The Journal of Supercomputing最新文献

Existing segmentation-based text detection methods generally face the problems of insufficient receptive fields, insufficient text information filtering, and difficulty in balancing detection accuracy and speed, limiting their ability to detect arbitrary-shaped text in complex backgrounds. To address these problems, we propose a new text detection method fusing the pure ConvNet model InceptionNeXt and the multi-scale attention mechanism. Firstly, we propose a text information reinforcement module to fully extract effective text information from features of different scales while preserving spatial position information. Secondly, we construct the InceptionNeXt Block module to compensate for insufficient receptive fields without significantly reducing speed. Finally, the INA-DBNet network structure is designed to fuse local and global features and achieve the balance of accuracy and speed. Experimental results demonstrate the efficacy of our method. Particularly, on the MSRA-TD500 and Total-text datasets, INA-DBNet achieves 91.3% and 86.7% F-measure while maintaining real-time inference speed. Code is available at: https://github.com/yuyu678/INANET.

In the development of industrial internet of things (IIoT), the application of mobile edge computing (MEC) has significantly enhanced the efficiency of task computation offloading. However, the risk of data privacy leakage persists during the computation offloading process. Considering the diversity of task data sensitivity and the variability in server security protection capabilities, this paper proposes a method for assessing privacy satisfaction. To ensure offloading efficiency while effectively enhancing privacy security, we have proposed an IIoT cloud-edge-device computation offloading algorithm based on dual-Q network deep reinforcement learning, named D2PCO, to optimize the computation offloading process in IIoT tasks. The incorporation of the dual-Q network notably enhances the algorithm’s learning ability and efficiency in dealing with complex decision-making problems. Experimental results show that the proposed D2PCO algorithm significantly improves user privacy satisfaction while ensuring low delay. Compared with MA3MCO, DDPG, shortest distance priority, and random scheduling algorithms, it reduces the average offloading delay by 4.15%, 9.98%, 13.2%, and 26.47% and increases privacy satisfaction by 0.8%, 4.26%, 10.15%, and 30.30%, respectively.

Multimodal sentiment analysis is an important and active research field. Most methods construct fusion modules based on unimodal representations generated by pretrained models, which lack the deep interaction of multimodal information, especially the rich semantic-emotional information embedded in text. In addition, previous studies have focused on capturing modal coherence information and ignored differentiated information. We propose a text-enhanced multi-interactive attention and multitask learning network (TEMM). First, syntactic dependency graphs and sentiment graphs of the text are constructed, and additional graph embedding representations of the text are obtained using graph convolutional networks and graph attention networks. Then, self-attention and cross-modal attention are applied to explore intramodal and intermodal dynamic interactions, using text as the main cue. Finally, a multitask learning framework is constructed to exert control over the information flow by monitoring the mutual information between the unimodal and multimodal representations and exploiting the classification properties of the unimodal modality to achieve a more comprehensive focus on modal information. The experimental results on the CMU-MOSI, CMU-MOSEI, and CH-SIMS datasets show that the proposed model outperforms state-of-the-art models.

Multi-access edge computing (MEC), as a computing model that provides services on the user side, can effectively solve the problems of high delay and resource shortage in traditional cloud computing when processing massive data. However, existing edge computing resources are still limited, and difficult to provide services to users in inaccessible remote areas. Considering that unmanned aerial vehicle (UAV) has the advantages of easy deployment, high flexibility and low cost, a UAV-assisted MEC hierarchical computation offloading framework is proposed. Firstly, contract theory is used to solve the information asymmetry problem between the platform and the UAV, and the UAV is encouraged to provide computing services. By analyzing the attributes and conditions of feasible contracts, the optimal contract is designed using the Lagrange multiplier method. Secondly, by constructing the preference set between UAV and mobile user (MU), a mobile user and unmanned aerial vehicle bilateral matching (MUBM) algorithm is proposed to establish the connection between user tasks and UAV computing resources. Finally, the feasibility and effectiveness of the contract were verified through experiments. The experimental results also prove the stability of the MUBM.

In this paper, we focus on examining the (K_{1,2})-structure-connectivity of any connected graph. Let G be a connected graph with n vertices, we show that (kappa (G; K_{1,2})) is well defined if (hbox {diam}(G)ge 4), or (nequiv 1pmod 3), or (Gnotin {C_{5},K_{n}}) when (nequiv 2pmod 3), or there exist three vertices u, v, w such that (N_{G}(u)cap (N_{G}({v,w})cup {v,w})=emptyset) when (nequiv 0pmod 3). Furthermore, if G has (K_{1,2})-structure-cut, we prove (kappa (G)/3le kappa (G; K_{1,2})le kappa (G)).

