Search

Convolutional Neural Networks (CNNs) are a kind of artificial neural network designed to extract features and find out patterns for tasks such as segmentation, recognizing objects, and drawing up classification. Within a CNNs architecture, pooling operations are used until the number of parameters and the computational complexity are reduced. Numerous papers have focused on investigating the impact of pooling on the performance of Convolutional Neural Networks (CNNs), leading to the development of various pooling models. Recently, a fuzzy pooling operation based on type-1 fuzzy sets was introduced to cope with the local imprecision of the feature maps. However, in fuzzy set theory, it is not always accurate to assume that the degree of non-membership of an element in a fuzzy set is simply the complement of the degree of membership. This is due to the potential existence of a hesitation degree, which implies a certain level of uncertainty. To overcome this limitation, intuitionistic fuzzy sets (IFS) were introduced to incorporate the concept of a degree of hesitation. In this paper, we introduce a novel pooling operation based on intuitionistic fuzzy sets to incorporate the degree of hesitation heretofore neglected by a fuzzy pooling operation based on classical fuzzy sets, and we investigate its performance in the context of image classification. Intuitionistic pooling is performed in four steps: bifuzzification (by the transformation of data through the use of membership and non-membership maps), first aggregation (through the transformation of the IFS into a standard fuzzy set, second aggregation (through the transformation and use of a sum operator), and the defuzzification of feature map neighborhoods by using a max operator. IFS pooling is used for the construction of an intuitionistic pooling layer that can be applied as a drop-in replacement for the current, fuzzy (type-1) and crisp, pooling layers of CNN architectures. Various experiments involving multiple datasets demonstrate that an IFS-based pooling can enhance the classification performance of a CNN. A benchmarking study reveals that this significantly outperforms even the most recent pooling models, especially in stochastic environments. [ABSTRACT FROM AUTHOR]

Deep learning, a crucial technique for achieving artificial intelligence (AI), has been successfully applied in many fields. The gradual application of the latest architectures of deep learning in the field of time series forecasting (TSF), such as Transformers, has shown excellent performance and results compared to traditional statistical methods. These applications are widely present in academia and in our daily lives, covering many areas including forecasting electricity consumption in power systems, meteorological rainfall, traffic flow, quantitative trading, risk control in finance, sales operations and price predictions for commercial companies, and pandemic prediction in the medical field. Deep learning-based TSF tasks stand out as one of the most valuable AI scenarios for research, playing an important role in explaining complex real-world phenomena. However, deep learning models still face challenges: they need to deal with the challenge of large-scale data in the information age, achieve longer forecasting ranges, reduce excessively high computational complexity, etc. Therefore, novel methods and more effective solutions are essential. In this paper, we review the latest developments in deep learning for TSF. We begin by introducing the recent development trends in the field of TSF and then propose a new taxonomy from the perspective of deep neural network models, comprehensively covering articles published over the past five years. We also organize commonly used experimental evaluation metrics and datasets. Finally, we point out current issues with the existing solutions and suggest promising future directions in the field of deep learning combined with TSF. This paper is the most comprehensive review related to TSF in recent years and will provide a detailed index for researchers in this field and those who are just starting out. [ABSTRACT FROM AUTHOR]

Pattern recognition systems always misclassify anomalies, which can be dangerous for uninformed users. Therefore, anomalies must be filtered out from each classification. The main challenge for the anomaly filter design is the huge number of possible anomaly samples compared with the number of samples in the training set. Tailoring the filter for the given classifier is just the first step in this reduction. Paper tests the hypothesis that the filter trained in avoiding "near" anomalies will also refuse the "far" anomalies, and the anomaly detector is then just a classifier distinguishing between "far real" and "near anomaly" samples. As a "far real" samples generator was used, a Generative Adversarial Network (GAN) fake generator that transforms normally distributed random seeds into fakes similar to the training samples. The paper proves the assumption that seeds unused in fake training will generate anomalies. These seeds are distinguished according to their Chebyshev norms. While the fakes have seeds within the hypersphere with a given radius, the near anomalies have seeds within the sphere near cover. Experiments with various anomaly test sets have shown that GAN-based anomaly detectors create a reliable anti-anomaly shield using the abovementioned assumptions. The proposed anomaly detector is tailored to the given classifier, but its limitation is due to the need for the availability of the database on which the classifier was trained. [ABSTRACT FROM AUTHOR]

Recent research has revealed that subtle imperceptible perturbations can deceive well-trained neural network models, leading to inaccurate outcomes. These instances, known as adversarial examples, pose significant threats to the secure application of machine learning techniques in safety-critical systems. In this paper, we delve into the study of one-pixel attacks in deep neural networks, recently reported as a kind of adversarial examples. To identify such one-pixel attacks, most existing methodologies rely on the differential evolution method, which utilizes random selection from the current population to escape local optima. However, the differential evolution technique might waste search time and overlook good solutions if the number of iterations is insufficient. Hence, in this paper, we propose a gradient ascent with momentum approach to efficiently discover good solutions for the one-pixel attack problem. As our method takes a more direct route to the goal compared to existing methods relying on blind random walks, it can effectively identify one-pixel attacks. Our experiments conducted on popular CNNs demonstrate that, in comparison with existing methodologies, our technique can detect one-pixel attacks significantly faster. [ABSTRACT FROM AUTHOR]

This correction notice provides updates and corrections to a research paper on deep machine learning model-based cyber-attacks detection in smart power systems. The paper proposes a deep learning-based attack detection model that addresses system disturbances caused by natural events and cyber-attacks. It discusses the use of principal component analysis (PCA) for feature selection and the potential of deep learning-based and decision tree classifiers for detecting cyber-attacks in intelligent energy grids. The notice also includes updates to references, clarifications on the experimental data set, and feature selection based on PCA. The study evaluates the effectiveness of the proposed model using various assessment indexes and compares it to conventional methods, concluding that the proposed model is effective in detecting cyber-attacks and demonstrates good performance in detecting destructive attacks. [Extracted from the article]

This paper studies an artificial neural network (ANN) for multi-asset European options. Firstly, a simple three-layer ANN-3 is established with undetermined weights and bias. Secondly, the time–space discrete PDE of the multi-asset option is given and the corresponding discrete data are fed into the ANN-3. Then, using least squares error as the objective function, the weights and bias of ANN-3 are trained well. Numerical examples are carried out to confirm the stability, accuracy and efficiency. Experiments show the ANN's relative error is about  0.8 % . This method can be extended into multi-layer ANN- q (q > 3)  and extended into American options. [ABSTRACT FROM AUTHOR]

