Search

Photonic neuromorphic computing is raising a growing interest as it promises to provide massive parallelism and low power consumption. In this paper, we demonstrate for the first time a feed-forward neural network via an 8 × 8 Indium Phosphide cross-connect chip, where up to 8 on-chip weighted addition circuits are co-integrated, based on semiconductor optical amplifier technology. We perform the weight calibration per neuron, resulting in a normalized root mean square error smaller than 0.08 and a best case dynamic range of 27 dB. The 4 input to 1 output weighted addition operation is executed on-chip and is part of a neuron, whose non-linear function is implemented via software. A three feedback loop optimization procedure is demonstrated to enable an output neuron accuracy improvement of up to 55%. The exploitation of this technology as neural network is evaluated by implementing a trained 3-layer photonic deep neural network to solve the Iris flower classification problem. Prediction accuracy of 85.8% is achieved, with respect to the 95% accuracy obtained via a computer. A comprehensive analysis of the error evolution in our system reveals that the electrical/optical conversions dominate the error contribution, which suggests that an all optical approach is preferable for future neuromorphic computing hardware design.

Load forecasting is a pivotal part of the power utility companies. To provide load-shedding free and uninterrupted power to the consumer, decision-makers in the utility sector must forecast the future demand for electricity with a minimum error percentage. Load prediction with less percentage of error can save millions of dollars to the utility companies. There are numerous Machine Learning (ML) techniques to amicably forecast electricity demand, among which the hybrid models show the best result. Two or more than two predictive models are amalgamated to design a hybrid model, each of which provides improved performances by the merit of individual algorithms. This paper reviews the current state-of-the-art of electric load forecasting technologies and presents recent works pertaining to the combination of different ML algorithms into two or more methods for the construction of hybrid models. A comprehensive study of each single and multiple load forecasting model is performed with an in-depth analysis of their advantages, disadvantages, and functions. A comparison between their performance in terms of Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and Mean Absolute Percentage Error (MAPE) values are developed with pertinent literature of several models to aid the researchers with the selection of suitable models for load prediction.

Chest X-ray (CXR) interpretations are conducted in hospitals and medical facilities on daily basis. If the interpretation tasks were performed correctly, various vital medical conditions of patients can be revealed such as pneumonia, pneumothorax, interstitial lung disease, heart failure and bone fracture. The current practices often involve tedious manual processes dependent on the expertise of radiologist or consultant, thus, the execution is easily prone to human errors of being misdiagnosed. With the recent advances of deep learning and increased hardware computational power, researchers are working on various networks and algorithms to develop machines learning that can assists radiologists in their diagnosis and reduce the probability of misdiagnosis. This paper presents a review of deep learning advancements made in the field of chest radiography. It discusses single and multi-level localization and segmentation techniques adopted by researchers for higher accuracy and precision.

Although a number of fault diagnosis algorithms for inertial sensors have been proposed in previous decades, the performance of these algorithms needs to be improved with regard to small faults. In this paper, we introduce a data driven-based algorithm, namely, SaPD, for the anomaly detection and output reconstruction of a redundant inertial measurement unit (RIMU). SaPD implements the fault identification of an inertial apparatus by combining an artificial neural network with the Q contribution plots method in parity space. To improve the performance of the fault detection part, in particular for small faults, we introduce a novel hyperplane that measures the distances between inputs and the primary-neuron set obtained from a self-organizing incremental neural network (SOINN). We also employ the Q contribution plots of sensors in the fault isolation part by analyzing historical data with principal component analysis (PCA). We perform quantitative evaluations in a realistic simulation environment, which demonstrates that the proposed SaPD algorithm outperforms other related algorithms in terms of the fault identification accuracy of tiny faults with an acceptable computational complexity.

This paper presents a new method for extracting 3D point clouds from multi-focus images of a fibrous web acquired on an optical microscope to analyze microscopic structures of a fibrous web. The algorithm consists of two major parts: (1) utilizing a convolutional neural network (CNN) to extract in-focus objects from multi-focus images, and (2) a depth identification module (DIM) which is a frequency domain-based model used to identify the depths of object points. The network, namely the multi-focus image deblurring network (MIDN), was designed by introducing gradient features into the network to deblur images and generate the ranges of focal depths of object points. Based on the results of MIDN, DIM was constructed to calculates the focal plane depth for each point. The experiments show that the combination of MIDN and DIM provides a practical way to generate complete, accurate 3D structures of nonwoven.

In this paper, we propose a novel machine learning pipeline to detect QRS complexes in very noisy wearable electrocardiogram (ECG) devices. The machine learning pipeline consists of a Butterworth filter, two wavelet convolutional neural networks (WaveletCNNs) autoencoders, an optional QRS complex inverter, a Monte Carlo k-nearest neighbours (k-NN), and a convolutional long short-term memory (ConvLSTM). WaveletCNN autoencoders filter out electrode contact noise, instrumentation noise, and motion artifact noise by using the advantages of wavelet filters and convolutional neural networks. The QRS complex inverter flips inverted QRS complexes. Monte Carlo k-NN performs automatic gain control on the ECG signals in order to normalize it. The ConvLSTM executes the final QRS complex detection by using the power of a convolutional neural network and a long short-term memory. The MIT-BIH, the European ST-T, and the Long Term ST database Noise Stress Test databases provide the training and testing ECG recordings. The proposed machine learning pipeline performs 3 standard deviations better than the state of the art QRS complex detection algorithms in terms of F1 score for very noisy environments.

