The Journal of Supercomputing最新文献

Due to the varying data distributions in different domains, transferring sentiment classification models across domains is often infeasible. Additionally, labeling data in specific domains can be both costly and time-consuming. To address these challenges, multi-source cross-domain sentiment classification leverages knowledge from multiple source domains to aid in sentiment classification in the target domain, utilizing labeled data from these sources. This paper introduces a novel multi-source cross-domain sentiment classification method that leverages collaborative training and selective ensemble classification. By utilizing unlabeled data from the target domain and labeled data from multiple source domains, our method reduces the need for manual labeling and enhances classification accuracy. Empirical evaluations on the Amazon multi-domain review dataset show that our approach achieves an average accuracy of 0.8932 ± 0.012 (0.95 confidence interval), demonstrating significant improvements in robustness and efficiency.

Monocular depth estimation is essential in computer vision and robotics applications, including localization, mapping, and 3D object detection. In recent years, supervised learning algorithms that model large amounts of data have been successful in depth estimation. However, obtaining dense ground truth depth labels remains a challenge in supervised training. Therefore, unsupervised methods trained using monocular image sequences have gained wider attention. However, the depth estimation results of most existing models often produce blurred edges. Therefore, we propose various effective improvement strategies to construct a depth estimation network TL-Depth. (1) We propose a tower connection structure that utilizes convolutional processing to facilitate feature fusion, achieve precise semantic classification of pixels, and yield more accurate depth results. (2) We employ a Laplacian-filtering residual to focus on boundary information and enhance detailed results. (3) During the feature extraction stage, multiple pooling excitations are used by embedding them in the convolutional layer. This reduces redundant information while enhancing the network’s feature extraction capability. The experimental results on the KITTI dataset and the Make3D dataset demonstrate that this method achieves good results compared to current methods.

White blood cells (WBCs) are crucial microscopic defenders of the human immune system in combating transmittable conditions triggered by germs, infections, and various other human pathogens. Timely and appropriate WBC detection and classification are decisive for comprehending the immune system’s standing and its feedback to various pathologies, assisting in diagnosing and monitoring illness. Nevertheless, the manual classification of WBCs is strenuous, extensive, and prone to errors, while automated approaches can be cost-prohibitive. Within artificial intelligence, deep learning (DL) approaches have become an appealing option for automating WBC recognition. The existing DL techniques for WBC classification face several limitations and computational difficulties, such as overfitting, limited scalability, and design complexity, often battling with function variety in WBC images and requiring considerable computational resources. This research study recommends an ingenious hybrid inception-xception Convolutional Semantic network (CNN) designed to deal with constraints in existing DL versions. The proposed network incorporates inception and depth-separable convolution layers to successfully catch attributes across many ranges, therefore minimizing concerns related to model complexity and overfitting. In contrast to traditional CNN designs, the proposed network lessens the layers made use of and increases their function removal capacities, hence enhancing the performance of WBC classification, which needs a wide variety of attribute abilities. Furthermore, the proposed model was trained, validated and tested on three popular and widely recognized datasets, namely, Leukocyte Images for Segmentation and Classification (LISC), Blood Cell Count and Detection (BCCD), and Microscopic PBS (PBS-HCB), where it demonstrates the generalization and robustness and superiority of our proposed model. The model depicted an outstanding average accuracy rate of 99.25%, 99.65%, and 98.6% on a five-fold cross-validation test for the respective datasets, surpassing existing models as detailed. The model’s robustness and superior performance, validated across diverse datasets, underscore its potential as an advanced tool for accurate and efficient WBC classification in medical diagnostics.

The fog computing model is a new computing model that has been proposed in recent years by increasing the number of requests sent to the cloud to reduce the delay and workload of the cloud computing model. In addition to its advantages, the fog computing model also has challenges, among which we can mention the issue of service placement in this computing model, which is very effective in the performance of the computing model. So far, many works have been presented to solve the problem of service deployment by considering different goals such as energy consumption, end-to-end delay, load balancing, resource efficiency, etc. Considering the importance of all the mentioned parameters, it is very important to provide a multi-objective method. In multi-objective problems, the method of evaluating the generated solutions is a separate challenge. Therefore, in this paper, a service placement method is presented by considering end-to-end delay criteria and energy consumption based on the modified Harris hawks algorithm to solve multi-objective problems. To increase accuracy, in the proposed method called multi-objective Harris hawks optimization, a multi-objective problem is modeled as several single-objective problems. The simulation results in CloudSim show that the proposed method has achieved better results than other algorithms in terms of reducing energy consumption, end-to-end delay, and network utilization.

In recent years, the rapid integration of Internet of Things (IoT) devices into the healthcare sector has brought about revolutionary advancements in patient care and data management. While these technological innovations hold immense promise, they concurrently raise critical security concerns, particularly in safeguarding medical data against potential cyber threats. The sensitive nature of health-related information requires robust measures to ensure patient data's confidentiality, integrity, and availability within IoT-enabled medical environments. Addressing the imperative need for enhanced security in IoT-based healthcare systems, we propose a comprehensive method encompassing three distinct phases. In the first phase, we implement blockchain-enabled request and transaction encryption to fortify the security of data transactions, providing an immutable and transparent framework. Subsequently, in the second phase, we introduce request pattern recognition check, leveraging diverse data sources to identify and thwart potential unauthorized access attempts. Finally, the third phase incorporates feature selection and the BiLSTM network to enhance the accuracy and efficiency of intrusion detection through advanced machine-learning techniques. We compared the simulation results of the proposed method with three recent related methods, namely AIBPSF-IoMT, OMLIDS-PBIoT, and AIMMFIDS. The evaluation criteria encompass detection rates, false alarm rates, precision, recall, and accuracy, crucial benchmarks in assessing the overall performance of intrusion detection systems. Notably, our findings reveal that the proposed method outperforms these existing methods across all evaluated criteria, underscoring its superiority in enhancing the security posture of IoT-based healthcare systems.