With the rapid development of sensor technology, the anomaly detection of multi-source time series data becomes more and more important. Traditional anomaly detection methods deal with the temporal and spatial information in the data independently, and fail to make full use of the potential of spatio-temporal information. To address this issue, this paper proposes a novel integration method that combines sensor embeddings and temporal representation networks, effectively exploiting spatio-temporal dynamics. In addition, the graph neural network is introduced to skillfully simulate the complexity of multi-source heterogeneous data. By applying a dual loss function—consisting of a reconstruction loss and a prediction loss—we further improve the accuracy of anomaly detection. This strategy not only promotes the ability to learn normal behavior patterns from historical data, but also significantly improves the predictive ability of the model, making anomaly detection more accurate. Experimental results on four multi-source sensor datasets show that our proposed method performs better than the existing models. In addition, our approach enhances the ability to interpret anomaly detection by analyzing the sensors associated with the detected anomalies. [ABSTRACT FROM AUTHOR]

Wireless traffic prediction is essential to developing intelligent communication networks that facilitate efficient resource allocation. Along this line, decentralized wireless traffic prediction under the paradigm of federated learning is becoming increasingly significant. Compared to traditional centralized learning, federated learning satisfies network operators' requirements for sensitive data protection and reduces the consumption of network resources. In this paper, we propose a novel communication-efficient federated learning framework, named FedCE, by developing a gradient compression scheme and an adaptive aggregation strategy for wireless traffic prediction. FedCE achieves gradient compression through top-K sparsification and can largely relieve the communication burdens between local clients and the central server, making it communication-efficient. An adaptive aggregation strategy is designed by quantifying the different contributions of local models to the global model, making FedCE aware of spatial dependencies among various local clients. We validate the effectiveness of FedCE on two real-world datasets. The results demonstrate that FedCE can improve prediction accuracy by approximately 27% with only 20% of communications in the baseline method. [ABSTRACT FROM AUTHOR]

Addressing real-time optimization problems becomes increasingly challenging as their complexity continues to escalate over time. So bio-optimization algorithms (BoAs) come into the picture to solve such problems due to their global search capability, adaptability, versatility, parallelism, and robustness. This article aims to perform hyperparameter tuning of machine learning (ML) models by integrating them with BoAs. Aiming to maximize the accuracy of the hybrid bio-optimized defect prediction (HBoDP) model, this research paper develops four novel hybrid BoAs named the gravitational force Lévy flight grasshopper optimization algorithm (GFLFGOA), the gravitational force Lévy flight grasshopper optimization algorithm–sparrow search algorithm (GFLFGOA-SSA), the gravitational force grasshopper optimization algorithm–sparrow search algorithm (GFGOA-SSA), and the Lévy flight grasshopper optimization algorithm–sparrow search algorithm (LFGOA-SSA). These aforementioned algorithms are proposed by integrating the good exploration capacity of the SSA with the faster convergence of the LFGOA and GFGOA. The performances of the GFLFGOA, GFLFGOA-SSA, GFGOA-SSA, and LFGOA-SSA are verified by conducting two different experiments. Firstly, the experimentation was conducted on nine benchmark functions (BFs) to assess the mean, standard deviation (SD), and convergence rate. The second experiment focuses on boosting the accuracy of the HBoDP model through the fine-tuning of the hyperparameters in the artificial neural network (ANN) and XGBOOST (XGB) models. To justify the effectiveness and performance of these hybrid novel algorithms, we compared them with four base algorithms, namely the grasshopper optimization algorithm (GOA), the sparrow search algorithm (SSA), the gravitational force grasshopper optimization algorithm (GFGOA), and the Lévy flight grasshopper optimization algorithm (LFGOA). Our findings illuminate the effectiveness of this hybrid approach in enhancing the convergence rate and accuracy. The experimental results show a faster convergence rate for BFs and improvements in software defect prediction accuracy for the NASA defect datasets by comparing them with some baseline methods. [ABSTRACT FROM AUTHOR]

Financial time series data are characterized by non-linearity, non-stationarity, and stochastic complexity, so predicting such data presents a significant challenge. This paper proposes a novel hybrid model for financial forecasting based on CEEMDAN-SE and ARIMA- CNN-LSTM. With the help of the CEEMDAN-SE method, the original data are decomposed into several IMFs and reconstructed via sample entropy into a lower-complexity stationary high-frequency component and a low-frequency component. The high-frequency component is predicted by the ARIMA statistical forecasting model, while the low-frequency component is predicted by a neural network model combining CNN and LSTM. Compared to some classical prediction models, our algorithm exhibits superior performance in terms of three evaluation indexes, namely, RMSE, MAE, and MAPE, effectively enhancing model accuracy while reducing computational overhead. [ABSTRACT FROM AUTHOR]

Deep neural networks (DNNs) have achieved breakthrough progress in various fields, largely owing to the support of large-scale datasets with manually annotated labels. However, obtaining such datasets is costly and time-consuming, making high-quality annotation a challenging task. In this work, we propose an improved noisy sample selection method, termed "sample selection framework", based on a mixup loss and recalibration strategy (SMR). This framework enhances the robustness and generalization abilities of models. First, we introduce a robust mixup loss function to pre-train two models with identical structures separately. This approach avoids additional hyperparameter adjustments and reduces the need for prior knowledge of noise types. Additionally, we use a Gaussian Mixture Model (GMM) to divide the entire training set into labeled and unlabeled subsets, followed by robust training using semi-supervised learning (SSL) techniques. Furthermore, we propose a recalibration strategy based on cross-entropy (CE) loss to prevent the models from converging to local optima during the SSL process, thus further improving performance. Ablation experiments on CIFAR-10 with 50% symmetric noise and 40% asymmetric noise demonstrate that the two modules introduced in this paper improve the accuracy of the baseline (i.e., DivideMix) by 1.5% and 0.5%, respectively. Moreover, the experimental results on multiple benchmark datasets demonstrate that our proposed method effectively mitigates the impact of noisy labels and significantly enhances the performance of DNNs on noisy datasets. For instance, on the WebVision dataset, our method improves the top-1 accuracy by 0.7% and 2.4% compared to the baseline method. [ABSTRACT FROM AUTHOR]

Breast cancer is one of the most common causes of death in women. Early signs of breast cancer can be an abnormality depicted on breast images like breast ultrasonography. Unfortunately, ultrasound images contain a lot of noise, which greatly increases the difficulty for doctors to interpret them. In recent years, computer-aided diagnosis (CAD) has been widely used in medical images, reducing the workload of doctors and the probability of misdiagnosis. However, it still faces the following challenges in clinical practice: one is the lack of interpretability, and another is that the accuracy is not high enough. In this paper, we propose a classification model of breast ultrasound images that leverages tumor boundaries as prior knowledge and strengthens the model to guide classification. Furthermore, we employ the advantages of convolutional neural network (CNN) to extract local features and Transformer to extract global features to achieve information balance and complementarity between the two neural network models which increase the recognition performance of the model. Additionally, an explanation method is used to generate visual results, thereby improving the poor interpretability of deep learning models. Finally, we evaluate the model on the BUSI dataset and compare it with other CNN and Transformer models. Experimental results show that the proposed model obtains an accuracy of 0.9870 and an F1 score of 0.9872, achieving state-of-the-art performance. [ABSTRACT FROM AUTHOR]