In this paper, we propose a promising nonlinear noise power estimation method based on correlation functions using artificial neural networks (ANN), which is robust against both amplifier spontaneous emission noise and the symbol patterns. Error vector correlation (EVC), together with the amplitude noise correlation (ANC) and the phase noise correlation (PNC), is used as the input of ANN in the proposed method. 378 cases of 224 Gb/s polarization-multiplexed 16-quadrature amplitude modulated (PM-16-QAM) signal transmitted over a wide range of conditions by varying launch power, OSNR and transmission distance are used to train and test the ANN. With the launch power varying from 0 to 8 dBm and the transmission distance as long as 2400 km, the results of tested samples demonstrate that the maximum absolute deviation (MAD), mean absolute error (MAE) and root mean square error (RMSE) of the estimated nonlinear noise power are 0.65 dB, 0.20 dB and 0.27 dB, respectively. In order to verify the independence of symbol patterns, 150 new cases with 30 pseudo-random binary sequence (PRBS) seeds are used to test the trained ANN. The results show that the MAE, MAD and RMSE of estimated nonlinear noise power are 0.86 dB, 0.25 dB and 0.33 dB respectively, meaning that the trained ANN is valid even if the test samples are not covered by the trained process. The results in this work verified that the ANN with EVC, ANC and PNC as input can make the trained model feasible to accurately estimate the nonlinear noise power in high speed optical coherent communication systems.

Text correction systems (e.g., spell checkers) have been used to improve the quality of computerized text by detecting and correcting errors. However, the task of performing spelling correction and word normalization (text correction) for Thai social media text has remained largely unexplored. In this paper, we investigated how current text correction systems perform on correcting errors and word variances in Thai social texts and propose a method designed for this task. We have found that currently available Thai text correction systems are insufficiently robust for correcting spelling errors and word variances, while the text correctors designed for English grammatical error correction suffer from overcorrections (text rewrites). Thus, we proposed a neural-based text corrector with a two-stage structure to alleviate issues of overcorrections while exploiting the benefits of a neural Seq2Seq corrector. Our method consists of a neural-based error detector and a Seq2Seq neural error corrector with contextual attention. This novel architecture allows the Seq2Seq network to produce corrections based on both the erroneous text and its context without the need for an end-to-end structure. Our method outperformed all the other evaluated text correction systems. When compared to the second-best result (copy-augmented transformer), our method further reduced the word error rate (WER) from 2.51% to 2.07%, improved the generalized language evaluation understanding (GLEU) score from 0.9409 to 0.9502 on the Thai text correction task, and improved the GLEU score from 0.7409 to 0.7539 on the English spelling correction task.

International audience; Machine learning is yielding unprecedented interest in research and industry, due to recent success in many applied contexts such as image classification and object recognition. However, the deployment of these systems requires huge computing capabilities, thus making them unsuitable for embedded systems. To deal with this limitation, many researchers are investigating brain-inspired computing, which would be a perfect alternative to the conventional Von Neumann architecture based computers (CPU/GPU) that meet the requirements for computing performance, but not for energy-efficiency. Therefore, neuromorphic hardware circuits that are adaptable for both parallel and distributed computations need to be designed. In this paper, we focus on Spiking Neural Networks (SNNs) with a comprehensive study of information coding methods and hardware exploration. In this context, we propose a framework for neuromorphic hardware design space exploration, which allows to define a suitable architecture based on application-specific constraints and starting from a wide variety of possible architectural choices. For this framework, we have developed a behavioral level simulator for neuromorphic hardware architectural exploration named NAXT. Moreover, we propose modified versions of the standard Rate Coding technique to make trade-offs with the Time Coding paradigm, which is characterized by the low number of spikes propagating in the network. Thus, we are able to reduce the number of spikes while keeping the same neuron’s model, which results in an SNN with fewer events to process. By doing so, we seek to reduce the amount of power consumed by the hardware. Furthermore, we present three neuromorphic hardware architectures in order to quantitatively study the implementation of SNNs. One of these architectures integrates a novel hybrid structure: a highly-parallel computation core for most solicited layers, and time-multiplexed computation units for deeper layers. These architectures are derived from a novel funnel-like Design Space Exploration framework for neuromorphic hardware.

Prediction models are used for knowing the behavior of highly related complex data. The prediction of morphological structures, and especially the mandible from cranio-maxillary variables, has clinical and investigative odontological usefulness. For example, in cases of trauma, pathologies and in forensic sciences, especially when it is necessary to¬ individualize a missing person, using facial reconstruction. The aim of this paper is to predict mandibular morphology through artificial neuronal networks, using cranio-maxillary measures in posterior-anterior radiographs.

