Search

The introduction of hybrid alternating current (AC)/direct current (DC) distribution networks led to several developments in smart grid and decentralized power system technology. The paper concentrates on several topics related to the operation of hybrid AC/DC networks. Such as optimization methods, control strategies, energy management, protection issues, and proposed solutions. The implementation of neural network optimization methods has great importance for the successful integration of multiple energy sources, dynamic energy management, establishment of system stability and reliability, power distribution optimization, management of energy storage, and online fault detection and diagnosis in hybrid networks like the hybrid AC–DC microgrids (MG). Taking advantage of renewable energy generation and cost-cutting through the neural network optimization technique holds the key to these progressions. Besides identifying the challenges in the operation of a hybrid system, the paper also compares this system to conventional MGs and shows the benefits of this type of system over different MG structures. This review compares the different topologies, particularly looking at the AC–DC coupled hybrid MGs, and shows the important role of the interlinking of converters that are used for efficient transmission between AC and DC MGs and generally used to implement the different control and optimization techniques. Overall, this review paper can be regarded as a reference, pointing out the pros and cons of integrating hybrid AC/DC distribution networks for future study and improvement paths in this developing area. [ABSTRACT FROM AUTHOR]

Blockchain is recognized for its robust security features, and its integration with Internet of Things (IoT) systems presents scalability and operational challenges. Deploying Artificial Intelligence (AI) within blockchain environments raises concerns about balancing rigorous security requirements with computational efficiency. The prime motivation resides in integrating AI with blockchain to strengthen IoT security and withstand multiple variants of lethal threats. With the increasing number of IoT devices, there has also been a spontaneous increase in security vulnerabilities. While conventional security methods are inadequate for the diversification of IoT devices, adopting AI can assist in identifying and mitigating such threats in real time, whereas integrating AI with blockchain can offer more intelligent decentralized security measures. The paper contributes to a three-layered architecture encompassing the device/sensory, edge, and cloud layers. This structure supports a novel method for assessing legitimacy scores and serves as an initial security measure. The proposed scheme also enhances the architecture by introducing an Ethereum-based data repositioning framework as a potential trapdoor function, ensuring maximal secrecy. To complement this, a simplified consensus module generates a conclusive evidence matrix, bolstering accountability. The model also incorporates an innovative AI-based security optimization utilizing an unconventional neural network model that operates faster and is enhanced with metaheuristic algorithms. Comparative benchmarks demonstrate that our approach results in a 48.5% improvement in threat detection accuracy and a 23.5% reduction in processing time relative to existing systems, marking significant advancements in IoT security for smart cities. [ABSTRACT FROM AUTHOR]

In the current era, the government consistently emphasizes the pursuit of high-quality development, as evidenced by the ongoing increase in the tertiary industry's GDP share. As a crucial component of the modern service sector, logistics plays a pivotal role in determining the operational efficiency and overall quality of the industrial economy. This study focuses on constructing a Chongqing logistics express demand prediction index system. It employs an improved BP neural network model to forecast the logistics express demand for Chongqing over the next five years. Given the limited express demand data sequence and the normalized characteristics of the data, the selected training method is the Bayesian regularization approach, with the LeCun Tanh function serving as the hidden layer activation function. Additionally, a genetic algorithm is designed to optimize the initial weights and thresholds of the BP neural network, thereby enhancing prediction accuracy and reducing the number of iterations. The experimental results of the improved GA-BP network are analyzed and compared, demonstrating that the improved BP neural network, utilizing GA optimization, can more reliably and accurately predict regional logistics express demand. According to the findings, the forecast indicates that the logistics express demand for Chongqing in 2026 will be 2,171,642,700 items. [ABSTRACT FROM AUTHOR]

The color features of strawberries at different growth stages vary slightly and occluded during growth. To address these challenges, this study proposes a lightweight multi-stage detection method based on You Only Look Once version 7-tiny (YOLOv7-tiny) for strawberries in complex environments. First, the size of the model is reduced by replacing the ordinary convolution of the neck network used for deep feature extraction and fusion with lightweight Ghost convolution. Then, by introducing the Coordinate Attention (CA) module, the model's focus on the target detection area is enhanced, thereby improving the detection accuracy of strawberries. The Wise Intersection over Union (WIoU) loss function is integrated to accelerate model convergence and enhance the recognition accuracy of occluded targets. The advanced Adaptive nesterov momentum algorithm (Adan) is utilized for gradient descent, processing averaged sample data. Additionally, considering the small size of strawberry targets, a detection head specifically for small targets is added, performing detection on a 160 × 160 × 64 feature map, which significantly improves the detection performance for small strawberries. Experimental results demonstrate that the improved network model achieves an m A P @ 0.5 of 88.2% for multi-stage strawberry detection, which is 2.44% higher than the original YOLOv7-tiny algorithm. Meanwhile, G F L O P s and P a r a m s are reduced by 1.54% and 12.10%, respectively. In practical detection and inference, the improved model outperforms current mainstream target detection models, enabling a quicker and more accurate identification of strawberries at different growth stages, thus providing technical support for intelligent strawberry picking. [ABSTRACT FROM AUTHOR]

With the advent of the digital era, the quantity and variety of online higher education learning resources have expanded rapidly. The efficient adaptation of suitable resources to meet the needs of learners with specific requirements has become crucial for improving learning outcomes. Although current online learning resource recommendation systems have made some progress in matching resources, they still face challenges related to the inadequate integration of resource features and a superficial understanding of learners' needs. These challenges hinder the achievement of personalized and precise matching, affecting learners' study effi- ciency and the effective utilization of educational resources. This study first analyzes the importance of adapting online higher education learning resources and the limitations of existing research. Subsequently, a novel neural network optimization strategy is proposed. The research comprises two main parts. Firstly, the self-attention-convolutional neural network (SA-CNN) model is employed for the deep integration of the content features of online learning resources. This aims to enhance the comprehensiveness of resource descriptions. Secondly, a deep-metric attention model is introduced to accurately model and adapt to learners' needs. This approach not only optimizes the feature representation of learning resources but also enhances the sensitivity and accuracy of the recommendation system towards learners' requirements. This study is of significant importance for improving the performance of higher education online learning resource recommendation systems. It also provides new insights into the construction of personalized learning paths and ensuring the balanced allocation of educational resources. [ABSTRACT FROM AUTHOR]

Optimization of the “mature” fields development in machine learning algorithms is one of the urgent problems nowadays. The task is set to extend the effective operation of wells, optimize production management at the late stage of field development. Based on the task set, the article provides an overview of possible solutions in waterflooding management problems. Production management technology is considered as an alternative to intensification of operation, which is associated with an increase in the produciton rate and involves finding solutions aimed at reducing the water cut of well production. The practical implementation of the “Neural technologies for production improvement” includes the following steps: evaluation, selection, predictive analytics. The result is a digital technological regime of wells that corresponds to the set goal and the solution of the optimization problem in artificial intelligence algorithms using the software and hardware complex “Atlas – Waterflood Management”.“Neural technologies for production improvement” have been successfully tested at the pilot project site of the productive formation of the Vatyeganskoe field. The article provides a thorough and detailed analysis of the work performed, describes the algorithms and calculation results of the proxy model using the example of the pilot area, as well as the integration of the “Atlas – Waterflood Management” and the organization of the workflow with the field professionals of the Territorial Production Enterprise Povkhneftegaz.

