The Journal of Supercomputing最新文献

Cloud application programming interface (API) is a software intermediary that enables applications to communicate and transfer information to one another in the cloud. As the number of cloud APIs continues to increase, developers are inundated with a plethora of cloud API choices, so researchers have proposed many cloud API recommendation methods. Existing cloud API recommendation methods can be divided into two types: content-based (CB) cloud API recommendation and collaborative filtering-based (CF) cloud API recommendation. CF methods mainly consider the historical information of cloud APIs invoked by mashups. Generally, CF methods have better recommendation performances on head cloud APIs due to more interaction records, and poor recommendation performances on tail cloud APIs. Meanwhile, CB methods can improve the recommendation performances of tail cloud APIs by leveraging the content information of cloud APIs and mashups, but their overall performances are not as good as those of CF methods. Moreover, traditional cloud API recommendation methods ignore the complementarity relationship between mashups and cloud APIs. To address the above issues, this paper first proposes the complementary function vector (CV) based on tag co-occurrence and graph convolutional networks, in order to characterize the complementarity relationship between cloud APIs and mashups. Then we utilize the attention mechanism to systematically integrate CF, CB, and CV methods, and propose a model named Content and Complementarity enhanced Attentional Collaborative Filtering (CCeACF). Finally, the experimental results show that the proposed approach outperforms the state-of-the-art cloud API recommendation methods, can effectively alleviate the long tail problem in the cloud API ecosystem, and is interpretable.

Markov chains and Markov decision processes have been widely used to model the behavior of computer systems with probabilistic aspects. Numerical and iterative methods are commonly used to analyze these models. Many efforts have been made in recent decades to improve the efficiency of these numerical methods. In this paper, focusing on Markov models with non-sparse structure, a new set of heuristics is proposed for prioritizing model states with the aim of reducing the total computation time. In these heuristics, a set of simulation runs are used for statistical analysis of the effect of each state on the required values of the other states. Under this criterion, the priority of each state in updating its values is determined. The proposed heuristics provide a state ordering that improves the value propagation among the states. The proposed methods are also extended for very large models where disk-based techniques are required to analyze the models. Experimental results show that our proposed methods in this paper reduce the running times of the iterative methods for most cases of non-sparse models.

The increasing growth of the Internet of Things (IoT) and its open and shared character has exponentially led to a rise in new attacks. Consequently, quick and adaptive detection of attacks in IoT environments is essential. The Intrusion Detection System (IDS) is responsible for protecting and detecting the type of attacks. Creating an IDS that works in real time and adapts to environmental changes is critical. In this paper, we propose a Deep Reinforcement Learning-based (DRL) self-learning IDS that addresses the mentioned challenges. DRL-based IDS helps to create a decision agent, who controls the interaction with the indeterminate environment and performs binary detection (normal/intrusion) in fog. We use the ensemble method to classify multi-class attacks in the cloud. The proposed approach was evaluated on the CIC-IDS2018 dataset. The results demonstrated that the proposed model achieves a superior performance in detecting intrusions and identifying attacks to compare other machine learning techniques and state-of-the-art approaches. For example, our suggested method can detect Botnet attacks with an accuracy of 0.9999% and reach an F-measure of 0.9959 in binary detection. It can reduce the prediction time to 0.52 also. Overall, we proved that combining multiple methods can be a great way for IDS.

The integration of generative artificial intelligence (GAI) and internet of vehicles (IoV) will transform vehicular intelligence from conventional analytical intelligence to service-specific generative intelligence, enhancing vehicular services. In this context, computing force networks (CFNs), capable of flexibly scheduling widespread, multi-domain, multi-layer, and distributed resources, can cater to the demands of the IoV for ultra-high-density computing power and ultra-low latency. In CFNs, the integration of GAI and IoV consumes enormous energy, and GAI servers need to purchase energy from energy suppliers (ESs). However, the information asymmetry between GAI servers and ESs makes it difficult to price energy fairly and distributed ESs and GAI servers constitute a complex trading environment where malicious ESs may intentionally provide low-quality services. In this paper, to facilitate efficient and secure energy trading, and supply for ubiquitous AIGC services, we initially introduce an innovative CFNs-based GAI energy trading system architecture; present an energy consumption model for AIGC services, cost model of ESs, and reputation evaluation model of ESs; and obtain utility functions of GAI servers and ESs based on contract theory. Then, we propose a secure incentive mechanism in IoV, including designing an optimal contract scheme based on contract feasibility conditions and a safety guarantee mechanism based on blockchain. Simulation results demonstrate the feasibility and superiority of our energy trading mechanism.

This paper introduces a cutting-edge deep learning-based model aimed at enhancing the short-term performance of microgrids by simultaneously minimizing operational costs and emissions in the presence of distributed energy resources. The primary focus of this research is to harness the potential of demand response programs (DRPs), which actively engage a diverse range of consumers to mitigate uncertainties associated with renewable energy sources (RES). To facilitate an effective demand response, this study presents a novel incentive-based payment strategy packaged as a pricing offer. This approach incentivizes consumers to actively participate in DRPs, thereby contributing to overall microgrid optimization. The research conducts a comprehensive comparative analysis by evaluating the operational costs and emissions under scenarios with and without the integration of DRPs. The problem is formulated as a challenging mixed-integer nonlinear programming problem, demanding a robust optimization technique for resolution. In this regard, the multi-objective particle swarm optimization algorithm is employed to efficiently address this intricate problem. To showcase the efficacy and proficiency of the proposed methodology, a real-world smart microgrid case study is chosen as a representative example. The obtained results demonstrate that the integration of deep learning-based demand response with the incentive-based pricing offer leads to significant improvements in microgrid performance, emphasizing its potential to revolutionize sustainable and cost-effective energy management in modern power systems. Key numerical results demonstrate the efficacy of our approach. In the case study, the implementation of our demand response strategy results in a cost reduction of 12.5% and a decrease in carbon emissions of 14.3% compared to baseline scenarios without DR integration. Furthermore, the optimization model shows a notable increase in RES utilization by 22.7%, which significantly reduces reliance on fossil fuel-based generation.