This paper analyzes the potential of Artificial Neural Networks (ANNs) for the modeling and optimization of magnetic components and, specifically, inductors. After reviewing the basic properties of ANNs, several potential modeling and design workflows are presented. A hybrid method, which combines the accuracy of 3D Finite Element Method (FEM) and the low computational cost of ANNs, is selected and implemented. All relevant effects are considered (3D magnetic and thermal field patterns, detailed core loss data, winding proximity losses, coupled loss-thermal model, etc.) and the implemented model is extremely versatile (30 input and 40 output variables). The proposed ANN-based model can compute 50'000 designs per second with less than 3% deviation with respect to 3D FEM simulations. Finally, the inductor of a 2 kW DC-DC buck converter is optimized with the ANN-based workflow. From the Pareto fronts, a design is selected, measured, and successfully compared with the results obtained with the ANNs. The implementation (source code and data) of the proposed workflow is available under an open-source license., IEEE Open Journal of Power Electronics, 1, ISSN:2644-1314

In this paper, novel computing paradigm by exploiting the strength of feed-forward artificial neural networks (ANNs) with Levenberg-Marquardt Method (LMM), and Bayesian Regularization Method (BRM) based backpropagation is presented to find the solutions of initial value problems (IVBs) of linear/nonlinear pantograph delay differential equations (LP/NP-DDEs). The dataset for training, testing and validation is created with reference to known standard solutions of LP/NP-DDEs. ANNs are implemented using the said dataset for approximate modeling of the system on mean squared error based merit functions, while learning of the adjustable parameters is conducted with efficacy of LMM (ANN-LMM) and BRMs (ANN-BRM). The performance of the designed algorithms ANN-LMM and ANN-BRM on IVPs of first, second and third order NP-FDEs are verified by attaining a good agreement with the available solutions having accuracy in the range from 10-5 to 10-8 and are further endorsed through error histograms and regression measures.

In this paper, we propose a Generative Translation Classification Network (GTCN) for improving visual classification accuracy in settings where classes are visually similar and data is scarce. For this purpose, we propose joint learning from a scratch to train a classifier and a generative stochastic translation network end-to-end. The translation network is used to perform on-line data augmentation across classes, whereas previous works have mostly involved domain adaptation. To help the model further benefit from this data-augmentation, we introduce an adaptive fade-in loss and a quadruplet loss. We perform experiments on multiple datasets to demonstrate the proposed method's performance in varied settings. Of particular interest, training on 40% of the dataset is enough for our model to surpass the performance of baselines trained on the full dataset. When our architecture is trained on the full dataset, we achieve comparable performance with state-of-the-art methods despite using a light-weight architecture., Comment: 14 pages, 17 figures, 13 tables

International audience; The presence of nonlinearities such as hysteresis and creep increases the difficulty in the dynamic modeling and control of piezoelectric micromanipulators, in spite of the fact that the application of such devices requires high accuracy. Moreover, sensing in the microscale is expensive, making model feedback the only viable option. On the other hand, data-driven dynamic models are powerful tools within system identification that may be employed to construct models for a given plant. Recently, considerable effort has been devoted in extending the huge success of deep learning models to the identification of dynamic systems. In the present paper, we present the results of the successful application of deep learning based black-boxmodels for characterizing the dynamic behavior of micromanipulators. The excitation signal is a multisine spanning the frequency band of interest and the selected model is validated with semi static individual sinusoidal curves. Various architectures are tested to achieve a reasonable result and we try to summarize the best approach for the fine tuning required for such application. The results indicate the usefulness and predictive power for deep learning based models inthe field of system identification and in particular hysteresis modeling and compensation in micromanipulation applications.

Current encryption systems are run with a hybrid mode in which symmetric and asymmetric key methods are mixed. This hybrid mode is devised to employ the fast processing speed of the symmetric key while circumventing the difficulty of providing services due to computational complexity of the asymmetric key method by limiting its use to secured key exchange. To implement a secured symmetric key-based cryptographic system to circumvent the crack problem due to the emergence of quantum computers in asymmetric key exchanges, the problem of sharing a key transferred in the most secure manner should be solved while symmetric keys in use are kept up-to-date. This paper proposes a three-dimensional (3D) cube algorithm that can be used by creating the up-to-date key in a symmetric key encryption system to provide security resistance in a quantum computer environment. More specifically, it presents a solution to the secured key sharing method that can minimize the damage due to the pre-shared key (PSK) leakage. This is done by using the method of inducing the symmetric key creation based on deep neural network learning without sharing the PSK between systems to securely maintain the key while creating and using the key that is variably used through the symmetric key encryption system of the 3D cube algorithm. Thus, the proposed algorithm secures confidentiality and integrity of data transferred over a network because information that can be obtained by malicious attackers is small (because the symmetric key used in encryption and decryption is induced without exchanging the PSK with the application of deep neural network learning).