In Structural Health Monitoring (SHM) of bridges, accurately assessing damage is critical to maintaining the safety and integrity of a structure. One of the primary challenges in damage assessment is the precise localization and quantification of defects, which is essential for making timely maintenance decisions and reducing the risk of structural failures. This paper introduces a novel damage detection method for SHM of a truss bridge by coupling a Deep Neural Network (DNN) model with an evolved Artificial Rabbit Optimization (EVARO) algorithm. The integration of DNN with the stochastic search capability of the EVARO algorithm helps to avoid local minima, thereby ensuring more accurate and reliable results. Additionally, the optimization algorithm's effectiveness is further enhanced by incorporating evolving predator features and the Cauchy motion search mechanism. The proposed method is first validated using various data benchmark problems, demonstrating its effectiveness compared to other well-known algorithms. Secondly, a case study involving the Chuong Duong truss bridge under different simulated damage scenarios further confirms the superiority of the proposed method in both localizing and quantifying damages. [ABSTRACT FROM AUTHOR]

Three-wheeled omnidirectional mobile robots (TOMRs) are widely used to accomplish precise transportation tasks in narrow environments owing to their stability, flexible operation, and heavy loads. However, these robots are susceptible to slippage. For wheeled robots, almost all faults and slippage will directly affect the power consumption. Thus, using the energy consumption model data and encoder data in the healthy condition as a reference to diagnose robot slippage and other system faults is the main issue considered in this paper. We constructed an energy model for the TOMR and analyzed the factors that affect the power consumption in detail, such as the position of the gravity center. The study primarily focuses on the characteristic relationship between power consumption and speed when the robot experiences slippage or common faults, including control system faults. Finally, we present the use of a table-based artificial neural network (ANN) to indicate the type of fault by comparing the modeled data with the measured data. The experiments proved that the method is accurate and effective for diagnosing faults in TOMRs. [ABSTRACT FROM AUTHOR]

Patient-specific 3D models of the human mandible are finding increasing utility in medical fields such as oral and maxillofacial surgery, orthodontics, dentistry, and forensic sciences. The efficient creation of personalized 3D bone models poses a key challenge in these applications. Existing solutions often rely on 3D statistical models of human bone, offering advantages in rapid bone geometry adaptation and flexibility by capturing a range of anatomical variations, but also a disadvantage in terms of reduced precision in representing specific shapes. Considering this, the proposed parametric model allows for precise manipulation using morphometric parameters acquired from medical images. This paper highlights the significance of employing the parametric model in the creation of a personalized bone model, exemplified through a case study targeting mandibular prognathism reconstruction. A personalized model is described as 3D point cloud determined through the utilization of series of parametric functions, determined by the application of geometrical morphometrics, morphology properties, and artificial neural networks in the input dataset of human mandible samples. With 95.05% of the personalized model's surface area displaying deviations within −1.00–1.00 mm relative to the input polygonal model, and a maximum deviation of 2.52 mm, this research accentuates the benefits of the parametric approach, particularly in the preoperative planning of mandibular deformity surgeries. [ABSTRACT FROM AUTHOR]

Featured Application: The Deep Neural Network model employed in this work will help build a system for the pre-diagnosis of retinopathies that can lead to blindness. The main intention of this type of system is to support the work performed by ophthalmology specialists. Computer-based pre-diagnosis of diseases through medical imaging is a task worked on for many years. The so-called fundus images stand out since they do not have uniform illumination and are highly sensitive to noise. One of the diseases that can be pre-diagnosed through fundus images is age-related macular degeneration, which initially manifests as the appearance of lesions called drusen. Several ways of pre-diagnosing macular degeneration have been proposed, methods based entirely on the segmentation of drusen with prior image processing have been designed and applied, and methods based on image pre-processing and subsequent conversion to feature vectors, or patterns, to be classified by a Machine-Learning model have also been developed. Finally, in recent years, the use of Deep-Learning models, particularly Convolutional Networks, has been proposed and used in classification problems where the data are only images. The latter has allowed the so-called transfer learning, which consists of using the learning achieved in the solution of one problem to solve another. In this paper, we propose the use of transfer learning through the Xception Deep Convolutional Neural Network to detect age-related macular degeneration in fundus images. The performance of the Xception model was compared against six other state-of-the-art models with a dataset created from images available in public and private datasets, which were divided into training/validation and test; with the training/validation set, the training was made using 10-fold cross-validation. The results show that the Xception neural network obtained a validation accuracy that surpasses other models, such as the VGG-16 or VGG-19 networks, and had an accuracy higher than 80% in the test set. We consider that the contributions of this work include the use of a Convolutional Neural Network model for the detection of age-related macular degeneration through the classification of fundus images in those affected by AMD (drusen) and the images of healthy patients. The performance of this model is compared against other methods featured in the state-of-the-art approaches, and the best model is tested on a test set outside the training and validation set. [ABSTRACT FROM AUTHOR]

Interior permanent magnet synchronous machines (IPMSMs) driven with a square-wave control (i.e., six-step, block, or 120° control), known commonly as brushless direct current (BLDC) drives, are used widely due to their high power density and control simplicity. The advance firing (AF) angle is employed to achieve improved operation characteristics of the drive. The AF angle is, in general, applied to compensate for the commutation effects. In the case of an IPMSM, the AF angle can also be adjusted to exploit reluctance torque. In this paper, a detailed study was performed to understand its effect on the drive's performance in regard to reluctance torque. Furthermore, a multi-objective optimization of the machine's cross-section using neural network models was conducted to enhance performance at a constant AF angle. The reference and improved machine designs were evaluated in a system-level simulation, where the impact was considered of the commutation of currents. A significant improvement in the machine performance was achieved after optimizing the geometry and implementing a fixed AF angle of 10°. [ABSTRACT FROM AUTHOR]

In this paper, the fixed-time and preassigned-time synchronization issues of fully quaternion-valued fuzzy memristive neural networks are studied based on the dynamic event-triggered control mechanism. Initially, the fuzzy rules are defined within the quaternion domain and the relevant properties are established through rigorous analysis. Subsequently, to conserve resources and enhance the efficiency of the controller, a kind of dynamic event-triggered control mechanism is introduced for the fuzzy memristive neural networks. Based on the non-separation analysis, fixed-time and preassigned-time synchronization criteria are presented and the Zeno phenomenon under the event-triggered mechanism is excluded successfully. Finally, the effectiveness of the theoretical results is verified through numerical simulations. [ABSTRACT FROM AUTHOR]

