查看更多>>摘要：Automatic sound classification attracts increasing research attention owing to its vast applications, such as robot navigation, environmental sensing, musical instrument classification, medical diagnosis, and surveillance. In this research, we propose an ensemble convolutional bidirectional Long Short-Term Memory (CBiLSTM) network with optimal hyper-parameter selection for undertaking sound classification. We first transform each audio signal into a spectrogram representation using the Short-time Fourier transform (STFT). A Particle Swarm Optimization (PSO) variant is subsequently proposed to optimize the learning rate, weight decay, numbers of filters and hidden units in the convolutional and BiLSTM layers, respectively, in order to extract effective spatial–temporal characteristics from the spectrogram inputs. To tackle the issue of stagnation in optimization, the proposed algorithm incorporates local exploitation using secant and Newton–Raphson methods, promising leader generation using regular and irregular super-ellipse formulae, and three-dimensional spherical search coefficients. Moreover, it takes into account multiple fused elite signals in conjunction with numerical analysis based exploitation to balance between diversification and intensification. A variety of CBiLSTM networks with distinctive optimized settings are devised. An ensemble model is then constructed by incorporating a set of three yielded networks based on a majority voting scheme. Evaluated using several audio data sets, our ensemble CBiLSTM networks outperform those with default and optimal settings identified by other search methods, existing deep architectures and state-of-the-art related studies. In addition to sound classification tasks, the proposed PSO algorithm also outperforms a number of classical and advanced search methods for solving diverse unimodal and multimodal benchmark functions with statistical significance.
查看更多>>摘要：Mobile health (mHealth) technologies, such as symptom tracking apps, are crucial for coping with the global pandemic crisis by providing near real-time, in situ information for the medical and governmental response. However, in such a dynamic and diverse environment, methods are still needed to support public health decision-making. This paper uses the lens of strong structuration theory to investigate networks of COVID-19 symptoms in the Belfast metropolitan area. A self-supervised machine learning method measuring information entropy was applied to the Northern Ireland COVIDCare app. The findings reveal: (1) relevant stratifications of disease symptoms, (2) particularities in health-wealth networks, and (3) the predictive potential of artificial intelligence to extract entangled knowledge from data in COVID-related apps. The proposed method proved to be effective for near real-time in-situ analysis of COVID-19 progression and to focus and complement public health decisions. Our contribution is relevant to an understanding of SARS-COV-2 symptom entanglements in localised environments. It can assist decision-makers in designing both reactive and proactive health measures that should be personalised to the heterogeneous needs of different populations. Moreover, near real-time assessment of pandemic symptoms using digital technologies will be critical to create early warning systems of emerging SARS-CoV-2 strains and predict the need for healthcare resources.
查看更多>>摘要：Conductive particle inspection is the crucial procedure in the circuit detection process of liquid crystal modules since only validly deformed particles have conductive effects in the circuit. The main task of inspection is to accurately locate and count the validly deformed particles. However, due to various difficulties such as the variety of deformed particles, the aggregation between valid and invalid particles, the different sizes to overlap of deformed particles, and uneven illumination, etc., robust real-time detection of valid particles becomes a challenging problem. In this paper, a novel Multi-Scale Deep Adversarial Network (MSDA-Net) composed of Generator and Discriminator is proposed for detecting valid particles in a liquid crystal module. Firstly, a compact Multi-branch network is adopted as Generator for detecting valid particles, which achieves well detection accuracy with less time by utilizing the lightweight structure and the Coarse-to-Fine multi-scale feature extraction form. Secondly, the designed Logical Focusing Attention module (LFA) is applied in Generator to enhance the center location precision of valid particles. Furthermore, the Multi-branch based Deep Adversarial Strategy is proposed to distinguish deeply the similarity of visual characteristics between valid and invalid particles, which achieves higher detection accuracy with fewer fake particles. The experiments on the real datasets demonstrate the effectiveness of the proposed methods for the real-time detection of valid particles compared to the state-of-the-art methods.