International audience; Unlabelled - The fact that a disproportionate share of the disease burden from COVID-19, including mortality, has been borne by racial and ethnic minority communities is well documented. The purpose of this paper is to reassess the "race/ethnicity effect" in COVID-19 mortality in the United States. Using an ecological regression framework and county-level data, the study aims to make two contributions. First, it estimates the race/ethnicity effect for major racial/ethnic groups at important junctures during the first year of the pandemic. In doing so, it seeks to provide the fullest possible picture of the nature and evolution of the race/ethnicity effect. Second, it estimates the race/ethnicity effect net of basic socioeconomic factors (SEF). This helps to identify the likely mechanisms through which the race/ethnicity effect operates. Racial/ethnic composition is flexibly measured in two ways-by percentage contributions to county population and by indicators of group plurality. The ecological regressions revealed a positive association between the size of three racial/ethnic groups-non-Hispanic Blacks, non-Hispanic American Indian and Alaskan Natives (AIAN) and Hispanics-and county mortality, although the association was stronger and more consistent for Blacks and AIANs. Furthermore, accounting for basic SEF had different impacts on the race/ethnicity-mortality association for the three groups. For Hispanics, it was almost fully mediated. For Blacks, it decreased but remained statistically significant [62-6% higher mortality associated with a 1-standard deviation increase in Black share of county population; 2.3-1.1 times higher mortality in the average Black plurality county]. For AIANs, it was largely unaffected or even increased [44-10% higher mortality associated with a 1-standard deviation increase in AIAN share; 6.2-1.8 times higher mortality in AIAN plurality county). For all groups, the race/ethnicity effect generally decreased as the pandemic wore on during the first year. Supplementary information - The online version contains supplementary material available at 10.1007/s44155-022-00019-9.

International audience; This paper introduces a new efficient autoprecoder (AP) based deep learning approach for massive multiple-input multiple-output (mMIMO) downlink systems in which the base station is equipped with a large number of antennas with energy-efficient power amplifiers (PAs) and serves multiple user terminals. We present AP-mMIMO, a new method that jointly eliminates the multiuser interference and compensates the severe nonlinear (NL) PA distortions. Unlike previous works, AP-mMIMO has a low computational complexity, making it suitable for a global energy-efficient system. Specifically, we aim to design the PA-aware precoder and the receive decoder by leveraging the concept of autoprecoder, whereas the end-to-end massive multiuser (MU)-MIMO downlink is designed using a deep neural network (NN). Most importantly, the proposed AP-mMIMO is suited for the varying block fading channel scenario. To deal with such scenarios, we consider a two-stage precoding scheme: 1) a NN-precoder is used to address the PA non-linearities and 2) a linear precoder is used to suppress the multiuser interference. The NN-precoder and the receive decoder are trained off-line and when the channel varies, only the linear precoder changes on-line. This latter is designed by using the widely used zero-forcing precoding scheme or its lowcomplexity version based on matrix polynomials. Numerical simulations show that the proposed AP-mMIMO approach achieves competitive performance with a significantly lower complexity compared to existing literature. Index Terms-multiuser (MU) precoding, massive multipleinput multiple-output (MIMO), energy-efficiency, hardware impairment, power amplifier (PA) nonlinearities, autoprecoder, deep learning, neural network (NN)

Recently, there has been a significant amount of interest in satellite telemetry anomaly detection (AD) using neural networks (NN). For AD purposes, the current approaches focus on either forecasting or reconstruction of the time series, and they cannot measure the level of reliability or the probability of correct detection. Although the Bayesian neural network (BNN)-based approaches are well known for time series uncertainty estimation, they are computationally intractable. In this paper, we present a tractable approximation for BNN based on the Monte Carlo (MC) dropout method for capturing the uncertainty in the satellite telemetry time series, without sacrificing accuracy. For time series forecasting, we employ an NN, which consists of several Long Short-Term Memory (LSTM) layers followed by various dense layers. We employ the MC dropout inside each LSTM layer and before the dense layers for uncertainty estimation. With the proposed uncertainty region and by utilizing a post-processing filter, we can effectively capture the anomaly points. Numerical results show that our proposed time series AD approach outperforms the existing methods from both prediction accuracy and AD perspectives.

Quantum Error Correction (QEC) is required in quantum computers to mitigate the effect of errors on physical qubits. When adopting a QEC scheme based on surface codes, error decoding is the most computationally expensive task in the classical electronic back-end. Decoders employing neural networks (NN) are well-suited for this task but their hardware implementation has not been presented yet. This work presents a space exploration of fully-connected feed-forward NN decoders for small distance surface codes. The goal is to optimize the neural network for high decoding performance, while keeping a minimalistic hardware implementation. This is needed to meet the tight delay constraints of real-time surface code decoding. We demonstrate that hardware based NN-decoders can achieve high decoding performance comparable to other state-of-the-art decoding algorithms whilst being well below the tight delay requirements $(\approx 440\ \mathrm{ns})$ of current solid-state qubit technologies for both ASIC designs $(, 19 pages, 21 figures, 5 papers

Operational Load Monitoring is an industrial process that allows to predict the remaining in-service life of a mechanical structure under variable loads. Data from sensors embedded or mounted on the structure is acquired and allows to estimate the number and amplitude of load cycles that the structure has withstood so far in its working environment. This process is especially important in the aerospace industry where mechanical structures of an aircraft are monitored in order to maximize their operating lifetime. Smart Operational Load Monitoring means implementation of artificial intelligence techniques to the process in order to make predictions based on measurements from reduced number of sensors. In this paper a composite lightweight structure of typical geometry used in aircraft structures is taken as an example for Smart Operational Load Monitoring. The predictions are made from measurements from six strain gauges mounted to the structure, using carefully prepared artificial intelligence-based models. Efficiency of the models is compared, in terms of their prediction accuracies and computational complexities., National Agency for Academic Exchange of Poland, Silesian University of Technology. Faculty of Mechanical Engineering