Stagnation at local optima represents a significant challenge in bio-inspired optimization algorithms, often leading to suboptimal solutions. This paper addresses this issue by proposing a hybrid model that combines the Orca predator algorithm with deep Q-learning. The Orca predator algorithm is an optimization technique that mimics the hunting behavior of orcas. It solves complex optimization problems by exploring and exploiting search spaces efficiently. Deep Q-learning is a reinforcement learning technique that combines Q-learning with deep neural networks. This integration aims to turn the stagnation problem into an opportunity for more focused and effective exploitation, enhancing the optimization technique's performance and accuracy. The proposed hybrid model leverages the biomimetic strengths of the Orca predator algorithm to identify promising regions nearby in the search space, complemented by the fine-tuning capabilities of deep Q-learning to navigate these areas precisely. The practical application of this approach is evaluated using the high-dimensional Heartbeat Categorization Dataset, focusing on the feature selection problem. This dataset, comprising complex electrocardiogram signals, provided a robust platform for testing the feature selection capabilities of our hybrid model. Our experimental results are encouraging, showcasing the hybrid strategy's capability to identify relevant features without significantly compromising the performance metrics of machine learning models. This analysis was performed by comparing the improved method of the Orca predator algorithm against its native version and a set of state-of-the-art algorithms. [ABSTRACT FROM AUTHOR]

This paper discusses a type of mixed-delay quaternion-valued neural networks (QVNNs) under impulsive and stochastic disturbances. The considered QVNNs model are treated as a whole, rather than as complex-valued neural networks (NNs) or four real-valued NNs. Using the vector Lyapunov function method, some criteria are provided for securing the mean-square exponential stability of the mixed-delay QVNNs under impulsive and stochastic disturbances. Furthermore, a type of chaotic QVNNs under stochastic and impulsive disturbances is considered using a previously established stability analysis method. After the completion of designing the linear feedback control law, some sufficient conditions are obtained using the vector Lyapunov function method for determining the mean-square exponential synchronization of drive–response systems. Finally, two examples are provided to demonstrate the correctness and feasibility of the main findings and one example is provided to validate the use of QVNNs for image associative memory. [ABSTRACT FROM AUTHOR]

The existence of chaos is particularly relevant, as the identification of a chaotic behavior in a time series could lead to reliable short-term forecasting. This paper evaluates the existence of nonlinearity and chaos in the underlying process of the spot prices of the Spanish electricity market. To this end, we used daily data spanning from 1 January 2013, to 31 March 2021 and we applied a comprehensive framework that encompassed a wide range of techniques. Nonlinearity was analyzed using the BDS method, while the existence of a chaotic structure was studied through Lyapunov exponents, recurrence plots, and quantitative recurrence analysis. While nonlinearity was detected in the underlying process, conclusive evidence supporting chaos was not found. In addition, the generalized autoregressive conditional heteroscedastic (GARCH) model accounts for part of the nonlinear structure that is unveiled in the electricity market. These findings hold substantial value for electricity market forecasters, traders, producers, and market regulators. [ABSTRACT FROM AUTHOR]

The squeeze film damper (SFD) is proven to be highly effective in mitigating rotor vibration as it traverses the critical speed, thus making it extensively utilized in the aeroengine domain. In this paper, we investigate the stiffness and damping of SFD using the Reynolds equation and neural network models. Our specific focus includes examining the structural and operating parameters of SFDs, such as clearance, feed pressure of oil, rotor whirl, and rotational speed. Firstly, the pressure distribution analytical model of the oil film inside the SFD based on the hydrodynamic lubrication theory is established, as described by the Reynolds equation. It obtained oil film forces, pressure, stiffness, and damping values under various sets of structural, lubrication, and operating parameters, including length, clearance, boundary pressure at both sides, rotational speed, and whirling motion, by applying difference computations to the Reynolds equation. Secondly, according to the significant analyses of the obtained oil film stiffness and damping, the following three parameters of the most significance are found: clearance, rotational speed, and rotor whirl. Furthermore, neural network models, including GA-BP and decision tree models, are established based on the obtained results of difference computation. The numerical simulation and calculation of these models are then applied to show their validity with all given parameters and the three significant parameters separately as two sets of model input. Regardless of either set of model inputs, these established neural network models are capable of predicting the nonlinear stiffness and damping of the oil film inside an SFD. These sensitive parameters merely require measurement, followed by the utilization of a neural network to predict stiffness and damping instead of the Reynolds equation. This process serves structural enhancement, facilitates parameter optimization in SFDs, and provides crucial support for refining the design parameters of SFDs. [ABSTRACT FROM AUTHOR]

Lightweight and adaptive adjustment are key research directions for deep neural networks (DNNs). In coal industry mining, frequent changes in raw coal sources and production batches can cause uneven distribution of appearance features, leading to concept drift problems. The network architecture and parameters should be adjusted frequently to avoid a decline in model accuracy. This poses a significant challenge for those without specialist expertise. Although the Neural Architecture Search (NAS) has a strong ability to automatically generate networks, enabling the automatic design of highly accurate networks, it often comes with complex internal topological connections. These redundant architectures do not always effectively improve network performance, especially in resource-constrained environments, where their computational efficiency is significantly reduced. In this paper, we propose a method called Topology Complexity Neural Architecture Search (TCNAS). TCNAS proposes a new method for evaluating the topological complexity of neural networks and uses both topological complexity and accuracy to guide the search, effectively obtaining lightweight and efficient networks. TCNAS employs an adaptive shrinking search space optimization method, which gradually eliminates poorly performing cells to reduce the search space, thereby improving search efficiency and solving the problem of space explosion. In the classification experiments of coal and gangue, the optimal network designed by TCNAS has an accuracy of 83.3%. And its structure is much simpler, which is about 1/53 of the parameters of the network dedicated to coal and gangue recognition. Experiments have shown that TCNAS is able to generate networks that are both efficient and simple for resource-constrained industrial applications. [ABSTRACT FROM AUTHOR]

