we take the type point, which is the type of medical images or text content, as a weakly supervised method to test the effectiveness of single modality, and guide multi-modal contrastive learning. In the fine-tuning stage, we propose a question-type classification method to learn different reasoning skills for close-ended and open-ended tasks respectively, and guide fusion via question-type attention. We evaluate the method on three public medical datasets and compare it with state-of-the-art models. Among them, the accuracy rate reached 78.5% on the VQA-RAD dataset, 84.2% on the Slake dataset, and 63.0% on the PathVQA dataset. Experimental results demonstrate the effectiveness of the method proposed in this paper.

The rapid development of autonomous vehicles and smart city has led to an exponential increase in data generation within Intelligent Transportation Systems (ITS). However, comprehensive extraction and utilization of these data are severely hindered by communication and energy constraints, security and privacy concerns, vehicle mobility limitations, and spatial distribution challenges. Using 6G and Digital Twin (DT) technologies offers a promising solution to these problems. In this paper, we propose a DT-based model training architecture for vehicular networks and introduce Federated Learning (FL) to preserve data privacy. While distributed model training and parameter transmission introduce challenges in delay and energy consumption, which conflict with real-time service requirements in ITS. In addition, the quality of the data and the processing capability of each vehicle varies widely, which will affect the efficiency of data sharing and model accuracy. Therefore, it is vital to select appropriate training nodes and optimize resource allocation under the constraints of task delay and energy consumption. We formulate an optimization model to improve the selection of FL participating nodes and energy management strategies, aiming to maximize accuracy while minimizing energy consumption. We then develop a DT-assisted deep reinforcement learning (DRL) method. Experiments show that our scheme achieves higher training accuracy and energy efficiency compared to the benchmark.

Using a deep learning method to detect buried pipes on ground penetrating radar (GPR) data is popular for its excellent performance potential. However, this needs a large amount of high-quality training data, which leads to time-consuming and labor-intensive data annotation work. Semi-supervised learning method provides a solution for the situation of small sample size. However, the original semi-supervised learning methods lack control over the quality of pseudo labels during model training. To address the problem, a new prior knowledge-guided semi-supervised deep learning method is proposed to improve the model performance under the small sample. In the method, a prior structure feature (PSF) is constructed to control the quality of pseudo label during semi-supervised learning. The structure response index (SRI) is designed to segment the structure subject out for eliminating disturbance from unstructured information. Then, the PSF is represented using histograms of oriented gradients (HOGs) and Fourier descriptors (FDs) to describe the structure’s edge orientation and shape. Based on the PSF, LightGBM is employed as a discriminator to screen out high-quality pseudo labels for model training. In the experiments, comparison experiments with some state-of-the-art supervised learning methods and semi-supervised learning methods were carried out to validate the method’s performance. The ablation studies on labeled dataset size, pseudo-label confidence threshold, and PSF were conducted to validate the effectiveness of the prior structural feature in controlling the pseudo label’s quality. The results show that the PSF is effective, and the proposed method outperforms the other methods.

BACKGROUND: Synaptogyrin-2 (SYNGR2), as a member of synaptogyrin gene family, is overexpressed in several types of cancer. However, the role of SYNGR2 in pan-cancer is largely unexplored. METHODS: From the TCGA and GEO databases, we obtained bulk transcriptomes, and clinical information. We examined the expression patterns, prognostic values, and diagnostic value of SYNGR2 in pan-cancer, and investigated the relationship of SYNGR2 expression with tumor mutation burden (TMB), microsatellite instability (MSI), immune infiltration, and immune checkpoint (ICP) genes. The gene set enrichment analysis (GSEA) software was used to perform pathway analysis. Besides, we built a nomogram of liver hepatocellular carcinoma patients (LIHC) and validated its prediction accuracy. RESULTS: SYNGR2 was highly expressed in most cancers. The high expression of SYNGR2 significantly reduced the overall survival (OS), disease-specific survival (DSS), disease-free interval (DFI), and progression-free interval (PFI) in multiple types of cancer. Also, receiver operating characteristic (ROC) curve analysis demonstrated that SYNGR2 showed high accuracy in distinguishing cancerous tissues from normal ones. Moreover, SYNGR2 expression was correlated with TMB, MSI, immune scores, and immune cell infiltrations. We also analyzed the association of SYNGR2 with immunotherapy response in LIHC. Finally, a nomogram including SYNGR2 and pathologic T, N, M stage was built and exhibited good predictive power for the OS, DSS, and PFI of LIHC patients. CONCLUSION: Overall, SYNGR2 is a critical oncogene in various tumors. SYNGR2 participates in the carcinogenic progression, and may contribute to the immune infiltration in tumor microenvironment. Our study suggests that SYNGR2 can serve as a predictor related to prognosis in pan-cancer, especially LIHC.