查看更多>>摘要：The distribution of the rotor slot wedge eddy current loss in large nuclear power turbo generators is complex and is influenced by many factors. Excessive eddy current loss leads to severe rotor heating, potentially leading to thermal accidents; therefore, the design precision of large generators must be improved. In this paper, a Fuzzy C-Means Deep Gaussian Process Regression (FCM-DGPR) method is proposed to predict the eddy current loss of a large generator in order to solve the problem of the insufficient accuracy of deep Gaussian process regression (DGPR) with increasing number of the data samples. First, the original dataset is obtained by the finite element method (FEM) and then normalized to construct the samples of the eddy current loss of a large nuclear power generator. Second, the training set is automatically clustered into different subsets by the fuzzy c-means algorithm, and each subset is used to train the DGPR model to obtain different sub models. The membership degree of each data point in the test set is calculated and used to evaluate the sub model of the data. Then, the sub model is used to predict the eddy current loss. Finally, the result is obtained by concatenating the results of each sub model. The results show that the goodness of fit (R2) is 0.9809, the root mean square error (RMSE) is 0.0271, the prediction error is small, and the model exhibits good prediction performance. Further experimental results show that the FCM-DGPR method is superior to the existing DGPR models and other models and is more suitable for predicting the eddy current loss of large generators.
查看更多>>摘要：Sustainability has become increasingly important over the last three decades and has proven to be a key enabler for constructing resilient supply chains. Customers who want their products to be authenticated for sustainability put pressure on Original Equipment Manufacturers and suppliers to become more sustainable on a global scale. Moreover, social sustainability issues have become more challenging to address, and a growing number of stakeholders put emphasis on societal concerns. To this end, decision-makers are becoming increasingly interested in applying disruptive technologies to address societal, environmental, and economic concerns and accomplish sustainability goals. Researchers argue that disruptive technologies such as blockchain may be implemented to assist supply chains towards building sustainability. However, our literature analysis concluded that existing research has not quantitatively examined the critical functions of sustainable supply chain (SSC) for blockchain applicability using a decision framework. Therefore, this research, through Fuzzy SWARA-COPRAS-EDAS and COPELAND-based framework, is aimed at investigating the most feasible functions of a SSC for potential blockchain implementations. Using this framework, the critical functions of a SSC were ranked against the benefits of blockchain. The findings of this study implied that while sourcing, delivery, transformation and product recovery proved to be the most appropriate functions of SSCs for blockchain applications, customers and product use was the least feasible one. This study aids decision-makers in gaining a more thorough understanding of where in a SSC blockchain may create additional value.
查看更多>>摘要：Finding a feasible solution set for optimization problems in conflict with objective functions poses significant challenges. Moreover, in such problems, the level of complexity may increase depending on the geometry of the objective and decision spaces. The most effective methods in solving multi-objective problems having high levels of complexity are search algorithms using the Pareto-based archiving approach. Recently, the crowding distance approach has been used to improve the performance of the Pareto-based archiving method. This article presents research conducted on the development of a method that can find the optimum solution set for a multi-objective optimal power flow (MOOPF) problem whose objective functions are in conflict. For this purpose, a powerful and effective method was developed using the Pareto archiving approach based on crowding distance. The performance of the developed method was tested on twenty-four benchmark problems of different types and difficulty levels and compared with competing algorithms. The data obtained from the experimental trials and four different performance metrics were analyzed using statistical test methods. Analysis results showed that the proposed method yielded a competitive performance on different types of multi-objective optimization problems and was able to find the best solutions in the literature for the real-world MOOPF problem.
查看更多>>摘要：It is very important to obtain spatio-temporal information in video deblurring based on deep learning. The existing methods usually jointly learn the spatio-temporal information of blurred videos through single-stream networks, which inevitably limit spatio-temporal information learning and video deblurring performance of networks. Therefore, we propose a dual-stream spatio-temporal decoupling network (STDN), which can learn the spatio-temporal information of blurred videos more flexibly and efficiently with the decoupled temporal stream and spatial stream, for solving this problem. Firstly, in the temporal stream of STDN, we propose a video deblurring pipeline, that is motion compensation plus 3D CNNs, for solving the drawback of 3D CNNs that its receptive field cannot effectively cover the same but misplaced contents of different frames. Thus, the temporal stream can aggregate temporal information of frame sequences and handle inter-frame misalignments more effectively. Specifically, we design a novel deformable convolution compensation module (DCCM) to achieve motion compensation of this pipeline more accurately. Then, we develop a 3DConv module optimized by the designed temporal, spatial, and channel decoupling attention block, named the CTS, to achieve 3D CNNs of this pipeline. Secondly, we design a spatial stream in which two types of wide-activation residual modules are stacked, for learning more spatial features of the central frame to supplement the temporal stream. Finally, extensive experiments on the baseline datasets demonstrate that the proposed STDN has better performance than the latest methods. Remarkably, using the proposed temporal stream alone already can achieve competitive video deblurring performance than the existing methods.
