Journal of Supercomputing最新文献

Financial time series have been extensively studied within the past decades; however, the advent of machine learning and deep neural networks opened new horizons to apply supercomputing techniques to extract more insights from the underlying patterns of price data. This paper presents a tri-state labeling approach to classify the underlying patterns in price data into up, down and no-action classes. The introduction of a no-action state in our novel approach alleviates the burden of denoising the dataset as a preprocessing task. The performance of our labeling algorithm is experimented with using machine learning and deep learning models. The framework is augmented by applying the Bayesian optimization technique for the selection of the best tuning values of the hyperparameters. The price trend prediction module generates the required trading signals. The results show that the average annualized Sharpe ratio as the trading performance metric is about 2.823, indicating the framework produces excellent cumulative returns.

Rumor detection aims to judge the authenticity of posts on social media (such as Weibo and Twitter), which can effectively prevent the spread of rumors. While many recent rumor detection methods based on graph neural networks can be conducive to extracting the global features of rumors, each node of the rumor propagation structure learned from graph neural networks is considered to have multiple individual scalar features, which are insufficient for reflecting the deep-level rumor properties. To address the above challenge, we propose a novel model named graph attention capsule network on dynamic propagation structures (GACN) for rumor detection. Specifically, GACN consists of two components: a graph attention network enforced by capsule network that can encode static graphs into substructure classification capsules for mining the deep-level properties of rumor, and a dynamic network framework that can divide the rumor structure into multiple static graphs in chronological order for capturing the dynamic interactive features in the evolving process of the rumor propagation structure. Moreover, we use the capsule attention mechanism to combine the capsules generated from each substructure to focus more on informative substructures in rumor propagation. Extensive validation on two real-world datasets demonstrates the superiority of GACN over baselines.

Information dissemination occurs through the 'word of media' in the digital world. Fraudulent and deceitful content, such as misinformation, has detrimental effects on people. An implicit fact-based automated fact-checking technique comprising information retrieval, natural language processing, and machine learning techniques assist in assessing the credibility of content and detecting misinformation. Previous studies focused on linguistic and textual features and similarity measures-based approaches. However, these studies need to gain knowledge of facts, and similarity measures are less accurate when dealing with sparse or zero data. To fill these gaps, we propose a 'Content Similarity Measure (CSM)' algorithm that can perform automated fact-checking of URLs in the healthcare domain. Authors have introduced a novel set of content similarity, domain-specific, and sentiment polarity score features to achieve journalistic fact-checking. An extensive analysis of the proposed algorithm compared with standard similarity measures and machine learning classifiers showed that the 'content similarity score' feature outperformed other features with an accuracy of 88.26%. In the algorithmic approach, CSM showed improved accuracy of 91.06% compared to the Jaccard similarity measure with 74.26% accuracy. Another observation is that the algorithmic approach outperformed the feature-based method. To check the robustness of the algorithms, authors have tested the model on three state-of-the-art datasets, viz. CoAID, FakeHealth, and ReCOVery. With the algorithmic approach, CSM showed the highest accuracy of 87.30%, 89.30%, 85.26%, and 88.83% on CoAID, ReCOVery, FakeHealth (Story), and FakeHealth (Release) datasets, respectively. With a feature-based approach, the proposed CSM showed the highest accuracy of 85.93%, 87.97%, 83.92%, and 86.80%, respectively.

[This corrects the article DOI: 10.1007/s11227-021-04222-4.].

Harris hawks optimization (HHO) is a new meta-heuristic algorithm that builds a model by imitating the predation process of Harris hawks. In order to solve the problems of poor convergence speed caused by uniform choice position update formula in the exploration stage of basic HHO and falling into local optimization caused by insufficient population richness in the later stage of the algorithm, a Harris hawks optimization based on global cross-variation and tent mapping (CRTHHO) is proposed in this paper. Firstly, the tent mapping is introduced in the exploration stage to optimize random parameter q to speed up the convergence in the early stage. Secondly, the crossover mutation operator is introduced to cross and mutate the global optimal position in each iteration process. The greedy strategy is used to select, which prevents the algorithm from falling into local optimal because of skipping the optimal solution and improves the convergence accuracy of the algorithm. In order to investigate the performance of CRTHHO, experiments are carried out on ten benchmark functions and the CEC2017 test set. Experimental results show that the CRTHHO algorithm performs better than the HHO algorithm and is competitive with five advanced meta-heuristic algorithms.