Many dynamic random-access memory (DRAM) manufacturing companies encounter significant resistance value deviations during the PVD sputtering process for manufacturing Ti thin films. These resistance values are influenced by the thickness of the thin films. Current mitigation strategies focus on adjusting film thickness to reduce resistance deviations, but this approach affects product structure profile and performance. Additionally, varying Ti thin film thicknesses across different product structures increase manufacturing complexity. This study aims to minimize resistance value deviations across multiple film thicknesses with minimal resource utilization. To achieve this goal, we propose the TSDTM-ANN-GA framework, which integrates the two-stage dynamic Taguchi method (TSDTM), artificial neural networks (ANN), and genetic algorithms (GA). The proposed framework requires significantly fewer experimental resources than traditional full factorial design and grid search method, making it suitable for resource-constrained and low-power computing environments. Our TSDTM-ANN-GA framework successfully identified an optimal production condition configuration for five different Ti thin film thicknesses: Degas temperature = 245 °C, Ar flow = 55 sccm, DC power = 5911 W, and DC power ramp rate = 4009 W/s. The results indicate that the deviation between the resistance values and their design values for the five Ti thin film thicknesses decreased by 86.8%, 94.1%, 95.9%, 98.2%, and 98.8%, respectively. The proposed method effectively reduced resistance deviations for the five Ti thin film thicknesses and simplified manufacturing management, allowing the required design values to be achieved under the same manufacturing conditions. This framework can efficiently operate on resource-limited and low-power computers, achieving the goal of real-time dynamic production parameter adjustments and enabling DRAM manufacturing companies to improve product quality promptly. [ABSTRACT FROM AUTHOR]

The inter-shaft bearing is the core component of a high-pressure rotor support system of a high-thrust aero engine. One of the most challenging tasks for a PHM is monitoring its working condition. However, considering that in the bearing rotor system of a high-thrust aero engine bearings are prone to wear failure due to unbalanced or misaligned faults of the rotor system, especially in harsh environments, such as those at high operating loads and high rotation speeds, bearing wear can easily evolve into serious faults. Compared with aero engine fault diagnosis and RUL prediction, relatively little research has been conducted on bearing condition monitoring. In addition, considering how to evaluate future performance states with limited time series data is a key problem. At the same time, the current deep neural network model has the technical challenge of poor interpretability. In order to fill the above gaps, we developed a new framework of a residual space–time feature fusion focusing module named RTCA-Net, which focuses on solving the key problem. It is difficult to accurately monitor the wear state of aero engine inter-shaft bearings under special working conditions in practical engineering. Specifically, firstly, a residual space–time structure module was innovatively designed to capture the characteristic information of the metal dust signal effectively. Secondly, a feature-focusing module was designed. By adjusting the change in the weight coefficient during training, the RTCA-Net framework can select the more useful information for monitoring the wear condition of inter-shaft bearings. Finally, the experimental dataset of metal debris was verified and compared with seven other methods, such as the RTC-Net. The results showed that the proposed RTCA-Net framework has good generalization, superiority, and credibility. [ABSTRACT FROM AUTHOR]

With the development of edge computing and deep learning, intelligent human behavior recognition has spawned extensive applications in smart worlds. However, current edge computing technology faces performance bottlenecks due to limited computing resources at the edge, which prevent deploying advanced deep neural networks. In addition, there is a risk of privacy leakage during interactions between the edge and the server. To tackle these problems, we propose an effective, privacy-preserving edge–cloud collaborative interaction scheme based on WiFi, named P-CA, for human behavior sensing. In our scheme, a convolutional autoencoder neural network is split into two parts. The shallow layers are deployed on the edge side for inference and privacy-preserving processing, while the deep layers are deployed on the server side to leverage its computing resources. Experimental results based on datasets collected from real testbeds demonstrate the effectiveness and considerable performance of the P-CA. The recognition accuracy can maintain 88%, although it could achieve about 94.8% without the mixing operation. In addition, the proposed P-CA achieves better recognition accuracy than two state-of-the-art methods, i.e., FedLoc and PPDFL, by 2.7% and 2.1%, respectively, while maintaining privacy. [ABSTRACT FROM AUTHOR]

Even if the SARS-CoV-2 pandemic recedes, research regarding the effectiveness of government policies to contain the spread of the pandemic remains important. In this study, we analyze the impact of a set of epidemiological factors on the spread of SARS-CoV-2 in 30 European countries, which were applied from early 2020 up to mid-2022. We combine four data sets encompassing each country's non-pharmaceutical interventions (NPIs, including 66 government intervention types), distributions of 31 virus types, and accumulated percentage of vaccinated population (by the first five doses) as well as the reported infections, each on a daily basis. First, a Bayesian deep learning model is trained to predict the reproduction rate of the virus one month ahead of each day. Based on the trained deep learning model, the importance of relevant influencing factors and the magnitude of their effects on the outcome of the neural network model are computed by applying explainable machine learning algorithms. Second, in order to re-examine the results of the deep learning model, a Bayesian statistical analysis is implemented. In the statistical analysis, for each influencing input factor in each country, the distributions of pandemic growth rates are compared for days where the factor was active with days where the same factor was not active. The results of the deep learning model and the results of the statistical inference model coincide to a significant extent. We conclude with reflections with regard to the most influential factors on SARS-CoV-2 spread within European countries. [ABSTRACT FROM AUTHOR]

The purpose of this work is to design a novel process based on the deep neural network (DNN) process to solve the dynamical human immunodeficiency virus (HIV-1) infection system in cancer patients (HIV-1-ISCP). The dual hidden layer neural network structure using the combination of a radial basis and sigmoid function with twenty and forty neurons is presented for the solution of the nonlinear HIV-1-ISCP. The mathematical form of the model is divided into three classes named cancer population cells (T), healthy cells (H), and infected HIV (I) cells. The validity of the designed novel scheme is proven through the comparison of the results. The optimization is performed using a competent scale conjugate gradient procedure, the correctness of the proposed numerical approach is observed through the reference results, and negligible values of the absolute error are around 10−3 to 10−4. The database numerical solutions are achieved from the Runge–Kutta numerical scheme, and are used further to reduce the mean square error by taking 72% of the data for training, while 14% of the data is taken for testing and substantiations. To authenticate the credibility of this novel procedure, graphical plots using different performances are derived. [ABSTRACT FROM AUTHOR]

Credit risk refers to the possibility of borrower default, and its assessment is crucial for maintaining financial stability. However, the journey of credit risk data generation is often gradual, and machine learning techniques may not be readily applicable for crafting evaluations at the initial stage of the data accumulation process. This article proposes a credit risk modeling methodology, TED-NN, that first constructs an indicator system based on expert experience, assigns initial weights to the indicator system using the Analytic Hierarchy Process, and then constructs a neural network model based on the indicator system to achieve a smooth transition from an empirical model to a data-driven model. TED-NN can automatically adapt to the gradual accumulation of data, which effectively solves the problem of risk modeling and the smooth transition from no to sufficient data. The effectiveness of this methodology is validated through a specific case of credit risk assessment. Experimental results on a real-world dataset demonstrate that, in the absence of data, the performance of TED-NN is equivalent to the AHP and better than untrained neural networks. As the amount of data increases, TED-NN gradually improves and then surpasses the AHP. When there are sufficient data, its performance approaches that of a fully data-driven neural network model. [ABSTRACT FROM AUTHOR]