This paper provides an overview of enhanced network services, while emphasizing on the role of Unmanned Aerial Vehicles (UAVs) as core network equipment with radio and backhaul capabilities. Initially, we elaborate the various deployment options, focusing on UAVs as airborne radio, backhaul and core network equipment, pointing out the benefits and limitations. We then analyze the required enhancements in the Service-Based Architecture (SBA) to support UAV services including UAV navigation and air traffic management, weather forecasting and UAV connectivity management. The use of airborne UAVs network services is assessed via qualitative means, considering the impact on vehicular applications. Finally, an evaluation has been conducted via a testbed implementation, to explore the performance of UAVs as edge cloud nodes, hosting an Aerial Control System (ACS) function responsible for the control and orchestration of a UAV fleet.

Precise online lithium-ion battery state of health estimation is critical for the correct operation and management of battery-based energy storage systems such as microgrids and electric vehicles. However, in such applications it is necessary to maintain standard operation and therefore difficult to experimentally determine. Advancements in machine learning techniques and capabilities allow for precise and efficient data-driven predictions. In this paper we propose a simple, yet effective state of health estimation model based on the extraction of features observed from patterns in the voltage, current and temperature profiles during the charging process, which then through artificial neural networks allow for per cycle estimations. We then apply this model to two groups of batteries from the NASA Ames PCoE Battery data set. Results show that the proposed model is capable of estimating the state of health of batteries discharged under varied conditions with resulting coefficients of determination between 0.896 and 0.992 while also employing significantly less input data than other works. This work was partially supported by the Junta de Andalucia, Spain under project UMA18-FEDERJA-150 and Gobierno de España, Spain under project RTI2018-093421-B-100. Funding for open access charge: Universidad de Málaga–CBUA .

This paper addresses the gap in the scientific literature regarding construction cost estimates for the construction of underground metro stations. It provides preliminary cost estimation models using linear regression for use by the Greek underground metro public transport authority for planning future extensions to the Athens and Thessaloniki networks. At the same time, it contributes to the body of knowledge by proposing material quantity prediction models and presents a two-stage preliminary cost estimation model for the construction of civil engineering works of underground metro stations. Stage one uses the construction cost budgets of six metro stations in Greece to develop a multilinear regression equation for the prediction of the overall cost for construction of civil engineering works; stage two provides estimates of material quantities using linear regression, key quantity ratios, and artificial neural networks. The data analyzed are from the prior measurements of quantities for the construction of the Chaidari to Piraeus extension of the Athens Metro Line 3. After comparing the actual values of costs and quantities with the corresponding predictions, acceptable discrepancies are observed. All models provide estimates within ±25% discrepancies, which are acceptable at the conceptual planning phase in order to initiate project funding quests.

This document describes the results of the study contributing to the methods and tools applicable in plastic waste sorting systems that exploit the multistatic ultra-wideband impulse radar enforced with a deep learning signal processing back-end. The novelty of the research is the use of synthetic data for the development of a trained neural network before real data are available, and the use of a multistatic radar for the improvement of the training data set. The study results are described in multiple publications; the current paper shows the applicability of the described approach. The main results are as follows: a monostatic impulse radar can be used for the determination of material properties, such as thickness, dielectric permittivity, and losses, with limited accuracy; multistatic radar configuration increases the accuracy of the material property estimation; an open source finite difference time domain simulator can be used to simulate electromagnetic wave propagation in dielectric structures in order to generate synthetic data for development of optimized artificial neuron network structures used for the estimation of dielectric material properties, and the developed network can successfully be used for multistatic radar data processing.

Local stress concentration occurs when faults are present in pipelines under pressure. An example of such defects is the problem of corrosion caused by the environment in the field of pipeline installation. In the first part of this paper, we attempt to model the corrosion in the hemi-ellipsoidal form in order to study the locations of stress concentration in the specimens by several experimental cases and their influence on the stress resistance. The Gurson-Tvergaard-Needleman (GTN) mesoscopic damage model is used to simulate the specimens with good accuracy. In the second part, the investigation is extended to a pipe under static pressure with and without the presence of a glass fibre reinforced polymer (GFRP) composite patch. The maximum stress and percent stress reduction in a defected pipe with a hemi-ellipsoidal defect are determined using a 3D finite element model. This part examines the impact of the geometry of the composite patches on the percentage reduction of the maximum stresses in a section of pipeline subjected to static pressure. In the third part, the stresses and the percentage reduction in the maximum stresses are predicted using an artificial neural network (ANN). An inverse problem using ANN and Jaya algorithm is proposed to predict the group level of different sizes of defects under composite patches based on the maximum stress and percentage reduction of stress that the pipe withstands. The new method relates directly to real-world pipeline construction and repair applications. It could be also used for structural safety monitoring.