Neural networks (NNs) have shown outstanding performance in solar photovoltaic (PV) power forecasting due to their ability to effectively learn unstable environmental variables and their complex interactions. However, NNs are limited in their practical industrial application in the energy sector because the optimization of the model structure or hyperparameters is a complex and time-consuming task. This paper proposes a two-stage NN optimization method for robust solar PV power forecasting. First, the solar PV power dataset is divided into training and test sets. In the training set, several NN models with different numbers of hidden layers are constructed, and Optuna is applied to select the optimal hyperparameter values for each model. Next, the optimized NN models for each layer are used to generate estimation and prediction values with fivefold cross-validation on the training and test sets, respectively. Finally, a random forest is used to learn the estimation values, and the prediction values from the test set are used as input to predict the final solar PV power. As a result of experiments in the Incheon area, the proposed method is not only easy to model but also outperforms several forecasting models. As a case in point, with the New-Incheon Sonae dataset—one of three from various Incheon locations—the proposed method achieved an average mean absolute error (MAE) of 149.53 kW and root mean squared error (RMSE) of 202.00 kW. These figures significantly outperform the benchmarks of attention mechanism-based deep learning models, with average scores of 169.87 kW for MAE and 232.55 kW for RMSE, signaling an advance that is expected to make a significant contribution to South Korea's energy industry. [ABSTRACT FROM AUTHOR]

The rapid advancements of Artificial Intelligence (AI) technologies, particularly Large Language Models (LLMs), have brought and accelerated significant innovations across various domains. Regardless of their widespread usefulness, the scalability of LLMs poses considerable challenges, primarily due to their substantial demands on computational and energy resources. This article explores the importance of developing and applying effective compression techniques to mitigate these numerous challenges. Techniques such as pruning, quantization, and knowledge distillation are analyzed for their potential to decrease a LLM's size and its associated computational demands, while striving to maintain performance integrity. Each technique inherently presents unique trade-offs between model efficiency and accuracy, requiring a nuanced understanding of their applications. We have made an in-depth analysis into the complexities of implementing these techniques, highlighting the balance required between performance and compression, along with the complex process of customization to specific LLM architectures. The article further analyzes the very important validation and testing phases that are much needed for ensuring that compressed models perform adequately in real-world applications. We have also considered the future adaptability of compression techniques to evolving AI models and architectures. The conducted study emphasizes the ongoing need for innovative research in model compression in order to make AI technologies more sustainable and accessible across various sectors, thereby expanding their potential benefits while addressing the limitations and risks associated with their deployment. [ABSTRACT FROM AUTHOR]

This study refines the YOLOv7-tiny model through structured pruning and architectural fine-tuning, specifically for real-time eye state detection. By focusing on enhancing the model's efficiency, particularly in environments with limited computational resources, this research contributes significantly to advancing driver monitoring systems, where timely and accurate detection of eye states such as openness or closure can prevent accidents caused by drowsiness or inattention. Structured pruning was utilized to simplify the YOLOv7-tiny model, reducing complexity and storage requirements. Subsequent fine-tuning involved adjustments to the model's width and depth to further enhance processing speed and efficiency. The experimental outcomes reveal a pronounced reduction in storage size, of approximately 97%, accompanied by a sixfold increase in frames per second (FPS). Despite these substantial modifications, the model sustains high levels of precision, recall, and mean average precision (mAP). These improvements indicate a significant enhancement in both the speed and efficiency of the model, rendering it highly suitable for real-time applications where computational resources are limited. [ABSTRACT FROM AUTHOR]

Parameter pruning is one of the primary methods for compressing CNN models, aiming to reduce redundant parameters, the complexity of time and space, and the calculation resources of the network, all while ensuring minimal loss in the network's performance. Currently, most existing parameter pruning methods adopt equal pruning rates across all layers. Different from previous methods, this paper focuses on the optimal combination of each layer's pruning rates within a given pruning rate of the whole model. Genetic algorithm is used to determine the pruning rate for each layer. It's worth noting that while the pruning rate for individual layers may vary, the average pruning rate across all layers does not exceed the given pruning rate. Experimental validation is conducted on CIFAR10 and ImageNet ILSVRC2012 datasets using VGGNet and ResNet architectures. The results show that the accuracy loss and the FLOPs of the pruned model using our method are superior to those pruned using previous methods. [ABSTRACT FROM AUTHOR]

Purpose: The constitutive parameters of viscoelastic materials, such as storage modulus and loss factor, usually have frequency-dependent characteristics. The combination of polymers with different reinforcement and fillers usually exhibits various mechanical characteristics, which makes the identification of the material properties of viscoelastic materials a challenging task. The present study proposes an inverse identification technique based on a neural network optimization (NNO) algorithm to characterize the frequency-dependent material properties of a viscoelastic material. Methods: To this end, a symmetric three-layered sandwich plate is considered having face layers of isotropic elastic material and a core layer of viscoelastic material. The experimental free vibration tests are performed using the impact hammer method to determine resonant frequencies and modal loss factors for various eigenmodes. In addition, a numerical model of the sandwich plate is developed to determine vibrational responses utilizing the finite element method. The vibration-based material parameter identification technique is implemented based on the NNO algorithm. The identified material parameters are then compared with the experimental dynamic mechanical analysis (DMA) test results. Furthermore, a numerical parametric study is performed considering the optimized viscoelastic material properties to investigate the influence of various geometrical and structural factors on the free vibration response of the sandwich plate. Results: The identified results are in excellent agreement with the experimental DMA test results affirming the robustness of the proposed inverse technique. The parametric study not only investigates the effect of various structural and geometric parameters on the dynamic response of the sandwich plate but also verifies that the calibrated properties are both realistic and physically meaningful. Conclusions: The proposed algorithm is a useful and efficient tool for inverse identification of the constitutive properties, and this approach can be extended for the calibration of other parameters (constitutive or not) for a variety of viscoelastic materials in any field of application. This study is a critical step forward in understanding viscoelastic materials and their frequency-dependent behaviorur. [ABSTRACT FROM AUTHOR]

The quality of micro-milling parts is greatly influenced by surface roughness. However, the evaluation of surface roughness solely based on two-dimensional (2D) parameters is limited. In order to more comprehensively reflect the processing quality and surface performance in the micro-milling process, this paper selected three representative parameters, namely Sa, Sq, and Sdr, as the characterization parameters for three-dimensional surface roughness (3D-SR). The mapping relationship between the 3D-SR and spindle rotational velocity, feed velocity, radial machining depth, axial machining depth was experimentally investigated, then the relationship models were established using response surface method (RSM). Subsequently, this study employed the backpropagation neural network (BPNN) to predict the 3D-SR. While maintaining the same BPNN topology, we introduced particle swarm optimization (PSO) as well as genetic algorithm (GA) to enhance the initial thresholds and connection weights of the network. This allowed us to establish three prediction models for 3D-SR. The results indicate that the PSO-BPNN prediction models have better prediction performance than other models of this paper. [ABSTRACT FROM AUTHOR]