The economic time series prediction literature has seen an increase in research leveraging artificial neural networks (ANNs), particularly the multilayer perceptron (MLP) and, more recently, transformer networks. These ANN models have shown superior accuracy compared to traditional techniques such as autoregressive integrated moving average (ARIMA) models. The most recent models in the prediction of this type of neural network, such as recurrent or Transformers models, are composed of complex architectures that require sufficient processing capacity to address the problems, while MLP is based on densely connected layers and supervised learning. A deep understanding of the limitations is necessary to appropriately choose the ideal model for each of the prediction tasks. In this article, we show how a simple architecture such as the MLP allows a better adjustment than other models, including a shorter prediction time. This research is based on the premise that the use of the most recent models will not always allow better results. [ABSTRACT FROM AUTHOR]

Deep neural networks (DNNs) have gained considerable attention for their expressive capabilities, but unfortunately they have serious robustness risks. Formal verification is an important technique to ensure network reliability. However, current verification techniques are unsatisfactory in time performance, which hinders the practical applications. To address this issue, we propose an efficient optimization method based on parallel acceleration with more computing resources. The method involves the speedup configuration of a partition-based verification aligned with the structures and robustness formal specifications of DNNs. A parallel verification framework is designed specifically for neural network verification systems, which integrates various auxiliary modules and accommodates diverse verification modes. The efficient parallel scheduling of verification queries within the framework enhances resource utilization and enables the system to process a substantial volume of verification tasks. We conduct extensive experiments on multiple commonly used verification benchmarks to demonstrate the rationality and effectiveness of the proposed method. The results show that higher efficiency is achieved after parallel optimization integration. [ABSTRACT FROM AUTHOR]

The complexity of deep neural network models (DNNs) severely limits their application on devices with limited computing and storage resources. Knowledge distillation (KD) is an attractive model compression technology that can effectively alleviate this problem. Multi-teacher knowledge distillation (MKD) aims to leverage the valuable and diverse knowledge distilled by multiple teacher networks to improve the performance of the student network. Existing approaches typically rely on simple methods such as averaging the prediction logits or using sub-optimal weighting strategies to fuse distilled knowledge from multiple teachers. However, employing these techniques cannot fully reflect the importance of teachers and may even mislead student's learning. To address this issue, we propose a novel Decoupled Multi-Teacher Knowledge Distillation based on Entropy (DE-MKD). DE-MKD decouples the vanilla knowledge distillation loss and assigns adaptive weights to each teacher to reflect its importance based on the entropy of their predictions. Furthermore, we extend the proposed approach to distill the intermediate features from multiple powerful but cumbersome teachers to improve the performance of the lightweight student network. Extensive experiments on the publicly available CIFAR-100 image classification benchmark dataset with various teacher-student network pairs demonstrated the effectiveness and flexibility of our approach. For instance, the VGG8|ShuffleNetV2 model trained by DE-MKD reached 75.25%|78.86% top-one accuracy when choosing VGG13|WRN40-2 as the teacher, setting new performance records. In addition, surprisingly, the distilled student model outperformed the teacher in both teacher-student network pairs. [ABSTRACT FROM AUTHOR]

For large-scale optimization that covers a wide range of optimization problems encountered frequently in machine learning and deep neural networks, stochastic optimization has become one of the most used methods thanks to its low computational complexity. In machine learning and deep learning problems, nonconvex problems are common, while convex problems are rare. How to find the global minimum for nonconvex optimization and reduce the computational complexity are challenges. Inspired by the phenomenon that the stagewise stepsize tuning strategy can empirically improve the convergence speed in deep neural networks, we incorporate the stagewise stepsize tuning strategy into the iterative framework of Nesterov's acceleration- and variance reduction-based methods to reduce the computational complexity, i.e., the stagewise stepsize tuning strategy is incorporated into randomized stochastic accelerated gradient and stochastic variance-reduced gradient. The proposed methods are theoretically derived to reduce the complexity of the nonconvex and convex problems and improve the convergence rate of the frameworks, which have the complexity O (1 / μ ϵ) and O (1 / μ ϵ) , respectively, where μ is the PL modulus and L is the Lipschitz constant. In the end, numerical experiments on large benchmark datasets validate well the competitiveness of the proposed methods. [ABSTRACT FROM AUTHOR]

The timely and precise prediction of cardiovascular disease (CVD) risk is essential for effective prevention and intervention. This study proposes a novel framework that integrates the two-phase Taguchi method (TPTM), the hyperparameter artificial neural network (HANN), and a genetic algorithm (GA) called TPTM-HANN-GA. This framework efficiently optimizes hyperparameters for an artificial neural network (ANN) model during the training stage, significantly enhancing prediction accuracy for cardiovascular disease (CVD) risk. The proposed TPTM-HANN-GA framework requires far fewer experiments than a traditional grid search, making it highly suitable for application in resource-constrained, low-power computers, and edge artificial intelligence (edge AI) devices. Furthermore, the proposed TPTM-HANN-GA framework successfully identified the optimal configurations for the ANN model's hyperparameters, resulting in a hidden layer of 4 nodes, a tanh activation function, an SGD optimizer, a learning rate of 0.23425849, a momentum rate of 0.75462782, and seven hidden nodes. This optimized ANN model achieves 74.25% accuracy in predicting the risk of cardiovascular disease, which exceeds the existing state-of-the-art GA-ANN and TSTO-ANN models. The proposed TPTM-HANN-GA framework enables personalized CVD prediction to be efficiently conducted on low-power computers and edge-AI devices, achieving the goal of point-of-care testing (POCT) and empowering individuals to manage their heart health effectively. [ABSTRACT FROM AUTHOR]

Online multi-object tracking (MOT) techniques are instrumental in monitoring workers' positions and identities in construction settings. Traditional approaches, which employ deep neural networks (DNNs) for detection followed by body similarity matching, often overlook the significance of clear head features and stable head motions. This study presents a novel bidirectional tracking method that integrates intra-frame processing, which combines head and body analysis to minimize false positives and inter-frame matching to control ID assignment. By leveraging head information for enhanced body tracking, the method generates smoother trajectories with reduced ID errors. The proposed method achieved a state-of-the-art (SOTA) performance, with a multiple-object tracking accuracy (MOTA) of 95.191%, higher-order tracking accuracy (HOTA) of 78.884% and an identity switch (IDSW) count of 0, making it a strong baseline for future research. [ABSTRACT FROM AUTHOR]