In this paper the use of a visualization tool that helps researchers discover knowledge in databases is described. The paper presents the scientific and technological maps associated with bibliometric indicators. The maps are based on the concept of Self-Organizing Maps (SOM) which is a particularly robust form of unsupervised neural networks. Examples illustrating the visualization of information are included in the paper. En este trabajo se describe el uso de una herramienta de visualización que facilita descubrir conocimientos en bases de datos. Se presentan los mapas científico-tecnológicos asociados a los indicadores bibliométricos. Los mapas están basados en el concepto de los mapas auto-organizados (SOM, Self-Organizing Maps) y basados en la tecnología de las redes neuronales no supervisadas. Se incluyen ejemplos que permiten ilustrar la visualización de los resultados.

International audience; In this paper, a novel system is proposed for automating the process of brain tumor classification in magnetic resonance (MR) images. The proposed system has been validated on a database composed of 90 brain MR images belonging to different persons with several types of tumors. The images were arranged into 6 classes of brain tumors with 15 samples for each class. Each MR image of the brain is represented by a feature vector composed of several parameters extracted by two methods: the image entropy and the seven Hu's invariant moments. These two methods are applied on selected zones obtained by sliding a window along the MR image of the brain. The size of the used sliding window is 16x16 pixels for the first method (image entropy) and 64x64 pixels for the second method (seven Hu's invariant moments). To implement the classification, a multilayer perceptron trained with the gradient backpropagation algorithm has been used. The obtained results are very encouraging; the resulting system properly classifies 97.77% of the images of the used database.

© 2019 International Association of Computer Science and Information Technology. Human gesture is a form of body language usually used as a mean of communication and is very critical in human-robot interactions. Vision-based gesture recognition methods to detect hand motion are vital to support such interactions. Hand gesture recognition enables a convenient and usable interface between devices and users. In this paper, an approach is presented for hand gesture recognition based on image processing methods, namely Wavelets Transform (WT), Empirical Mode Decomposition (EMD), besides Artificial Intelligence classifier which is Artificial Neural Networks (ANN) and Convolutional Neural Network (CNN). The methods are evaluated based on many factors including execution time, accuracy, sensitivity, specificity, positive and negative predictive value, likelihood, receiver operating characteristic (ROC), area under roc curve (AUC) and root mean Square. Results indicate that WT have less execution time than EMD and CNN. In addition, CNN is more effective in extracting distinct features and classifying data accurately compared to EMD and WT. Ministry of Higher Education in Saudi Arabia

This paper analyzes the rapid and unexpected rise of deep learning within Artificial Intelligence and its applications. It tackles the possible reasons for this remarkable success, providing candidate paths towards a satisfactory explanation of why it works so well, at least in some domains. A historical account is given for the ups and downs, which have characterized neural networks research and its evolution from “shallow” to “deep” learning architectures. A precise account of “success” is given, in order to sieve out aspects pertaining to marketing or sociology of research, and the remaining aspects seem to certify a genuine value of deep learning, calling for explanation. The alleged two main propelling factors for deep learning, namely computing hardware performance and neuroscience findings, are scrutinized, and evaluated as relevant but insufficient for a comprehensive explanation. We review various attempts that have been made to provide mathematical foundations able to justify the efficiency of deep learning, and we deem this is the most promising road to follow, even if the current achievements are too scattered and relevant for very limited classes of deep neural models. The authors’ take is that most of what can explain the very nature of why deep learning works at all and even very well across so many domains of application is still to be understood and further research, which addresses the theoretical foundation of artificial learning, is still very much needed.

The direct torque control (DTC) system, which is based on induction machine drive is a developed and simple control method. It allows high dynamic performance with very simple hysteresis control scheme; However, its disadvantages are high current, torque, and flux ripple. In this paper, DTC of induction machine drive has been improved by using the applications of artificial intelligence (AI) approaches to reduce the current, torque, and flux ripples and also get better performance of the machines. At the conclusion of this study, the outcomes of traditional DTC and artificial intelligent methods are compared. By the program MATLAB/SIMULINK, the modeling and simulation results of the DTC system for induction machine (IM) have been proposed.

The treatment of wastewater is an essential factor in preventing pollutants and promoting the quality of the water. The inherent complexity, influential impact and the solid waste infrastructure all lead to confusion and variance in the primary clarifier for wastewater. These inconsistencies lead to variations in the purity and capacity constraints of wastewater and the existential impact of water receipt. Water treatment is a complicated task that has chemical, technical and biochemical aspects. A credible artificial neural network (ANN) method is necessary for another wastewater treatment plant to prevent the breakdown of the processes. Virtual reality seems to have become a strong solution for preventing waste management uncertainties and problems. This is not only due to extreme changes but also to significant external disturbances that water systems are subjected to when controlling challenges. Climate is among the most significant of such disturbances. Various environmental conditions actually include different influx frequencies and levels of substances. Water contamination has become one of the extremely serious growing concerns; sewage treatment plant identification is a key major issue here and the agencies enforce tighter requirements when operating wastewater software systems. This article plans to create models of achievement and prospects when possible future guidance of recent research borders for the use of artificial intelligence in wastewater treatment plants that concurrently deal with pollutants. This study has shown us that the composite ANN provides a greater level of competence in plant prediction and systemization. Highlight The ANN parameters are tuned with different weights and biases to obtain effective results and a solution to achieve the best results. The proposed research methodology also suggests that this research can be made more effective by fine-tuning ANN parameters using optimization techniques. This paper explores the effective utilization of artificial intelligence for wastewater. The composite ANN has been found to deliver a higher level of competence in plant prediction and systemization in this study. The behavior of input water in wastewater treatment plants must be predicted using a neural network.