The subject of study in this article is the technologies of Field Programmable Gate Array (FPGA), methods, and tools for prototyping of hardware accelerators of Artificial Intelligence (AI) and providing it as a service. The goal is to reduce the efforts of creation and modification of FPGA implementation of Artificial Intelligent projects and provide such solutions as a service. Task: to analyze the possibilities of heterogeneous computing for the implementation of AI projects; analyze advanced FPGA technologies and accelerator cards that allow the organization of a service; analyze the languages, frameworks, and integrated environments for the creation of Artificial Intelligence projects for FPGA implementation; propose a technique for modifiable FPGA project prototyping to ensure a long period of compatibility with integrated environments and target devices; propose a technique for the prototyping of FPGA services with high performance to improve the efficiency of FPGA based AI projects; propose a sequence of optimization of neural networks for FPGA implementation; and provide an example of the practical implementation of the research results. According to the tasks, the following results were obtained. Analysis of the biggest companies and vendors of FPGA technology is performed. Existing heterogeneous technologies and potential non-electronic mediums for AI computations are discussed. FPGA accelerator cards with a large amount of High Bandwidth Memory (HBM) on the same chip package for implementation of AI projects are analyzed and compared. Languages, frameworks, and technologies as well as the capabilities of libraries and integrated environments for prototyping of FPGA projects for the AI applications are analyzed in detail. The sequence of prototyping of FPGA projects that are stable to changes in the environment is proposed. The sequence of prototyping of highly efficient pipelined projects for data processing is proposed. The steps of optimization of neural networks for FPGA implementation of AI applications are provided. An example of practical use of the results of research, including the use of sequences is provided. Conclusions. One of the main contributions of this research is the proposed method of creation of FPGA based implementation of AI projects in the form of services. Proposed sequence of neural network optimization for FPGA allows the reduction of the complexity of the initial program model by more than five times for hardware implementation depending on the required accuracy. The described solutions allow the construction of completely scalable and modifiable FPGA implementations of AI projects to provide it as a service.

In the current era, the government consistently emphasizes the pursuit of high-quality development, as evidenced by the ongoing increase in the tertiary industry’s GDP share. As a crucial component of the modern service sector, logistics plays a pivotal role in determining the operational efficiency and overall quality of the industrial economy. This study focuses on constructing a Chongqing logistics express demand prediction index system. It employs an improved BP neural network model to forecast the logistics express demand for Chongqing over the next five years. Given the limited express demand data sequence and the normalized characteristics of the data, the selected training method is the Bayesian regularization approach, with the LeCun Tanh function serving as the hidden layer activation function. Additionally, a genetic algorithm is designed to optimize the initial weights and thresholds of the BP neural network, thereby enhancing prediction accuracy and reducing the number of iterations. The experimental results of the improved GA-BP network are analyzed and compared, demonstrating that the improved BP neural network, utilizing GA optimization, can more reliably and accurately predict regional logistics express demand. According to the findings, the forecast indicates that the logistics express demand for Chongqing in 2026 will be 2,171,642,700 items.

The color features of strawberries at different growth stages vary slightly and occluded during growth. To address these challenges, this study proposes a lightweight multi-stage detection method based on You Only Look Once version 7-tiny (YOLOv7-tiny) for strawberries in complex environments. First, the size of the model is reduced by replacing the ordinary convolution of the neck network used for deep feature extraction and fusion with lightweight Ghost convolution. Then, by introducing the Coordinate Attention (CA) module, the model’s focus on the target detection area is enhanced, thereby improving the detection accuracy of strawberries. The Wise Intersection over Union (WIoU) loss function is integrated to accelerate model convergence and enhance the recognition accuracy of occluded targets. The advanced Adaptive nesterov momentum algorithm (Adan) is utilized for gradient descent, processing averaged sample data. Additionally, considering the small size of strawberry targets, a detection head specifically for small targets is added, performing detection on a 160 × 160 × 64 feature map, which significantly improves the detection performance for small strawberries. Experimental results demonstrate that the improved network model achieves an mAP@0.5 of 88.2% for multi-stage strawberry detection, which is 2.44% higher than the original YOLOv7-tiny algorithm. Meanwhile, GFLOPs and Params are reduced by 1.54% and 12.10%, respectively. In practical detection and inference, the improved model outperforms current mainstream target detection models, enabling a quicker and more accurate identification of strawberries at different growth stages, thus providing technical support for intelligent strawberry picking.

Today, the evolution of art teaching evaluation systems faces significant challenges and opportunities, particularly with the limitations of subjective judgments in traditional methods impacting the accuracy of teaching quality assessments. This study leverages advancements in artificial intelligence, specifically neural networks enhanced by the Taruhai Sheath algorithm, to propose a novel and more objective approach for evaluating art teaching. Our optimized neural network model demonstrates a remarkable 92% accuracy in art teaching evaluations, outperforming traditional methods by 15%. Furthermore, our analysis reveals that the quality of practical teaching is a critical factor in boosting students’ artistic creativity. By applying the Taruhai Sheath algorithm for neural network optimization, this research offers a groundbreaking method to elevate the objectivity and rigor of art education evaluations, promising a significant leap forward in art teaching quality.

Artificial Intelligence (AI) techniques are considered the most advanced approaches for diagnosing faults in power transformers. Dissolved Gas Analysis (DGA) is the conventional approach widely adopted for diagnosing incipient faults in power transformers. The IEC-599 standard Ratio Method is an accurate method that evaluates the DGA. All the classical approaches have limitations because they cannot diagnose all faults accurately. Precisely diagnosing defects in power transformers is a significant challenge due to their extensive quantity and dispersed placement within the power network. To deal with this concern and to improve the reliability and precision of fault diagnosis, different Artificial Intelligence techniques are presented. In this manuscript, an artificial neural network (ANN) is implemented to enhance the accuracy of the Rogers Ratio Method. On the other hand, it should be noted that the complexity of an ANN demands a large amount of storage and computing power. In order to address this issue, an optimization technique is implemented with the objective of maximizing the accuracy and minimizing the architectural complexity of an ANN. All the procedures are simulated using the MATLAB R2023a software. Firstly, the authors choose the most effective classification model by automatically training five classifiers in the Classification Learner app (CLA). After selecting the artificial neural network (ANN) as the sufficient classification model, we trained 30 ANNs with different parameters and determined the 5 models with the best accuracy. We then tested these five ANNs using the Experiment Manager app and ultimately selected the ANN with the best performance. The network structure is determined to consist of three layers, taking into consideration both diagnostic accuracy and computing efficiency. Ultimately, a (100-50-5) layered ANN was selected to optimize its hyperparameters. As a result, following the implementation of the optimization techniques, the suggested ANN exhibited a high level of accuracy, up to 90.7%. The conclusion of the proposed model indicates that the optimization of hyperparameters and the increase in the number of data samples enhance the accuracy while minimizing the complexity of the ANN. The optimized ANN is simulated and tested in MATLAB R2023a—Deep Network Designer, resulting in an accuracy of almost 90%. Moreover, compared to the Rogers Ratio Method, which exhibits an accuracy rate of just 63.3%, this approach successfully addresses the constraints associated with the conventional Rogers Ratio Method. So, the ANN has evolved a supremacy diagnostic method in the realm of power transformer fault diagnosis. [ABSTRACT FROM AUTHOR]

This study refines the YOLOv7-tiny model through structured pruning and architectural fine-tuning, specifically for real-time eye state detection. By focusing on enhancing the model’s efficiency, particularly in environments with limited computational resources, this research contributes significantly to advancing driver monitoring systems, where timely and accurate detection of eye states such as openness or closure can prevent accidents caused by drowsiness or inattention. Structured pruning was utilized to simplify the YOLOv7-tiny model, reducing complexity and storage requirements. Subsequent fine-tuning involved adjustments to the model’s width and depth to further enhance processing speed and efficiency. The experimental outcomes reveal a pronounced reduction in storage size, of approximately 97%, accompanied by a sixfold increase in frames per second (FPS). Despite these substantial modifications, the model sustains high levels of precision, recall, and mean average precision (mAP). These improvements indicate a significant enhancement in both the speed and efficiency of the model, rendering it highly suitable for real-time applications where computational resources are limited.