Deep learning has achieved good performance in short-term traffic forecasting recently. However, the stochasticity and distribution imbalance are main characteristics to traffic flow, and these will bring the uncertainty and induce the network overfitting problem during deep learning. To deal with the problems, a new end-to-end hybrid deep learning network model, named M-B-LSTM, is proposed for short-term traffic flow forecasting in this paper. In the M-B-LSTM model, an online self-learning network is constructed as a data mapping layer to learn and equalize the traffic flow statistic distribution for reducing the effect of distribution imbalance and overfitting problem during network learning. Besides, the deep bidirectional long short-term memory network (DBLSTM) is introduced to reduce the uncertainty problem by forward and reverse contexts approximation process in the stochasticity reducing layer, and then the long short-term memory network (LSTM) is used to forecast the next traffic flow state in the forecasting layer. Furthermore, sufficient comparative experiments have been conducted and the results show the proposed model has better ability on solving uncertainty and overfitting problems than the state-of-art methods.

Recently, the performance of super-resolution has been improved by the stereo images since the additional information could be obtained from another view. However, it is a challenge to interact the cross-view information since disparities between left and right images are variable. To address this issue, we propose a disparity feature alignment module (DFAM) to exploit the disparity information for feature alignment and fusion. Specifically, we design a modified atrous spatial pyramid pooling module to estimate disparities and warp stereo features. Then we use spatial and channel attention for feature fusion. In addition, DFAM can be plugged into an arbitrary SISR network to super-resolve a stereo image pair. Extensive experiments demonstrate that DFAM incorporates stereo information with less inference time and memory cost. Moreover, RCAN equipped with DFAMs achieves better performance against state-of-the-art methods. The code can be obtained at https://github.com/JiawangDan/DFAM.

The architecture of the convolutional neural network (CNN) plays an important role in single image super-resolution (SISR). However, most models proposed in recent years usually transplant methods or architectures that perform well in other vision fields. Thence they do not combine the characteristics of super-resolution (SR) and ignore the key information brought by the recurring texture feature in the image. To utilize patch-recurrence in SR and the high correlation of texture, we propose a residual fractal convolutional block (RFCB) and expand its depth and width to obtain residual fractal network (RFN), which contains two variations, deep residual fractal network (DRFN) and wide residual fractal network (WRFN). RFCB is recursive with multiple branches of magnified receptive field. Through the phased feature fusion module, the network focuses on extracting high-frequency texture feature that repeatedly appear in the image. We also introduce residual in residual (RIR) structure to RFCB that enables abundant low-frequency feature feed into deeper layers and reduce the difficulties of network training. RFN is the first supervised learning method to combine the patch-recurrence characteristic in SISR into network design. Extensive experiments demonstrate that RFN outperforms state-of-the-art SISR methods in terms of both quantitative metrics and visual quality, while the amount of parameters has been greatly optimized. The source code and pre-trained models are released in https://github.com/JiahangGu/RFN.

Video captioning is one of research hotspots in computer vision. At present, video captioning algorithms mainly have following problems: First, traditional algorithms use equal-interval sampling to extract video features, which causes the loss of key frames containing a large amount of semantic information, thus leading to the inaccuracy of video captioning. Moreover, equal-interval sampling method results in lots of redundant frames, thereby increasing the amount of computation of algorithms extremely. Second, traditional algorithms only consider temporal information when extracting features. However, for the image and video, the spatial features also contain rich latent semantic information. Only extracting temporal features will lead to inaccurate natural language descriptions. To address these problems, we propose the video captioning method based on key semantic frames extraction and spatio-temporal attention mechanism (KSF-ST) in this paper. In order to extract key semantic frames, knowledge graph is adopted to obtain key semantic information of video frames, and knowledge reasoning is used to obtain the correlation among entities in the knowledge graph. In order to extract spatial latent semantic information of video frames, spatial attention mechanism is combined with temporal features to generate accurate natural language descriptions. We evaluate KSF-ST on two benchmark datasets. Extensive experiments have been conducted and the results demonstrate that our algorithm could achieve better video captioning performance than the state-of-the-art algorithms.