Early detection of lung nodules is essential for lung cancer screening and improving patient survival rates. Traditional object detection networks such as YOLO and Faster R-CNN have shown promising results in detecting lung nodules but often lack sufficient integration of extracted features to enhance accuracy and efficiency. Moreover, these methods typically do not retain the spatial information of lung nodules from the original CT images. To overcome these limitations, a novel lung nodule detection algorithm based on YOLOv7 is introduced. Firstly, to better preserve essential features and minimize interference from irrelevant background noise, a deformable attention module for feature fusion has been designed. Additionally, maximum intensity projection is employed to create projection images at various intensities, thereby enriching the spatial background information that is often missing in single CT slices. Thirdly, the WIoU loss function is utilized to replace the original YOLOv7 loss function, aiming to reduce the influence of low-quality samples on the gradient within the dataset. The proposed model was validated using the publicly available LUNA16 dataset and achieved a recall rate of 94.40% and an AP value of 95.39%. These results demonstrate the enhanced precision and efficiency of lung nodule detection.

Efficient utilization of processors in heterogeneous CPU–GPU systems is crucial for improving overall application performance by reducing workload completion time. This article introduces a framework designed to achieve maximum performance in scheduling the processing of sparse matrix-based applications within a heterogeneous CPU–GPU system. The framework suggests splitting the matrix into chunks, employing machine learning to find the optimal chunk size for scheduling efficiency, with the number of GPU streams regarded as a critical factor. The scheduling algorithm introduced is inspired by the concept of quartiles in statistics and is designed to operate in real-time, thereby striving to impose minimal overhead on the system. The evaluation of the proposed framework focused on the SpMV (Sparse Matrix–Vector Multiplication) kernel, essential for various applications such as matrix-based graph processing. This evaluation was conducted using a system equipped with an NVIDIA GTX 1070 GPU. Testing on real-world sparse matrices showed that the proposed scheduling algorithm significantly outperforms scenarios with no offloading, full offloading, and the Alternate Assignment method.

Researchers often use machine learning and deep learning to predict price trends in the financial markets, aiming to achieve high returns. However, accurately predicting market prices is challenging due to their nonlinear and seemingly random nature. Improving the accuracy of the prediction model is the common focus of researchers, yet it is crucial to also consider the data used in training. Traditional labeling methods used in most price trend prediction studies are not robust as they are sensitive to small price changes, leading to inefficient model training. To address this issue, this study introduces a Dynamic Threshold Breakout (DTB) labeling system that labels data based on the price percentage change during a specific period. This proposed labeling system was then integrated into an automated trading system using LightGBM and evaluated using three different markets. The results showed that the DTB labeling method is effective for trading in financial markets in terms of winning ratio, payoff ratio, profit factor, accuracy and ROI in trading performance.

The security and privacy of data are major concerns in the mobile crowdsensing (MCS) environment due to the huge amount of heterogeneous data received from various users and devices automatically or manually regarding their surrounding environment. User participation in the MCS approach is highly essential to have a vast dataset for analysis that will provide the required information or beneficial solution for society. However, it is difficult to achieve due to huge energy consumption, the need for internet connectivity for data transmission, and the security and privacy of data. Therefore, it is essential to have a network coverage model in which data transmission can be done with minimal energy consumption and the need for internet connectivity can be removed from the user’s side. The user’s sensitive data needs to be protected from internal and external attackers to improve the efficiency of the solution provided by the MCS environment with genuine data. This work is based on data collection from users based on their experience for a certain location using the hybrid network coverage model based on clustering, in which each location may have just one or multiple heterogeneous cluster heads. Discrete event-based CrowdSenSim Simulator has been used to design a simulation environment in urban spaces in which 2000 users will move to any location randomly among considered 40 locations and provide feedback data for the location. In this paper, a novel security mechanism based on multiple heterogeneous cluster heads per location has been presented, and it provides better security against attackers than the security model with one cluster head per location. The proposed multiple-cluster heads per location (MCHL)-based mechanism has been compared with the vulnerable one-cluster head per location (OCHL)-based mechanism on the basis of the average number of rounds attackers attacked, average number of locations attackers attacked, average coverage and average efficiency of attackers, and average efficiency of system security.

Triangular (gamma)-graphyne has a special carbon–carbon bonding arrangement, which results in outstanding electrical characteristics. It is a potential material for semiconductors and conductors in nanoelectronic devices. The number of vertices at a distance of 2 from a vertex is known as the connection number (CN) of that vertex. In this paper, we computed Zagreb-type indices based on connection numbers. In order to give us a better knowledge of the structural properties of molecules or networks, these indices are calculated. Following the computation of these indices, we investigated their use in computing entropy, providing important new information about the thermodynamic characteristics and complexity of the understudied systems. We used Python language to find the Pearson correlation coefficient between indices and entropy and show its heat map.