This paper is a discussion of the results of tests intended to (i) estimate the effects of component mix ratios and heat curing of an adhesive joint on the tensile strength, and (ii) to determine the adhesive component mix ratio for which heat curing is insignificant to the strength of adhesive butt joints. Experimental tests were carried out at ambient temperature and elevated temperature during which adhesive butt joints were loaded with a tensile force until failure. The variables were the mix ratio of epoxy adhesive components and the application of heat holding at the adhesive curing stage. An LSTM (long short-time memory) forecast was used to determine the point corresponding to the mix ratio of adhesive components at which heat holding of the adhesive joint no longer has a positive and significant importance to the final tensile strength of the joint.

This paper provides a novel procedure to estimate the education level of social network (SN) users by leveraging artificial neural networks (ANN). Additionally, it provides a robust methodology to extract explanatory insights from ANN models. It also contributes to the study of socio-demographic phenomena by utilizing less explored data sources, such as social media. It proposes Twitter data as an alternative data source for in-depth social studies, and ANN for complex patterns recognition. Moreover, cutting edge technology, such as face recognition, on social media data are applied to explain the social characteristics of country-specific users. We use nine variables and three hidden layers of neurons to identify high-skilled users. The resulted model describes well the level of education by correctly estimating it with an accuracy of 95% on the training set and an accuracy of 92% on a testing set. Approximately 30% of the analyzed users are highly skilled and this share does not differ among the two genders. However, it tends to be lower among users younger than 30 years old. Supplementary Information The online version contains supplementary material available at 10.1007/s13278-021-00832-1.

The synchronous generator is one of the most important active components in current electric power systems. New control methods should be designed to guarantee an efficient dynamic performance of the synchronous generator in strongly interconnected nonlinear power systems over a wide range of variable operating conditions. In this context, active suppression capability for different uncertainties and external disturbances represents a current trend in the development of new control design methodologies. In this paper, a new adaptive neural control scheme based on differential flatness with a modified structure including B-spline Neural Networks for transient stabilization and tracking of power-angle reference profiles for synchronous generators in interconnected electric power systems is introduced. These features are attained due to the advantages extracted of these two approaches: (a) a control design stage based on a power system model by differential flatness and (b) an adaptive performance using a correct design of B-spline Neural Networks, minimizing parameter dependency. The effectiveness of the proposed algorithm is demonstrated by simulation results in two test systems: single machine infinite bus and an interconnected power system. Transient stability and robust power-angle reference profile tracking are both verified.

This study proposes a heterogeneous hardware-based framework for network intrusion detection using lightweight artificial neural network models. With the increase in the volume of exchanged data, IoT networks’ security has become a crucial issue. Anomaly-based intrusion detection systems (IDS) using machine learning have recently gained increased popularity due to their generation’s ability to detect unseen attacks. However, the deployment of anomaly-based AI-assisted IDS for IoT devices is computationally expensive. A high-performance and ultra-low power consumption anomaly-based IDS framework is proposed and evaluated in this paper. The framework has achieved the highest accuracy of 98.57% and 99.66% on the UNSW-NB15 and IoT-23 datasets, respectively. The inference engine on the MAX78000EVKIT AI-microcontroller is 11.3 times faster than the Intel Core i7-9750H 2.6 GHz and 21.3 times faster than NVIDIA GeForce GTX 1650 graphics cards, when the power drawn was 18mW. In addition, the pipelined design on the PYNQ-Z2 SoC FPGA board with the Xilinx Zynq xc7z020-1clg400c device is optimised to run at the on-chip frequency (100 MHz), which shows a speedup of 53.5 times compared to the MAX78000EVKIT.

In medical and health sciences, the detection of cell injury plays an important role in diagnosis, personal treatment and disease prevention. Despite recent advancements in tools and methods for image classification, it is challenging to classify cell images with higher precision and accuracy. Cell classification based on computer vision offers significant benefits in biomedicine and healthcare. There have been studies reported where cell classification techniques have been complemented by Artificial Intelligence-based classifiers such as Convolutional Neural Networks. These classifiers suffer from the drawback of the scale of computational resources required for training and hence do not offer real-time classification capabilities for an embedded system platform. Field Programmable Gate Arrays (FPGAs) offer the flexibility of hardware reconfiguration and have emerged as a viable platform for algorithm acceleration. Given that the logic resources and on-chip memory available on a single device are still limited, hardware/software co-design is proposed where image pre-processing and network training were performed in software, and trained architectures were mapped onto an FPGA device (Nexys4DDR) for real-time cell classification. This paper demonstrates that the embedded hardware-based cell classifier performs with almost 100% accuracy in detecting different types of damaged kidney cells.