In this paper, we focus on a cellular network with a energy-constrained unmanned aerial vehicle (UAV), which serves as a relay for all the users to improve their uplink rates via cooperative communication. Uplink resource allocation in terms of power and time allocation among users is investigated to optimize the uplink sum-rate. By leveraging the local search strategy, the sum-rate optimization problem is decomposed into two subproblems, i.e., energy allocation and time allocation. We derived the optimal solution for both amplify-and-forward (AF) and decode- and-forward (DF) protocols with in-depth analysis. Numerical results show that path loss exponent has non- trivial impact on the sum-rate while higher energy replenishment rate of the UAV relay brings little performance gain. In addition, location of the UAV is also non-negligible and the cellular network can significantly benefit from proper deployment of the UAV relay.

In this paper, we consider a multiuser downlink communication system, where ground users are served by a UAV base station flying on an aerial plane. For the sake of computational complexity and energy economy, we propose a polygonal-line trajectory for the UAV base station and each ground user is served one by one while the UAV base station is travelling along one of the line segments on the trajectory. With the polygonal- line trajectory, joint optimization for power control and trajectory design for the UAV base station is investigated to maximize total capacity of all the ground users. Such a problem is decomposed into three subproblems for single-user downlink power profile, transmission energy allocation and waypoints location selection, respectively, and solved via alternate optimization, with which one of the three subproblems is solved given solutions to the other two. Simulation results show that the proposed scheme with optimized power profile and trajectory for the UAV base station outperforms two baselines. We also show that trajectory design significantly impact the system performance and is indispensable in UAV-assisted communications.

In this paper, we focus on fundamental point-to-point wireless transmission with a reverse feedback channel for automatic repeat request (ARQ) and propose wireless energy replenishment (WER) for the transmitter with ACK or NACK signal sent back by the receiver. In such a case, transmission power control affects not only transmission reliability at the receiver but also WER performance at the transmitter and transmission sustainability. Hence, dynamic power control for the transmitter is investigated to optimize the expected number of transmitted bits in multiple timeslots and such a problem is modeled as a finite-horizon Markov decision process (MDP), which is further solved via the backward induction algorithm. We evaluate the transmission performance with impact of channel condition, WER performance, number of timeslots as well as modulation scheme and show that transmission with WER outperforms that without WER, especially for moderate channel condition.

In this paper, we consider a cooperative communication system, where the relay node is capable of simultaneous wireless information and power transfer (SWIPT) by splitting the signal from the source node into two power streams for energy harvesting and information relaying, and study the optimal design to maximize the cooperative capacity for both amplify-and-forward (AF) and decode-and- forward (DF) protocols. The closed-form optimal power-split ratio is derived for the two protocols with in-depth analysis and performance in cooperative capacity is evaluated for the optimal power splitting scheme with AF and DF protocols. Experiment results show that AF protocol benefits more from favorable cooperation link condition than the DF protocol and cooperation link condition contributes more to the optimal cooperative capacity compared with direct link condition. Asymptotic analysis also show that signal processing noise is also non-negligible to the optimal cooperative capacity.

In cooperative spectrum sensing, wireless reporting channels (from local sensing nodes to fusion center) may suffer severe unreliability, which would make a correct local sensing result incorrect while received by fusion center. In this study, the reliability of the sensing result transmission from local sensor to fusion center is considered, and to make spectrum sensing more energy efficient, appropriate nodes were selected and activated to participate in spectrum sensing while others remain idle. To further reduce energy consumption, transmission power control for sensing nodes was optimized. Total energy consumption minimization was modeled as a mixed discrete and continuous variable optimization problem and the binary particle swarm optimization with power control (BPSO-PC) was proposed. BPSO-PC adjusted the sensing node transmission power and properly selects nodes for cooperative spectrum sensing. Simulation results showed that total energy consumption was significantly reduced compared with BPSO and three other sensing nodes selection algorithms.