Motion Estimation is one of the main tasks behind any video encoder. It is a computationally costly task; therefore, it is usually delegated to specific or reconfigurable hardware, such as FPGAs. Over the years, multiple FPGA implementations have been developed, mainly using hardware description languages such as Verilog or VHDL. Since programming using hardware description languages is a complex task, it is desirable to use higher-level languages to develop FPGA applications.The aim of this work is to evaluate OpenCL, in terms of expressiveness, as a tool for developing this kind of FPGA applications. To do so, we present and evaluate a parallel implementation of the Block Matching Motion Estimation process using OpenCL for Intel FPGAs, usable and tested on an Intel Stratix 10 FPGA. The implementation efficiently processes Full HD frames completely inside the FPGA. In this work, we show the resource utilization when synthesizing the code on an Intel Stratix 10 FPGA, as well as a performance comparison with multiple CPU implementations with varying levels of optimization and vectorization capabilities. We also compare the proposed OpenCL implementation, in terms of resource utilization and performance, with estimations obtained from an equivalent VHDL implementation.

Iron is one of the trace elements that plays a vital role in the human immune system, especially against variants of SARS-CoV-2 virus. Electrochemical methods are convenient for the detection due to the simplicity of instrumentation available for different analyses. The square wave voltammetry (SQWV) and differential pulse voltammetry (DPV) are useful electrochemical voltammetric techniques for diverse types of compounds such as heavy metals. The basic reason is the increased sensitivity by lowering the capacitive current. In this study, machine learning models were improved to classify concentrations of an analyte depending on the voltammograms obtained alone. SQWV and DPV were used to quantify the concentrations of ferrous ions (Fe${}^{+2}$) in potassium ferrocyanide (K${}_4$Fe(CN)${}_6$), validated by machine learning models for the data classifications. The greatest classifier algorithms models Backpropagation Neural Networks, Gaussian Naive Bayes, Logistic Regression, K-Nearest Neighbors Algorithm, K-Means clustering, and Random Forest were used as data classifiers, based on the data sets obtained from the measured chemical. Once competed to other algorithms models used previously for the data classification, ours get greater accuracy, maximum accuracy of 100% was obtained for each analyte in 25 s for the datasets.

Bidirectional generative adversarial networks (BiGANs) and cycle generative adversarial networks (CycleGANs) are two emerging machine learning models that, up to now, have been used as generative models, i.e., to generate output data sampled from a target probability distribution. However, these models are also equipped with encoding modules, which, after weakly supervised training, could be, in principle, exploited for the extraction of hidden features from the input data. At the present time, how these extracted features could be effectively exploited for classification tasks is still an unexplored field. Hence, motivated by this consideration, in this paper, we develop and numerically test the performance of a novel inference engine that relies on the exploitation of BiGAN and CycleGAN-learned hidden features for the detection of COVID-19 disease from other lung diseases in computer tomography (CT) scans. In this respect, the main contributions of the paper are twofold. First, we develop a kernel density estimation (KDE)-based inference method, which, in the training phase, leverages the hidden features extracted by BiGANs and CycleGANs for estimating the (a priori unknown) probability density function (PDF) of the CT scans of COVID-19 patients and, then, in the inference phase, uses it as a target COVID-PDF for the detection of COVID diseases. As a second major contribution, we numerically evaluate and compare the classification accuracies of the implemented BiGAN and CycleGAN models against the ones of some state-of-the-art methods, which rely on the unsupervised training of convolutional autoencoders (CAEs) for attaining feature extraction. The performance comparisons are carried out by considering a spectrum of different training loss functions and distance metrics. The obtained classification accuracies of the proposed CycleGAN-based (resp., BiGAN-based) models outperform the corresponding ones of the considered benchmark CAE-based models of about 16% (resp., 14%).