Predicting CPU usage is crucial to cloud resource management. Precise CPU prediction, however, is a tough challenge due to the variable and dynamic nature of CPUs. In this paper, we introduce TrAdaBoost.WLP, a novel regression transfer boosting method that employs Long Short-Term Memory (LSTM) networks for CPU consumption prediction. Concretely, a dedicated Periodicity-aware LSTM (PA-LSTM) model is specifically developed to take into account the use of periodically repeated patterns in time series data while making predictions. To adjust for variations in CPU demands, multiple PA-LSTMs are trained and concatenated in TrAdaBoost.WLP using a boosting mechanism. TrAdaBoost.WLP and benchmarks have been thoroughly evaluated on two datasets: 160 Microsoft Azure VMs and 8 Google cluster traces. The experimental results show that TrAdaBoost.WLP can produce promising performance, improving by 32.4% and 59.3% in terms of mean squared error compared to the standard Probabilistic LSTM and ARIMA.

Ground point filtering on national-level datasets is a challenge due to the presence of multiple types of landscapes. This limitation does not simply affect to individual users, but it is in particular relevant for those national institutions in charge of providing national-level Light Detection and Ranging (LiDAR) point clouds. Each type of landscape is typically better filtered by different filtering algorithms or parameters; therefore, in order to get the best quality classification, the LiDAR point cloud should be divided by the landscape before running the filtering algorithms. Despite the fact that the manual segmentation and identification of the landscapes can be very time intensive, only few studies have addressed this issue. In this work, we present a multistage approach to automate the identification of the type of landscape using several metrics extracted from the LiDAR point cloud, matching the best filtering algorithms in each type of landscape. An additional contribution is presented, a parallel implementation for distributed memory systems, using Apache Spark, that can achieve up to (34times) of speedup using 12 compute nodes.

Existing methods for detecting communities in attributed social networks often rely solely on network topology, which leads to suboptimal accuracy in community detection, inefficient use of available data, and increased time required for identifying groups. This paper introduces the Dual Embedding-based Graph Convolution Network (DEGCN) to address these challenges. This new method uses graph embedding techniques in a new deep learning framework to improve accuracy and speed up community detection by combining the nodes’ content with the network’s topology. Initially, we compute the modularity and Markov matrices of the input graph. Each matrix is then processed through a graph embedding network with at least two layers to produce a condensed graph representation. As a result, a multilayer perceptron neural network classifies each node’s community based on these generated embeddings. We tested the suggested method on three standard datasets: Cora, CiteSeer, and PubMed. Then, we compared the outcomes to many basic and advanced approaches using five important metrics: F1-score, adjusted rand index (ARI), normalized mutual information (NMI), and accuracy. The findings demonstrate that the DEGCN accurately captures community structure, achieves superior precision, and has higher ARI, NMI, and F1 scores, significantly outperforming existing algorithms for identifying community structures in medium-scale networks.

This paper proposes an image encryption scheme based on an improved four-dimensional chaotic system. First, a 4D chaotic system is constructed by introducing new state variables based on the Chen chaotic system, and its chaotic behavior is verified by phase diagrams, bifurcation diagrams, Lyapunov exponents, NIST tests, etc. Second, the initial chaotic key is generated using the hash function SHA-512 and plain image information. Parity scrambling is performed on the plain image using the chaotic sequence generated by the chaotic system. The image is then converted into a hexadecimal character matrix, divided into two planes according to the high and low bits of the characters and scrambled by generating two position index matrices using chaotic sequences. The two planes are then restored to a hexadecimal character matrix, which is further converted into the form of an image matrix. Finally, different combined operation diffusion formulas are selected for diffusion according to the chaotic sequence to obtain the encrypted image. Based on simulation experiments and security evaluations, the scheme effectively encrypts gray images and shows strong security against various types of attacks.

In recent years, RGBT trackers based on the Siamese network have gained significant attention due to their balanced accuracy and efficiency. However, these trackers often rely on similarity matching of features between a fixed-size target template and search region, which can result in unsatisfactory tracking performance when there are dramatic changes in target scale or shape or occlusion occurs. Additionally, while these trackers often employ feature-level fusion for different modalities, they frequently overlook the benefits of decision-level fusion, which can diminish their flexibility and independence. In this paper, a novel Siamese tracker through graph attention and reliable modality weighting is proposed for robust RGBT tracking. Specifically, a modality feature interaction learning network is constructed to perform bidirectional learning of the local features from each modality while extracting their respective characteristics. Subsequently, a multimodality graph attention network is used to match the local features of the template and search region, generating more accurate and robust similarity responses. Finally, a modality fusion prediction network is designed to perform decision-level adaptive fusion of the two modality responses, leveraging their complementary nature for prediction. Extensive experiments on three large-scale RGBT benchmarks demonstrate outstanding tracking capabilities over other state-of-the-art trackers. Code will be shared at https://github.com/genglizhi/SiamMGT.