The over-parameterization of neural networks and the local optimality of backpropagation algorithm have been two major problems associated with deep-learning. In order to reduce the redundancy of neural network parameters, the conventional approach has been to prune branches with small weights. However, this only solves the problem of parameter redundancy, not providing any global optimality guarantees. In this paper, we overturn back-propagation and combine the sparse network optimization problem and the network weight optimization problem using a non-convex optimization method, namely Simulated Annealing. This method can complete network training under the premise of controlling the amount of parameters. Different from simply updating network parameters using gradient descent, our method simultaneously optimizes the topology of the sparse network. With the guarantee of global optimality of Simulated Annealing solution, the performance of the sparse network optimized by our method has exceeded the one trained by backpropagation only.

Through the stability evaluation of a slope, a landslide geological disaster can be identified, and the safety and risk control of a project can be ensured. This work proposes an improved sparrow search algorithm to optimize the slope safety factor prediction model (ISSA–BP) of a BP neural network, through an improvement in two aspects: introducing dynamic weight factors and reverse learning strategies to realize adaptive searches. The optimal value improves a defect in the traditional model, preventing it from easily falling into the local minimum. First, combined with 352 sets of actual slope data, three machine learning models were used to predict the safety factor of the slope. Then, the accuracy index was used for evaluation. Compared with other models, the MAPE, RMSE, and R2 of the ISSA-BP model were 1.64%, 0.0296, and 0.99, respectively, and the error was reduced by 78% compared with the BP neural network, showing better accuracy. Finally, the three models were applied to the slope stability analysis of Tianbao Port in Wenshan Prefecture. The research shows that the predicted value of the ISSA–BP model was the closest to the actual safety factor, which verified the experimental results. The improved ISSA–BP model can effectively predict the safety factor of slopes under different conditions, and it provides a new technology for slope disaster warning and control. [ABSTRACT FROM AUTHOR]

The subject of study in this article is the technologies of Field Programmable Gate Array (FPGA), methods, and tools for prototyping of hardware accelerators of Artificial Intelligence (AI) and providing it as a service. The goal is to reduce the efforts of creation and modification of FPGA implementation of Artificial Intelligent projects and provide such solutions as a service. Task: to analyze the possibilities of heterogeneous computing for the implementation of AI projects; analyze advanced FPGA technologies and accelerator cards that allow the organization of a service; analyze the languages, frameworks, and integrated environments for the creation of Artificial Intelligence projects for FPGA implementation; propose a technique for modifiable FPGA project prototyping to ensure a long period of compatibility with integrated environments and target devices; propose a technique for the prototyping of FPGA services with high performance to improve the efficiency of FPGA based AI projects; propose a sequence of optimization of neural networks for FPGA implementation; and provide an example of the practical implementation of the research results. According to the tasks, the following results were obtained. Analysis of the biggest companies and vendors of FPGA technology is performed. Existing heterogeneous technologies and potential non-electronic mediums for AI computations are discussed. FPGA accelerator cards with a large amount of High Bandwidth Memory (HBM) on the same chip package for implementation of AI projects are analyzed and compared. Languages, frameworks, and technologies as well as the capabilities of libraries and integrated environments for prototyping of FPGA projects for the AI applications are analyzed in detail. The sequence of prototyping of FPGA projects that are stable to changes in the environment is proposed. The sequence of prototyping of highly efficient pipelined projects for data processing is proposed. The steps of optimization of neural networks for FPGA implementation of AI applications are provided. An example of practical use of the results of research, including the use of sequences is provided. Conclusions. One of the main contributions of this research is the proposed method of creation of FPGA based implementation of AI projects in the form of services. Proposed sequence of neural network optimization for FPGA allows the reduction of the complexity of the initial program model by more than five times for hardware implementation depending on the required accuracy. The described solutions allow the construction of completely scalable and modifiable FPGA implementations of AI projects to provide it as a service. [ABSTRACT FROM AUTHOR]

Convolutional Neural Networks (CNNs) are widely used in various systems, including text resource-constrained embedded systems or IoT devices. In such systems, it is typical to deploy compressed or pruned CNNs, instead of original ones, at the cost of reduced accuracy. Existing CNN pruning techniques have primarily focused on minimizing resource requirements. However, today’s embedded systems are increasingly dynamic in both resource demands and availability. Thus, the previous techniques that only consider given static cases are no longer efficient. In this paper, we propose a novel text multi-phase CNN that enables a text multi-objective exploration of a number of pruning candidates out of a single CNN. In the proposed technique, a CNN can operate in various versions depending on which subsets of weights are used and can be transformed to the one best matches to the given constraint adaptively and efficiently. For that, a CNN is first pruned to the sparsest form; then a set of parameters (sub-network) is additionally supplemented as the phase goes by. As a result, a number of network versions for all different phases can be represented by a single network and they form a pareto solution over the accuracy and resource usage trade-off. In this work, we target CPU-based CNN inference engines as most embedded systems do not have the luxury of specialized text co-processor support such as GPUs or HW accelerators. The proposed technique has been implemented in a publicly available CPU inference engine, Darknet, and its effectiveness has been validated with a popular CNN in terms of design space exploration capability and runtime switchability.

In order to predict the temperature change of Laoshan scenic area in Qingdao more accurately, a new back propagation neural network (BPNN) prediction model is proposed in this study. Temperature change affects our lives in various ways. The challenge that neural networks tend to fall into local optima needs to be addressed to increase the accuracy of temperature prediction. In this research, we used an improved genetic algorithm (GA) to optimize the weights and thresholds of BPNN to solve this problem. The prediction results of BPNN and GA-BPNN were compared, and the prediction results showed that the prediction performance of GA-BPNN was much better. Furthermore, a screening test experiment was conducted using GA-BPNN for multiple classes of meteorological parameters, and a smaller number of parameter sets were identified to simplify the prediction inputs. The values of running time, root mean square error, and mean absolute error of GA-BPNN are better than those of BPNN through the calculation and analysis of evaluation metrics. This study will contribute to a certain extent to improve the accuracy and efficiency of temperature prediction in the Laoshan landscape.