查看更多>>摘要：Most of the scientific and engineering problems are defined as constrained optimization functions. It can be very difficult due to their complex structures. Artificial Bee Colony Algorithm (ABC) is a remarkable metaheuristic developed for global optimization problems. However, due to the inadequacy of ABC's search capability, it cannot handle constraint optimization problems very well. In this study, an ABC variant adapted for solving constrained optimization problems called Artificial Bee Colony Algorithm with Distant Savants (ABCDS) is proposed to overcome this deficiency. ABCDS is based on a new and adaptable search equation that enables learning with savants that are at a certain distance from each other. Also, the algorithm is hybridized with competitive local search mechanism. To test the performance of ABCDS, benchmark set for Constrained Real-Parameter Optimization defined in CEC 2017 conference (CEC2017COP) and some of the problems in the benchmark set on real-world problems defined in CEC 2020 conference (CEC2020) are used. The results obtained by the algorithm are compared with recent ABC algorithms and some state-of-the-art algorithms. According to the experimental results, ABCDS is better and competitive than the compared algorithms.
查看更多>>摘要：The automated analysis of eye fundus images is crucial towards facilitating the screening and early diagnosis of glaucoma. Nowadays, there are two common alternatives for the diagnosis of this disease using deep neural networks. One is the segmentation of the optic disc and cup followed by the morphological analysis of these structures. The other is to directly address the diagnosis as an image classification task. The segmentation approach presents the advantage of using pixel-level labels with precise morphological information for training. However, while this detailed training feedback is not available for the classification approach, in this case the image-level labels may allow the discovery of additional non-morphological cues that are also relevant for the diagnosis. In this work, we propose a novel multi-task approach for the simultaneous classification of glaucoma and segmentation of the optic disc and cup. This approach can improve the overall performance by taking advantage of both pixel-level and image-level labels during the network training. Additionally, the segmentation maps that are predicted together with the diagnosis allow the extraction of relevant biomarkers such as the cup-to-disc ratio. The proposed methodology presents two relevant technical novelties. First, a network architecture for simultaneous segmentation and classification that increases the number of shared parameters between both tasks. Second, a multi-adaptive optimization strategy that ensures that both tasks contribute similarly to the parameter updates during training, avoiding the use of loss weighting hyperparameters. To validate our proposal, an exhaustive experimentation was performed on the public REFUGE and DRISHTI-GS datasets. The results show that our proposal outperforms comparable multi-task baselines and is highly competitive with existing state-of-the-art approaches. Additionally, the provided ablation study shows that both the network architecture and the optimization approach are independently advantageous for multi-task learning.
查看更多>>摘要：The advantages of blockchain virtual currency are convenient circulation, low transaction costs, and decentralized power. At present, more and more investors have focused their investment in the blockchain virtual currency. Transaction data of blockchain virtual currency belongs to the financial time series, which is noisy and random, bringing challenges to the prediction of transaction trends. The improved deep belief network (IDBN) model and Echo state network (ESN) are constructed based on deep belief network (DBN) model to explore the long short-term memory (LSTM) model and the transaction prediction of blockchain virtual currency under the DBN and to improve the transaction prediction accuracy of the blockchain virtual currency. In addition, the parameters of IDBN model were optimized using particle swarm optimization (PSO) algorithm, which are verified with the transaction data of stocks and blockchain virtual currencies (Bitcoin, Bitcoin Cash, and Ethereum), and compared with other cash algorithms for analysis. The results show that the PSO-IDBN-based time series prediction model proposed in this study can be applied to predicting the high-latitude and high-complexity data, showing superior performance compared to the traditional time series prediction and other deep learning prediction methods.
NSTL
Elsevier