Investigating public attitudes on social media is crucial for opinion mining systems. Stance detection aims to predict the attitude towards a specific target expressed in a text. However, effective neural stance detectors require substantial training data, which are challenging to curate due to the dynamic nature of social media. Moreover, deep neural networks (DNNs) lack explainability, rendering them unsuitable for scenarios requiring explanations. We propose a distantly supervised explainable stance detection framework (DS-ESD), comprising an instruction-based chain-of-thought (CoT) method, a generative network, and a transformer-based stance predictor. The CoT method employs prompt templates to extract stance detection explanations from a very large language model (VLLM). The generative network learns the input-explanation mapping, and a transformer-based stance classifier is trained with VLLM-annotated stance labels, implementing distant supervision. We propose a label rectification strategy to mitigate the impact of erroneous labels. Experiments on three benchmark datasets showed that our model outperformed the compared methods, validating its efficacy in stance detection tasks. This research contributes to the advancement of explainable stance detection frameworks, leveraging distant supervision and label rectification strategies to enhance performance and interpretability. [ABSTRACT FROM AUTHOR]

The motivation behind this study is to overcome the complex mathematical formulation and time-consuming nature of traditional numerical methods used in solving differential equations. It seeks an alternative approach for more efficient and simplified solutions. A Deep Neural Network (DNN) is utilized to understand the intricate correlations between the oscillator's variables and to precisely capture their dynamics by being trained on a dataset of known oscillator behaviors. In this work, we discuss the main challenge of predicting the behavior of oscillators without depending on complex strategies or time-consuming simulations. The present work proposes a favorable modified form of neural structure to improve the strategy for simulating linear and nonlinear harmonic oscillators from mechanical systems by formulating an ANN as a DNN via an appropriate oscillating activation function. The proposed methodology provides the solutions of linear and nonlinear differential equations (DEs) in differentiable form and is a more accurate approximation as compared to the traditional numerical method. The Van der Pol equation with parametric damping and the Mathieu equation are adopted as illustrations. Experimental analysis shows that our proposed scheme outperforms other numerical methods in terms of accuracy and computational cost. We provide a comparative analysis of the outcomes obtained through our proposed approach and those derived from the LSODA algorithm, utilizing numerical techniques, Adams–Bashforth, and the Backward Differentiation Formula (BDF). The results of this research provide insightful information for engineering applications, facilitating improvements in energy efficiency, and scientific innovation. [ABSTRACT FROM AUTHOR]

Inverse modeling can estimate uncertain parameters in subsurface reservoirs and provide reliable numerical models for reservoir development and management. The traditional simulation-based inversion method usually requires numerous numerical simulations, which is time-consuming. Recently, deep learning-based surrogate models have been widely studied as an alternative to numerical simulation, which can significantly improve the solving efficiency of inversion. However, for reservoirs with complex fracture distribution, constructing the surrogate model of numerical simulation presents a significant challenge. In this work, we present a deep graph learning-based surrogate model for inverse modeling of fractured reservoirs. Specifically, the proposed surrogate model integrates the graph attention mechanisms to extract features of fracture network in reservoirs. The graph learning can retain the discrete characteristics and structural information of the fracture network. The extracted features are subsequently integrated with a multi-layer recurrent neural network model to predict the production dynamics of wells. A surrogate-based inverse modeling workflow is then developed by combining the surrogate model with the differential evolutionary algorithm. Numerical studies performed on a synthetic naturally fractured reservoir model with multi-scale fractures illustrate the performance of the proposed methods. The results demonstrate that the proposed surrogate model exhibits promising generalization performance of production prediction. Compared with tens of thousands of numerical simulations required by the simulation-based inverse modeling method, the proposed surrogate-based method only requires 1000 to 1500 numerical simulations, and the solution efficiency can be improved by ten times. [ABSTRACT FROM AUTHOR]

The primary objective of introducing metaheuristic algorithms into traditional systematic logic is to minimize the cost function. However, there is a lack of research on the impact of introducing metaheuristic algorithms on the cost function under different proportions of positive literals. In order to fill in this gap and improve the efficiency of the metaheuristic algorithm in systematic logic, we proposed a metaheuristic algorithm based on mutation tabu search and embedded it in probabilistic satisfiability logic in discrete Hopfield neural networks. Based on the traditional tabu search algorithm, the mutation operators of the genetic algorithm were combined to improve its global search ability during the learning phase and ensure that the cost function of the systematic logic converged to zero at different proportions of positive literals. Additionally, further optimization was carried out in the retrieval phase to enhance the diversity of solutions. Compared with nine other metaheuristic algorithms and exhaustive search algorithms, the proposed algorithm was superior to other algorithms in terms of time complexity and global convergence, and showed higher efficiency in the search solutions at the binary search space, consolidated the efficiency of systematic logic in the learning phase, and significantly improved the diversity of the global solution in the retrieval phase of systematic logic. [ABSTRACT FROM AUTHOR]

This study presents a novel event-triggered relearning framework for neural network modeling, designed to improve prediction precision in dynamic stochastic complex industrial systems under non-stationary and variable conditions. Firstly, a sliding window algorithm combined with entropy is applied to divide the input and output datasets across different operational conditions, establishing clear data boundaries. Following this, the prediction errors derived from the neural network under different operational states are harnessed to define a set of event-triggered relearning criteria. Once these conditions are triggered, the relevant dataset is used to recalibrate the model to the specific operational condition and predict the data under this operating condition. When the predicted data fall within the training input range of a pre-trained model, we switch to that model for immediate prediction. Compared with the conventional BP neural network model and random vector functional-link network, the proposed model can produce a better estimation accuracy and reduce computation costs. Finally, the effectiveness of our proposed method is validated through numerical simulation tests using nonlinear Hammerstein models with Gaussian noise, reflecting complex stochastic industrial processes. [ABSTRACT FROM AUTHOR]

Currently, direct surveys are used less and less to assess satisfaction with the quality of user services. One of the most effective methods to solve this problem is to extract user attitudes from social media texts using natural language text mining. This approach helps to obtain more objective results by increasing the representativeness and independence of the sample of service consumers being studied. The purpose of this article is to improve existing methods and test a method for classifying Russian-language text reviews of patients about the work of medical institutions and doctors, extracted from social media resources. The authors developed a hybrid method for classifying text reviews about the work of medical institutions and tested machine learning methods using various neural network architectures (GRU, LSTM, CNN) to achieve this goal. More than 60,000 reviews posted by patients on the two most popular doctor review sites in Russia were analysed. Main results: (1) the developed classification algorithm is highly efficient—the best result was shown by the GRU-based architecture (val_accuracy = 0.9271); (2) the application of the method of searching for named entities to text messages after their division made it possible to increase the classification efficiency for each of the classifiers based on the use of artificial neural networks. This study has scientific novelty and practical significance in the field of social and demographic research. To improve the quality of classification, in the future, it is planned to expand the semantic division of the review by object of appeal and sentiment and take into account the resulting fragments separately from each other. [ABSTRACT FROM AUTHOR]