Neural networks are usually trained with different variants of gradient descent-based optimization algorithms such as the stochastic gradient descent or the Adam optimizer. Recent theoretical work states that the critical points (where the gradient of the loss is zero) of two-layer ReLU networks with the square loss are not all local minima. However, in this work, we will explore an algorithm for training two-layer neural networks with ReLU-like activation and the square loss that alternatively finds the critical points of the loss function analytically for one layer while keeping the other layer and the neuron activation pattern fixed. Experiments indicate that this simple algorithm can find deeper optima than stochastic gradient descent or the Adam optimizer, obtaining significantly smaller training loss values on four out of the five real datasets evaluated. Moreover, the method is faster than the gradient descent methods and has virtually no tuning parameters. [ABSTRACT FROM AUTHOR]

The development of computer vision technology as a type of artificial intelligence is increasing rapidly in various fields. This method uses deep learning methods based on artificial neural networks, a well-performed algorithm in multi-parameter analysis. One of the development of computer vision models and algorithms is for a thematic digital image classification, such as environmental analysis. Remote sensing based digital image classification is one of the reliable tools for environmental quality analysis. This study aims to perform neural network optimization for the analysis of the urban environment comfort based on satellite data. The input data used are 4 types of geobiophysical indexes as urban environmental comfort parameters derived from cloud-free annual mosaics Landsat-8 remote sensing satellite data. The results obtained in this study indicate that the 1 hidden layer neural network architecture with 16 neurons for the classification of urban environmental comfort and 10 other land cover classes is quite good. The result of the classification using this optimized artificial neural network shows that the distribution of classes is very uncomfortable which dominates the Greater Jakarta area and its surroundings. For other classes in the study area, some are uncomfortable and rather comfortable. By using this method, we obtained a fast classification training time of 18 seconds for 145 iterations to achieve an RMS Error of 0.01, and has a fairly high classification accuracy overall 89% with a Kappa coefficient of 0.88, while the 2 hidden layer neural network architecture does not succeed in achieving convergence

Carbon emissions play a pivotal role in driving global warming, with commercial and residential sectors ranking as the third-largest source after industrial and transportation sectors. Consequently, establishing a framework for spatial characterizing and predicting carbon emissions from these sectors is vital for effective carbon reduction planning. This study proposes an optimized Backpropagation Neural Network, incorporating a Spatial Weight Matrix, to model the relationship between land use and carbon emissions, taking spatial effects into account. Additionally, an optimized Random Forest model with variable search step lengths is developed for large-scale and multi-class land use prediction. Beijing, China serves as the study site, yielding the following findings: (1) The optimized Backpropagation Neural Network outperforms traditional models, achieving goodness of fit of 0.999 and 0.997 in training and testing datasets, respectively. (2) The optimized Random Forest model, while reducing prediction precision for ecological areas (Root Mean Square Error increases by 4.12 %), enhances overall model performance by 73.95 % in a single iteration. (3) By 2025, commercial and residential carbon emissions in Beijing are projected to reach 8.91 million tons. (4) The study characterizes the spatial patterns of commercial and residential carbon emissions at a 1 km resolution for the next 15 years, revealing statistical features. (5) Expansion of mixed commercial and residential land relies on service-oriented development, positively impacting regional carbon emission intensity control. • A novel carbon emission prediction model based on land-use subdivision is proposed. • Backpropagation Neural Network is optimized by the Spatial Weight Matrix. • Commercial and residential emissions in Beijing over the next 15 years are predicted. • Future commercial & residential emissions are characterized at resolution of 250 m. [ABSTRACT FROM AUTHOR]

Artificial Intelligence (AI) techniques are considered the most advanced approaches for diagnosing faults in power transformers. Dissolved Gas Analysis (DGA) is the conventional approach widely adopted for diagnosing incipient faults in power transformers. The IEC-599 standard Ratio Method is an accurate method that evaluates the DGA. All the classical approaches have limitations because they cannot diagnose all faults accurately. Precisely diagnosing defects in power transformers is a significant challenge due to their extensive quantity and dispersed placement within the power network. To deal with this concern and to improve the reliability and precision of fault diagnosis, different Artificial Intelligence techniques are presented. In this manuscript, an artificial neural network (ANN) is implemented to enhance the accuracy of the Rogers Ratio Method. On the other hand, it should be noted that the complexity of an ANN demands a large amount of storage and computing power. In order to address this issue, an optimization technique is implemented with the objective of maximizing the accuracy and minimizing the architectural complexity of an ANN. All the procedures are simulated using the MATLAB R2023a software. Firstly, the authors choose the most effective classification model by automatically training five classifiers in the Classification Learner app (CLA). After selecting the artificial neural network (ANN) as the sufficient classification model, we trained 30 ANNs with different parameters and determined the 5 models with the best accuracy. We then tested these five ANNs using the Experiment Manager app and ultimately selected the ANN with the best performance. The network structure is determined to consist of three layers, taking into consideration both diagnostic accuracy and computing efficiency. Ultimately, a (100-50-5) layered ANN was selected to optimize its hyperparameters. As a result, following the implementation of the optimization techniques, the suggested ANN exhibited a high level of accuracy, up to 90.7%. The conclusion of the proposed model indicates that the optimization of hyperparameters and the increase in the number of data samples enhance the accuracy while minimizing the complexity of the ANN. The optimized ANN is simulated and tested in MATLAB R2023a—Deep Network Designer, resulting in an accuracy of almost 90%. Moreover, compared to the Rogers Ratio Method, which exhibits an accuracy rate of just 63.3%, this approach successfully addresses the constraints associated with the conventional Rogers Ratio Method. So, the ANN has evolved a supremacy diagnostic method in the realm of power transformer fault diagnosis.

In order to improve the efficiency and accuracy of corroded cracks detection and classification in reinforced concrete, a corroded cracks identification model Steel Corrosion Net (SCNet), based on deep learning Convolutional Neural Network (CNN), is proposed. Crack figures are collected by self-shooting, internet search and corrosion test, then the data set of 39,000 pictures is built by data enhancement. Afterward, a SCNet three-classification neural network model is built and tested using TensorFlow learning framework and Python. The SCNet combines massive initial data with a multi hidden layer neural network framework, and achieves feature learning and accurate classification through model training. According to the training and testing accuracy of the model, the structure and parameters of the SCNet network are optimized. The results of SCNet are compared with those obtained by two traditional testing methods. The results show that the proposed SCNet model achieves a classification accuracy of 96.8%, so it can effectively identify and classify the corroded cracks in reinforced concrete, with high accuracy and measurability. Under harsh condition of noise interference, such as shadows and distortions, the proposed SCNet model shows a relatively stable classification performance compared with two traditional methods. [ABSTRACT FROM AUTHOR]