In this paper, we consider an energy harvesting cognitive radio (CR) system operating in slotted mode, where the secondary user (SU) has no wired power supplies and is powered exclusively by energy harvested from ambient environment. The SU can only perform either energy harvesting, spectrum sensing or data transmission at a time due to hardware limitation such that a timeslot is segmented into three non-overlapping fractions. Considering a generalized multi-slot spectrum sensing paradigm and two types of fusion rules: data fusion and decision fusion, we focus on the “harvesting-sensing-throughput” tradeoff and joint optimization for save-ratio, sensing duration, sensing threshold as well as fusion rule to maximize the SU's expected achievable throughput while keeping primary users (PUs) sufficiently protected. For data-fusion spectrum sensing, we translate the original problem into a convex one and show that the optimal solutions for sample number, mini-slot number as well as sensing threshold are non-unique. For decision-fusion spectrum sensing, we propose a two-level algorithm to solve the original problem with in-depth analysis on the convexity of a simplified problem and experiments show that the proposed algorithm is more efficient than differential evolution algorithm. We find that despite the inherent difference between the two types of fusion rules, the optimal data-fusion and decision-fusion strategies both converge to single-slot spectrum sensing while the SU's maximal expected achievable throughput is attained. Simulation results show that the optimal single-slot spectrum sensing strategy outperforms three other multi-slot strategies as well as two existing strategies while the empirical probability of detection is limited under a predefined level.

In this paper, we consider a cognitive radio (CR) system operating in slotted mode, where the secondary user (SU) has no wired power supplies and is powered exclusively through extracting energy from ambient environment. Due to hardware limitation, the SU can only perform either of energy harvesting, spectrum sensing or data transmission at the same time and we assume that a timeslot is partitioned into three non-overlapping fractions. Considering a generalized multi-slot spectrum sensing paradigm with data fusion rule, we focus on the "saving-sensing-throughput" tradeoff and joint optimization for save-ratio, sensing duration as well as sensing threshold to maximize the SU's expected achievable throughput while keeping primary users (PUs) sufficiently protected. By translating the original optimization problem into a convex one, we show that the optimal solutions for sample number, mini-slot number as well as sensing threshold are non-unique and single-slot spectrum sensing is more practically preferable. We also investigate the impact of system parameters to the optimal sensing strategy.

In recent years, the excessive energy consumption in wireless communication systems has been increasingly critical, and environmental and financial considerations have motivated a trend in wireless communication technologies to resort to renewable energy sources. Energy harvesting is considered as a promising solution to alleviate such issues and has received extensive attentions. In this paper, we consider a cognitive radio system with one primary user (PU) and one secondary user (SU) and both of their transmitters operate in time-slotted mode. The SU, which harvests energy exclusively from ambient radio signal, follows a save-then-transmit protocol. In such a scenario, we investigate the SU's optimal cooperation strategy, namely, the optimal decision (to cooperate with the PU or not) and the optimal action (to spend how much time on energy harvesting and to allocate how much power for cooperative relay). We separately investigate the optimal action in non-cooperation and cooperation modes to maximize the SU's achievable throughout and derive the optimal closed-form solutions. Based on the analytical results of the optimal solutions, we propose the optimal cooperation protocol (OCP) to make the optimal decision, which simply involves a two-level test. Simulation results show that the proposed OCP outperforms the other two protocols (non-cooperation protocol and stochastic cooperation protocol) and the optimal underlay (OU) transmission mode.

For the sake of the increase of energy consumption, the emission of carbide can't be negligible. In this paper, we propose a mathematical model which adjusts the radius of base station, to figure out the problem of energy saving, while the quality of services such as the traffic volume of each base station and the service area should be satisfied. For the model, we use ant colony algorithm as a practical method to solve this problem. By using this mechanism, we can achieve the purpose of energy saving, when the traffic in a service area is low.

As one solution to the gradual ossification of the existing networks, network virtualization enables multiple service providers (SPs) to coexist on a shared infrastructure, and it is considered as an integral part of next generation architecture. During the run time, SPs have exclusive rights for the allocated bandwidth resources, which may result in poor performance of bandwidth utilization. In this paper, we introduce a novel bandwidth reuse (BR) mechanism to allow SPs lease their idle bandwidth resources to other SPs as a virtual infrastructure provider (InP). In essence, the BR mechanism is a truthful auction with optimal expected revenue generation, which incentivizes SPs to open up their idle bandwidth significantly. Simulation results demonstrate that the BR mechanism could efficiently generate revenue and improve bandwidth utilization.