We consider the inverse scattering problem to reconstruct an obstacle using partial far-field data due to one incident wave. A simple indicator function, which is negative inside the obstacle and positive outside of it, is constructed and then learned using a deep neural network (DNN). The method is easy to implement and effective as demonstrated by numerical examples. Rather than developing sophisticated network structures for the classical inverse operators, we reformulate the inverse problem as a suitable operator such that standard DNNs can learn it well. The idea of the DNN-oriented indicator method can be generalized to treat other partial data inverse problems. [ABSTRACT FROM AUTHOR]

Industry is the primary application for induction machines. As such, it is essential to calculate the induction devices' electrical properties accurately. With DC testing, no-load rotor tests, and locked rotor tests, one may empirically evaluate the electrical variables of induction motors. These tests are expensive and difficult to conduct, however. The information supplied by machine makers can also be used to accurately approximate the equivalent variables of the circuits in induction machines. This article has successfully predicted motor reactance (Xm) for both double- and single-cage models using artificial neural networks (ANN). Although ANNs have been investigated in the literature, the ANN structures were trained to use unmemorized training. Besides ANN, six other approaches have been suggested to address this issue: heap-based optimization (HBO), leagues championship algorithm (LCA), multi-verse optimization (MVO), osprey optimization algorithm (OOA), cuckoo optimization algorithm (COA), and sooty tern optimization algorithm (STOA). The efficaciousness of the suggested approaches was compared with each another. Regarding the obtained outcomes, the suggested MVO- multi-layer perceptron (MLP) technique performed better than the other five methods regarding reactance prediction, with R2 of 0.99598 and 0.9962, and RMSE of 20.31492 and 20.80626 in the testing and training phases, respectively. For the projected model, the suggested ANNs have produced great results. The novelty lies in the mentioned methods' ability to tackle the complexities and challenges associated with induction motor reactance optimization, providing innovative approaches to finding optimal or near-optimal solutions. As researchers continue to explore and refine these techniques, their impact on motor design and efficiency will likely grow, driving advancements in electrical engineering. [ABSTRACT FROM AUTHOR]

One of the tasks in the operation of electric power systems is the correct functioning of the protection system and emergency automation algorithms. Instrument voltage and current transformers, operating in accordance with the laws of electromagnetism, are most often used for information support of the protection system and emergency automation algorithms. Magnetic core saturation of the specified current transformers can occur during faults. As a result, the correct functioning of the protection system and emergency automation algorithms is compromised. The consequences of current transformers saturation are mostly reflected in the main protections of network elements operating on a differential principle. This work aims to consider the analysis of current transformer saturation detection methods. The problem of identifying current transformer saturation is reduced to binary classification, and methods for solving the problem based on artificial neural networks, support vector machine, and decision tree algorithms are proposed. Computational experiments were performed, and their results were analyzed with imbalanced (dominance of the number of current transformer saturation modes over the number of modes with its normal operation) and balanced classes 0 (no current transformer saturation) and 1 (current transformer saturation). [ABSTRACT FROM AUTHOR]

Context: YOLO (You Look Only Once) is an algorithm based on deep neural networks with real-time object detection capabilities. This state-of-the-art technology is widely available, mainly due to its speed and precision. Since its conception, YOLO has been applied to detect and recognize traffic signs, pedestrians, traffic lights, vehicles, and so on. Objective: The goal of this research is to systematically analyze the YOLO object detection algorithm, applied to traffic sign detection and recognition systems, from five relevant aspects of this technology: applications, datasets, metrics, hardware, and challenges. Method: This study performs a systematic literature review (SLR) of studies on traffic sign detection and recognition using YOLO published in the years 2016–2022. Results: The search found 115 primary studies relevant to the goal of this research. After analyzing these investigations, the following relevant results were obtained. The most common applications of YOLO in this field are vehicular security and intelligent and autonomous vehicles. The majority of the sign datasets used to train, test, and validate YOLO-based systems are publicly available, with an emphasis on datasets from Germany and China. It has also been discovered that most works present sophisticated detection, classification, and processing speed metrics for traffic sign detection and recognition systems by using the different versions of YOLO. In addition, the most popular desktop data processing hardwares are Nvidia RTX 2080 and Titan Tesla V100 and, in the case of embedded or mobile GPU platforms, Jetson Xavier NX. Finally, seven relevant challenges that these systems face when operating in real road conditions have been identified. With this in mind, research has been reclassified to address these challenges in each case. Conclusions: This SLR is the most relevant and current work in the field of technology development applied to the detection and recognition of traffic signs using YOLO. In addition, insights are provided about future work that could be conducted to improve the field. [ABSTRACT FROM AUTHOR]

Building deep learning models proposed by third parties can become a simple task when specialized libraries are used. However, much mystery still surrounds the design of new models or the modification of existing ones. These tasks require in-depth knowledge of the different components or building blocks and their dimensions. This information is limited and broken up in different literature. In this article, we collect and explain the building blocks used to design deep learning models in depth, starting from the artificial neuron to the concepts involved in building deep neural networks. Furthermore, the implementation of each building block is exemplified using the Keras library. [ABSTRACT FROM AUTHOR]

Artificial neural networks are successfully used to solve a wide variety of scientific and technical problems. The purpose of the study is to increase the efficiency of distributed solutions for problems involving structural-parametric synthesis of neural network models of complex systems based on GRID (geographically disperse computing resources) technology through the integrated application of the apparatus of evolutionary optimization and queuing theory. During the course of the research, the following was obtained: (i) New mathematical models for assessing the performance and reliability of GRID systems; (ii) A new multi-criteria optimization model for designing GRID systems to solve high-resource computing problems; and (iii) A new decision support system for the design of GRID systems using a multi-criteria genetic algorithm. Fonseca and Fleming's genetic algorithm with a dynamic penalty function was used as a method for solving the stated multi-constrained optimization problem. The developed program system was used to solve the problem of choosing an effective structure of a centralized GRID system that was configured to solve the problem of structural-parametric synthesis of neural network models. To test the proposed approach, a Pareto-optimal configuration of the GRID system was built with the following characteristics: average performance–103.483 GFLOPS, cost–500 rubles per day, availability rate–99.92%, and minimum performance–51 GFLOPS. [ABSTRACT FROM AUTHOR]