Neural networks have demonstrated their usefulness for solving complex regression problems in circumstances where alternative methods do not provide satisfactory results. Finding a good neural network model is a time-consuming task that involves searching through a complex multidimensional hyperparameter and weight space in order to find the values that provide optimal convergence. We propose a novel neural network optimizer that leverages the advantages of both an improved evolutionary competitive algorithm and gradient-based backpropagation. The method consists of a modified, hybrid variant of the Imperialist Competitive Algorithm (ICA). We analyze multiple strategies for initialization, assimilation, revolution, and competition, in order to find the combination of ICA steps that provides optimal convergence and enhance the algorithm by incorporating a backpropagation step in the ICA loop, which, together with a self-adaptive hyperparameter adjustment strategy, significantly improves on the original algorithm. The resulting hybrid method is used to optimize a neural network to solve a complex problem in the field of chemical engineering: the synthesis and swelling behavior of the semi- and interpenetrated multicomponent crosslinked structures of hydrogels, with the goal of predicting the yield in a crosslinked polymer and the swelling degree based on several reaction-related input parameters. We show that our approach has better performance than other biologically inspired optimization algorithms and generates regression models capable of making predictions that are better correlated with the desired outputs. [ABSTRACT FROM AUTHOR]

The paper discusses various approaches used to optimize the convolutional neural network (CNN) model with You Only Look Once (YOLO) version 3 architecture. At the same time, an implementation using graphic processing units (GPU) computing is used as a baseline performance estimate. A separate study was conducted on the Intel Core i5-8500 central processing unit (CPU). Both the classical methods of the neural network approach, such as model pruning, and effective methods for processing video data, in particular, the optical flow and speed prediction, are considered. The proposed algorithm made it possible to overclock the GPU computing of the NVIDIA RTX 2080 Super video card by 4 times, up to almost 30 frames per second. However, processing frames using Intel Open Visual Inference and Neural network Optimization (OpenVINO) toolkit allows us to achieve similar performance on the CPU without optical flow and speed extrapolation. In this case, the acceleration reaches almost 30 times. [ABSTRACT FROM AUTHOR]

A strain of Stenotrophomonas acidaminiphila, isolated from fermenting bean-processing wastewater, produced alkaline protease in pretreated cassava waste-stream, but with low yield. Strain improvement by alternate combinatorial random mutagenesis and bioprocess optimization using comparative statistical and neural network methods enhanced yield by 17.8-fold in mutant kGy-04-UV-25. Kinetics of production by selected mutant modeled by logistic and modified Gompertz functions revealed higher specific growth rate in mutant than in the parent strain, likewise volumetric and specific productivities. Purification by PEG/Na+ citrate aqueous two-phase system recovered 73.87% yield and 52.55-fold of protease. Its activity was stable at 5–35% NaCl, 45–75°C, and was significantly enhanced by 1–15 mM sodium dodecyl sulfate (SDS). The protease was inhibited by low concentrations of phenyl-methyl-sulfonyl fluoride but was activated by 1–5 mM Mn2+ suggesting a manganese-dependent serine‑protease. The 45.7 kDa thermo-halo-stable alkaline protease demonstrated keratinolytic and blood-stain removal potentials showing prospects in textile and detergent industries, respectively.

Through the stability evaluation of a slope, a landslide geological disaster can be identified, and the safety and risk control of a project can be ensured. This work proposes an improved sparrow search algorithm to optimize the slope safety factor prediction model (ISSA–BP) of a BP neural network, through an improvement in two aspects: introducing dynamic weight factors and reverse learning strategies to realize adaptive searches. The optimal value improves a defect in the traditional model, preventing it from easily falling into the local minimum. First, combined with 352 sets of actual slope data, three machine learning models were used to predict the safety factor of the slope. Then, the accuracy index was used for evaluation. Compared with other models, the MAPE, RMSE, and R2 of the ISSA-BP model were 1.64%, 0.0296, and 0.99, respectively, and the error was reduced by 78% compared with the BP neural network, showing better accuracy. Finally, the three models were applied to the slope stability analysis of Tianbao Port in Wenshan Prefecture. The research shows that the predicted value of the ISSA–BP model was the closest to the actual safety factor, which verified the experimental results. The improved ISSA–BP model can effectively predict the safety factor of slopes under different conditions, and it provides a new technology for slope disaster warning and control.

Motivation, BRCA1 and BRCA2 are genes with tumor suppressor activity. They are involved in a considerable number of biological processes. To help the biologist in tumor classification, we developed a deep learning algorithm. The question when we want to construct a neural network is how many hidden layers and neurons should we use. If the number of inputs and outputs is defined by the problem, the number of hidden layers and neurons is difficult to define. Hidden layers and neurons that make up each layer of the neural network influence the performance of system predictions. There are different methods for finding the optimal architecture. In this paper, we present the two packages that we have developed, the genetic algorithm (GA) and the particle swarm optimization (PSO) to optimize the parameters of the neural network for predicting BRCA1 and BRCA2 pathogenicity; Results, we will compare the results obtained by the two algorithms. We used datasets collected from our NGS analysis of BRCA1 and BRCA2 genes to train deep learning models. It represents a data collection of 11,875 BRCA1 and BRCA2 variants. Our preliminary results show that the PSO provided the most significant architecture of hidden layers and the number of neurons compared to grid search and GA; Conclusions, the optimal architecture found by the PSO algorithm is composed of 6 hidden layers with 275 hidden nodes with an accuracy of 0.98, precision 0.99, recall 0.98, and a specificity of 0.99. [ABSTRACT FROM AUTHOR]

The outstanding properties of metal-organic frameworks (MOFs) have proven that this type of crystalline adsorbent has great potential in CO 2 capture applications. Most of the MOF research studies on new functional MOFs are conducted to improve the performance of CO 2 gas adsorption. Combined studies of material evaluation and process design on engineering issues in CO 2 capture applications in industry are rarely carried out. In this study, the authors attempted to address engineering issues by developing a biometal-organic framework with the bioligand L -glutamic acid that has more practical fabrication cost than petrochemical MOFs. Herein, the demonstration of the prediction and optimization of CO 2 adsorption capacity, selectivity, and heat of adsorption using a multiobjective genetic algorithm (MOGA) combined with an artificial neural network (ANN). The success of the Bio-MOF fabrication was evaluated by scanning electron microscopy, N 2 adsorption-desorption isotherm analysis, thermal gravimetric analysis, X-ray diffraction, and Fourier transform infrared spectroscopy techniques. Furthermore, volumetric measurements were performed at several temperatures (27, 35 and, 50 °C). The isosteric heat of adsorption was then evaluated by an indirect method with the Clausius-Clapeyron (C-C) and Chakraborty, Saha, and Koyama (CSK) equations. Then, CO 2 /N 2 selectivity was analysed by IAST techniques by regressing the experimental data with the Langmuir-Freundlich isothermal equation. The computational study by ANN and MOGA also gives satisfying results in balancing three requirements criteria. Thus, this study paved the way for the development of low-cost scalable MOF fabrication in industry by applying the optimization and balancing principles of the three objective functions. [ABSTRACT FROM AUTHOR]

Given article is devoted features of the decision of a problem of integer nonlinear programming, by means of developed neural network method and algorithm of nonlinear optimization of means «decision Search» tabular processor Microsoft Excel. In offered neural network method the task in view decision is made by means of a recurrent neural network (RNN) matrix architecture with m neurons in each line and n neurons in each column. All neurons such network are connected with each other by communications, and the signal from an exit neuron can move on its input. Neural network method is characterized by that on inputs mentioned RNN the entrance vector of values of parameters of optimized nonlinear criterion function of a problem of distribution of a non-uniform resource moves, calculation of values of weight factors connected among themselves neurons is carried out and signal RNN is formed. This signal by means of nonlinear function will be transformed to the discrete target signal characterizing values quasi-optimal of the decision of the mentioned problem which size changes from 0 to 1. The estimation of efficiency of the decision of a considered problem was carried out at its various values of an indicator of efficiency on the basis of developed imitating model RNN. As indicators of efficiency of application offered neural network method were used – an average relative error and time of the decision of a problem. The value received by means of algorithm of nonlinear optimization of means was accepted to the exact decision «decision Search» tabular processor Microsoft Excel. The analysis of the received results of the experimental researches, offered neural network method, has allowed to make the conclusion that in comparison with an existing method of nonlinear optimization of tabular processor Microsoft Excel use offered neural network method allows essentially (in 9,4 times) to lower time of the decision of a problem dimension 10 × 8 (m × n) and thus to provide accuracy of its decision not less than 99,8 %.