To address the path planning problem for unmanned combat aerial vehicle (UCAV) more effectively, a novel two-stage path planning model is proposed. The first stage involves a longitudinal search primarily aimed at predicting the initial path, while the second stage is a horizontal search designed to correct the initial path. Furthermore, to tackle the UCAV path planning issue more effectively, this paper designs an improved elephant clan optimization (IECO) algorithm based on the average sample learning strategy, opposition-based learning, and Lévy flight disturbance strategy. Subsequently, IECO is integrated with the two-stage model (TSIECO) to address the UCAV path planning problem. Additionally, numerical experiments across 15 test functions reveal that IECO outperforms other algorithms in terms of optimization capability and convergence speed. Finally, the UCAV path planning experimental results indicate that the two-stage model based on IECO, as proposed in this paper, has significant advantages over traditional path planning models based on other swarm intelligence algorithms. Specifically, in three different simulated environments, the TSIECO has been tested on a total of 9 maps with varying parameters, yielding paths that are optimal in terms of cost and stability.

In the realm of heterogeneous computing, the efficient allocation of resources is pivotal for optimizing system performance. However, user-submitted tasks are often complex and have varied resource demands. Moreover, the dynamic nature of resource states in such platforms, coupled with variations in resource types and capabilities, results in significant intricacy of the system environment. To this end, we propose a scheduling algorithm based on hierarchical reinforcement learning, namely MD-HRL. Such an algorithm could simultaneously harmonize task completion time, device power consumption, and load balancing. It contains a high-level agent (H-Agent) for task selection and a low-level agent (L-Agent) for resource allocation. The H-Agent leverages multi-hop attention graph neural networks (MAGNA) and one-dimensional convolutional neural networks (1DCNN) to encode the information of tasks and resources. Kolmogorov–Arnold networks is then employed for integrating these representations while calculating subtask priority scores. The L-Agent exploits a double deep Q network to approximate the best strategy and objective function, thereby optimizing the task-to-resource mapping in a dynamic environment. Experimental results demonstrate that MD-HRL outperforms several state of the art baselines. It reduces makespan by 12.54%, improves load balancing by 5.83%, and lowers power consumption by 6.36% on average compared with the suboptimal method.

Medical image semantic segmentation is a crucial technique in medical imaging processing, providing essential diagnostic support by precisely delineating different tissue structures and pathological areas within an image. However, the pursuit of higher accuracy has led to increasingly complex architectures in existing networks, resulting in significant training overhead. In response, this study introduces a flattened, minimalist design philosophy and constructs the shallow super convolution U-shaped Net (SSCU-Net) based on this concept. Compared to the traditional four-layer U-shaped networks, SSCU-Net has a simplified two-layer structure, adhering to a lightweight research objective. On one hand, to address the issue of insufficient semantic feature extraction caused by the shallow network architecture, a parallel multi-branch feature extraction module called the super convolution block is designed to thoroughly extract diverse semantic information. On the other hand, to facilitate the transfer of critical semantic information between encoding and decoding, as well as across layers, the spatial convolution path, along with feature enhanced downsample and feature resolution upsample, are constructed. The performance of SSCU-Net was validated against 18 comparison models across seven metrics on five datasets. Results from metric analysis, image comparisons, and ablation tests collectively demonstrate that SSCU-Net achieves an average improvement of 15.9792% in the Dice coefficient compared to other models, confirming the model’s advantages in both lightweight design and accuracy. The network code is available at https://github.com/YF-W/SSCU-Net.

Traditional LSM-tree-based key-value storage systems face significant read amplification issues due to the multi-level structure of LSM-tree, the unordered SSTable files in Level 0, and the lack of an in-memory index structure for key-value pairs. We observed that, under the influence of workloads with locality features, key-value pairs exhibit a range-specific access intensity. Addressing the three reasons for LSM-tree read amplification, we have utilized the range-specific access intensity of workload to propose a multi-tree structure consisting of a B+ tree, a single-level hot tree, and an LSM-tree with partition-based Level 0. This aims to enhance the read performance of LSM-tree-based key-value storage systems. We designed the prototype, MTDB, based on LevelDB. The experimental results show that MTDB’s read performance is 1.62× to 2.02× that of LevelDB, and it approaches or exceeds the read performance of KVell and Bourbon while reducing memory overhead by 58.85%–86%.

Many-objective optimization problem is one of the most important and widely faced optimization problems in the real world. To solve many-objective optimization problems (MaOPs), numerous multi-objective evolutionary algorithms (MOEAs) have been developed to find a good convergence and well-distributed Pareto front. However, with the increase of dimensions, the distribution of solutions obtained by MOEAs becomes more complex and tends to be orthogonal, which significantly reduces the effectiveness of the algorithms. In this paper, we propose an improved many-objective evolutionary algorithm (MaOEA-MSAR), which incorporates a multi-strategy selection mechanism into an existing MOEA, and develops an adaptive reproduction operation to produce promising offspring individuals. Firstly, the selection strategy based on the angle-penalized distance is used to improve the coverage of the solutions in the objective space. Then, the selection strategy based on convergence rate is employed to strengthen the balance between diversity and convergence. Finally, an adaptive reproduction operation is used to select different reproduction strategies for the gene-level global exploration or local exploitation. A series of experiments are carried out against seven state-of-the-art many-objective optimization algorithms. Experimental results on commonly used 31 benchmark test problems with up to 15 objectives and a multi-objective vehicle routing problem have demonstrated that MaOEA-MSAR is competitive in handling various kinds of MaOPs.