With the development of machine learning and artificial intelligence (ML/AI) models, data-driven soft sensors, especially the neural network-based, have widespread utilization for the prediction of key water quality indicators in wastewater treatment plants (WWTPs). However, recent research indicates that the prediction performance and computational efficiency are greatly compromised due to the time-varying, nonlinear and high-dimensional nature of the wastewater treatment process. This paper proposes a neural network-based soft sensor with double-errors parallel optimization to achieve more accurate prediction for effluent variables timely. Firstly, relying on the Activity Based Classification (ABC) principle, an ensemble variable selection method that combines Pearson correlation coefficient (PCC) and mutual information (MI) is introduced to select the optimal process variables as auxiliary variables, thereby reducing the data dimensionality and simplifying the model complexity. Subsequently, a double-errors parallel optimization methodology with minimizing both point prediction error and distribution error simultaneously is proposed, aiming to enhancing the training efficiency and the fitting quality of neural networks. Finally, the effectiveness is quantitatively assessed in two datasets collected from the Benchmark Simulation Model no. 1 (BMS1) and an actual oxidation ditch WWTP. The experimental results illustrate that the proposed soft sensor achieves precise effluent variable prediction, with RMSE, MAE and R2 values being 0.0606, 0.0486, 0.99930, and 0.06939, 0.05381, 0.98040, respectively. Consequently, this soft sensor can expedite the convergence speed in the neural network training process and enhance the prediction performance, thereby contributing to the effective optimization management of WWTPs. The fundamental architecture of the proposed soft sensor. [Display omitted] • PCC and MI are combined to formulate an ensemble variable selection method. • A double-errors parallel optimization methodology is proposed for model training. • The proposed neural network-based soft sensor is applied in two WWTPs. [ABSTRACT FROM AUTHOR]

Neural networks are usually trained with different variants of gradient descent-based optimization algorithms such as the stochastic gradient descent or the Adam optimizer. Recent theoretical work states that the critical points (where the gradient of the loss is zero) of two-layer ReLU networks with the square loss are not all local minima. However, in this work, we will explore an algorithm for training two-layer neural networks with ReLU-like activation and the square loss that alternatively finds the critical points of the loss function analytically for one layer while keeping the other layer and the neuron activation pattern fixed. Experiments indicate that this simple algorithm can find deeper optima than stochastic gradient descent or the Adam optimizer, obtaining significantly smaller training loss values on four out of the five real datasets evaluated. Moreover, the method is faster than the gradient descent methods and has virtually no tuning parameters.

Neural networks have demonstrated their usefulness for solving complex regression problems in circumstances where alternative methods do not provide satisfactory results. Finding a good neural network model is a time-consuming task that involves searching through a complex multidimensional hyperparameter and weight space in order to find the values that provide optimal convergence. We propose a novel neural network optimizer that leverages the advantages of both an improved evolutionary competitive algorithm and gradient-based backpropagation. The method consists of a modified, hybrid variant of the Imperialist Competitive Algorithm (ICA). We analyze multiple strategies for initialization, assimilation, revolution, and competition, in order to find the combination of ICA steps that provides optimal convergence and enhance the algorithm by incorporating a backpropagation step in the ICA loop, which, together with a self-adaptive hyperparameter adjustment strategy, significantly improves on the original algorithm. The resulting hybrid method is used to optimize a neural network to solve a complex problem in the field of chemical engineering: the synthesis and swelling behavior of the semi- and interpenetrated multicomponent crosslinked structures of hydrogels, with the goal of predicting the yield in a crosslinked polymer and the swelling degree based on several reaction-related input parameters. We show that our approach has better performance than other biologically inspired optimization algorithms and generates regression models capable of making predictions that are better correlated with the desired outputs.

Neural Architecture Search (NAS) methods are widely employed to address the time-consuming and costly challenges associated with manual operation and design of deep convolutional neural networks (DCNNs). Nonetheless, prevailing methods still encounter several pressing obstacles, including limited network architecture design, excessively lengthy search periods, and insufficient utilization of the search space. In light of these concerns, this study proposes an optimization strategy for residual networks that leverages an enhanced Particle swarm optimization algorithm. Primarily, low-complexity residual architecture block is employed as the foundational unit for architecture exploration, facilitating a more diverse investigation into network architectures while minimizing parameters. Additionally, we employ a depth initialization strategy to confine the search space within a reasonable range, thereby mitigating unnecessary particle exploration. Lastly, we present a novel approach for computing particle differences and updating velocity mechanisms to enhance the exploration of updated trajectories. This method significantly contributes to the improved utilization of the search space and the augmentation of particle diversity. Moreover, we constructed a crime-dataset comprising 13 classes to assess the effectiveness of the proposed algorithm. Experimental results demonstrate that our algorithm can design lightweight networks with superior classification performance on both benchmark datasets and the crime-dataset. [ABSTRACT FROM AUTHOR]

• An innovative multi-generation system has been designed for liquid ammonia and hydrogen production. • Comparative examination of energy carriers (Hydrogen and Ammonia) is carried out. • Sensitivity analysis and Machine learning optimization are applied to the system. • Optimal liquid hydrogen and ammonia production rate and exergy efficiency are 0.026 kg/s, 0.151 and 41.24 %. • Steam Rankine cycle and the elecrolyzer have the highest exergy destructions. Power-to-fuel energy systems have an essential need for decarbonizing energy carriers. This study proposes a solar and wind energy-based system for producing liquid hydrogen and ammonia as energy carriers. The proposed system caters to urban requirements encompassing electricity, cooling, heating, and freshwater. Three different scenarios, only liquid hydrogen, only ammonia, and dual production, are considered and compared. An artificial neural network is employed for prediction and computational time reduction, coupled with a genetic algorithm for system optimization. For a case in which 40 % of the net power is used in the electrolyzer, and the system simultaneously produces liquid hydrogen and ammonia, the overall energy and exergy performance of the system are 56.78 % and 44.69 %, respectively. In this case, the system has the capability of producing 13.2 MW of net power, 0.057 kg/s of liquid hydrogen, 0.162 kg/s of cold ammonia, 3.63 kg/s of freshwater, and 3.4 MW cooling load. The exergy analysis of the system reveals that the Rankine cycle and the electrolyzer have the highest exergy destruction, at 27 % and 21 %, respectively. Using the Technique for Order Preference by Similarity to Ideal Solution method, the optimal liquid hydrogen rate, ammonia production rate, overall exergy efficiency, and net power are determined as 0.026 kg/s, 0.151 kg/s, 51.38 %, and 16.52 MW respectively are recognized as optimal values for dual production scenario. [ABSTRACT FROM AUTHOR]

Summary: At present, the application of communication network has spread to every area of social life. The progress of network technology has driven the development of information technology industry and the advancement of cloud computing. With the application of artificial intelligence and deep learning technology, the network becomes more and more intelligent, so the application of artificial intelligence theory in communication network optimization modeling is more extensive. In this paper, we propose the ubiquitous clouds framework and apply the artificial intelligence optimization scheme to the point‐to‐point (P2P) signal transmission network optimization scheme. For different application scenarios, this paper analyzes and experiments the proposed method. The experimental results show that the proposed method makes full use of the advantages of communication network resources which improves the efficiency of communication network optimization and also reduces the optimization cost to a certain extent compared with the state‐of‐the‐art approaches. [ABSTRACT FROM AUTHOR]

Political optimizer (PO) is a recently proposed human-behavior inspired meta-heuristic, which has shown tremendous performance on complex multimodal functions as well as engineering optimization problems. Good convergence speed and well-balanced exploratory and exploitative behavior of PO convince us to employ PO for the training of feedforward neural network (FNN). The FNN-training problem is formulated as an optimization problem in which the objective is to minimize the mean squared error (MSE) or cross entropy (CE). The weights and biases of the FNN are arranged in the form of a vector called a candidate solution. The performance of the proposed trainer is evaluated on 5 classification data-sets and 5 function-approximation data-sets, which have already been used in the literature. In recent years, grey wolf optimizer, moth flame optimization, multi-verse optimizer, sine-cosine algorithm, whale optimization algorithm, ant lion optimizer, and Salp swarm algorithm have successfully been applied on neural network training. In this paper, we compare the performance of PO with these algorithms and show that PO either outperforms them or performs equivalently. The MSE, CE, training set accuracy, and test set accuracy are used as metrics for the comparative analysis. The non-parametric Wilcoxon's rank-sum test is used to show the statistical significance of the results. Based on the tremendous performance, we highly recommend using PO for the training of artificial neural networks to solve the classification and regression problems. [ABSTRACT FROM AUTHOR]

Lately, robot swarm has widely employed in many applications like search and rescue missions, fire forest detection and navigation in hazard environments. Each robot in a swarm is supposed to move without collision and avoid obstacles while performing the assigned job. Therefore, a formation control is required to achieve the robot swarm three tasks. In this article, we introduce a decentralized formation control algorithm based on the potential field method for robot swarm. Our formation control algorithm is proposed to achieve the three tasks: avoid obstacles in the environment, keep a fixed distance among robots to maintain a formation and perform an assigned task. An artificial neural network is engaged in the online optimization of the parameters of the potential force. Then, real-time experiments are conducted to confirm the reliability and applicability of our proposed decentralized formation control algorithm. The real-time experiment results prove that the proposed decentralized formation control algorithm enables the swarm to avoid obstacles and maintain formation while performing a certain task. The swarm manages to reach a certain goal and tracks a given trajectory. Moreover, the proposed decentralized formation control algorithm enables the swarm to escape from local minima, to pass through two narrow placed obstacles without oscillation near them. From a comparison between the proposed decentralized formation control algorithm and the traditional PFM, we obtained that NN-swarm successes to reach its goal with average accuracy 0.14 m compared to 0.22 m for the T-swarm. The NN-swarm also keeps a fixed distance between robots with a higher swarming error reaches 34.83%, while the T-swarm reaches 23.59%. Also, the NN-swarm is more accurate in tracking a trajectory with a higher tracking error reaches 0.0086 m compared to min. error of T-swarm equals to 0.01 m. Besides, the NN-swarm maintains formation much longer than T-swarm while tracking trajectory reaches 94.31% while the T-swarm reaches 81.07% from the execution time, in environments with different numbers of obstacles. [ABSTRACT FROM AUTHOR]

Biometrics such as facial features, fingerprint, and iris are being used increasingly in modern authentication systems. These methods are now popular and have found their way into many portable electronics such as smartphones, tablets, and laptops. Furthermore, the use of biometrics enables secure access to private medical data, now collected in wearable devices such as smartwatches. In this work, we present an accurate low-power device authentication system that employs electrocardiogram (ECG) signals as the biometric modality. The proposed ECG processor consists of front-end signal processing of ECG signals and back-end neural networks (NNs) for accurate authentication. The NNs are trained using a cost function that minimizes intra-individual distance over time and maximizes inter-individual distance. Efficient low-power hardware was implemented by using fixed coefficients for ECG signal pre-processing and by using joint optimization of low-precision and structured sparsity for the NNs. We implemented two instances of ECG authentication hardware with 4X and 8X structurally-compressed NNs in 65 nm LP CMOS, which consume low power of 62.37  $\mu$ W and 75.41  $\mu$ W for real-time ECG authentication with a low equal error rate of 1.36% and 1.21%, respectively, for a large 741-subject in-house ECG database. The hardware was evaluated at 10 kHz clock frequency and 1.2 V voltage supply. [ABSTRACT FROM AUTHOR]

Style transfer is a task of migrating a style from one image to another. Recently, Full Convolutional Network (FCN) is adopted to create stylized images and make it possible to perform style transfer in real-time on advanced GPUs. However, problems are still existing in memory usage and time-consumption when processing high-resolution images. In this work, we analyze the architecture of the style transfer network and divide it into three parts: feature extraction, style transfer, and image reconstruction. And a novel way is proposed to accelerate the style transfer operation and reduce the memory usage at run-time by conducting the super-resolution style transfer network (SRSTN), which can generate super-resolution stylized images. Compared with other style transfer networks, SRSTN can produce competitive quality resulting images with a faster speed as well as less memory usage. [ABSTRACT FROM AUTHOR]

In the current industrial landscape, increasingly pervaded by technological innovations, the adoption of optimized strategies for asset management is becoming a critical key success factor. Among the various strategies available, the “Prognostics and Health Management” strategy is able to support maintenance management decisions more accurately, through continuous monitoring of equipment health and “Remaining Useful Life” forecasting. In the present study, convolutional neural network-based deep neural network techniques are investigated for the remaining useful life prediction of a punch tool, whose degradation is caused by working surface deformations during the machining process. Surface deformation is determined using a 3D scanning sensor capable of returning point clouds with micrometric accuracy during the operation of the punching machine, avoiding both downtime and human intervention. The 3D point clouds thus obtained are transformed into bidimensional image-type maps, i.e., maps of depths and normal vectors, to fully exploit the potential of convolutional neural networks for extracting features. Such maps are then processed by comparing 15 genetically optimized architectures with the transfer learning of 19 pretrained models, using a classic machine learning approach, i.e., support vector regression, as a benchmark. The achieved results clearly show that, in this specific case, optimized architectures provide performance far superior (MAPE = 0.058) to that of transfer learning, which, instead, remains at a lower or slightly higher level (MAPE = 0.416) than support vector regression (MAPE = 0.857).