One of the most critical problems in a communication system is losing information between the transmitter and the receiver. WiMAX (Worldwide Interoperability of Microwave Access) technology is gaining popularity and recognition as a Broadband Wireless Access (BWA) solution. At frequencies below 11 GHz, WiMAX can operate in line-of-sight (LOS) and non-line-of-sight (NLOS) scenarios. The implementation of WiMAX networks are rushed worldwide. Estimating path loss is crucial in the early stages of wireless network deployment and cell design. To anticipate propagation loss, several path loss models are available (e.g., Okumura Model Hata Model), but they are all bound by particular parameters. In this paper, we propose an MLP neural network-based path loss model with a well-structured implementation network design and grid search-based hyperparameter tuning. The proposed model optimally approximates mobile and base station path losses. Therefore, neurons number, learning rate, and hidden layers number are investigated to obtain the best model in terms of prediction accuracy. Path loss data is collected based on 14 networks in different microcellular settings. Simulations under Matlab environment showed that prediction errors were lower than standard log-distance-based path loss models.

Alkali-activated materials (AAMs) are qualitatively and quantitatively evaluated with an emphasis on the ultimate drying shrinkage. We systematically evaluated AAMs based on the mix design and curing conditions, utilizing a total of 452 AAM mixtures extracted from 44 papers. Finally, a predictive model for the ultimate drying shrinkage of AAMs was constructed using an artificial neural network (ANN) with high accuracy, in which the reactivity of binder, geopolymer paste volume, liquid-to-binder ratio, alkali activator modulus, aggregate volumetric ratio, curing temperature, relative humidity and specimen size were set as inputs. This model shows great generality by compiling various AAM mixtures and is easy-handling without preparation of samples for acquiring specific properties. Moreover, the efficiency of three commonly used models for predicting the drying shrinkage-the Bazant-Baweja model, Gardner and Lockman model, and multi-linear regression model-were evaluated and compared to the proposed ANN model, revealing a better prediction performance of ANN model. This study will advance the understanding of the drying shrinkage behaviors of AAMs and provide practical guidelines for designing AAM mixtures with high durability.

The operation and maintenance of a photovoltaic system is a challenging task that requires scientific soundness, and has significant economic impact. Faults in photovoltaic systems are a common phenomenon that demands fast diagnosis and repair. The effective and accurate diagnosis and categorization of faults is based on information received from the photovoltaic plant monitoring and energy management system. This paper presents the application of machine learning techniques in the processing of monitoring datasets of grid connected systems in order to diagnose faults. In particular, monitoring data from four photovoltaic parks located in Central Greece are analyzed. The existing data are divided for training and validation procedures. Different scenarios are examined first, in order to observe and quantify the behavior of artificial neural networks in already known faults. In this process, the faults are divided in three main categories. The system’s performance deviation against the prediction of the trained artificial neural network in each fault category is processed by health monitoring methodology in order to specify it quantitatively.

The transformer is one of the most important electrical machines in electrical systems. Its proper operation is fundamental for the distribution and transmission of electrical energy. During its service life, it is under continuous electrical and mechanical stresses that can produce diverse types of damage. Among them, short-circuited turns (SCTs) in the windings are one of the main causes of the transformer fault; therefore, their detection in an early stage can help to increase the transformer life and reduce the maintenance costs. In this regard, this paper proposes a signal processing-based methodology to detect early SCTs (i.e., damage of low severity) through the analysis of vibroacoustic signals in steady state under different load conditions, i.e., no load, linear load, nonlinear load, and both linear and nonlinear loads, where the transformer is adapted to emulate different conditions, i.e., healthy (0 SCTs) and with damage of low severity (1 and 2 SCTs). In the signal processing stage, the contrast index is analyzed as a fault indicator, where the Unser and Tamura definitions are tested. For the automatic classification of the obtained indices, an artificial neural network is used. It showed better results than the ones provided by a support vector machine. Results demonstrate that the contrast estimation is suitable as a fault indicator for all the load conditions since 89.78% of accuracy is obtained if the Unser definition is used.

The early detection of pathogen infections in plants has become an important aspect of integrated disease management. Although previous research demonstrated the idea of applying digital technologies to monitor and predict plant health status, there is no effective system for detecting pathogen infection before symptomatology appears. This paper presents the use of a low-cost and portable electronic nose coupled with machine learning (ML) models for early disease detection. Several artificial neural network models were developed to predict plant physiological data and classify processing tomato plants and soil samples according to different levels of pathogen inoculum by using e-nose outputs as inputs, plant physiological data, and the level of infection as targets. Results showed that the pattern recognition models based on different infection levels had an overall accuracy of 94.4–96.8% for tomato plants and between 94.81% and 96.22% for soil samples. For the prediction of plant physiological parameters (photosynthesis, stomatal conductance, and transpiration) using regression models or tomato plants, the overall correlation coefficient was 0.97–0.99, with very significant slope values in the range 0.97–1. The performance of all models shows no signs of under or overfitting. It is hence proven accurate and valid to use the electronic nose coupled with ML modeling for effective early disease detection of processing tomatoes and could also be further implemented to monitor other abiotic and biotic stressors.

The number and efficiency of seismic networks have steadily increase over time delivering large datasets to be analyzed for earthquake occurrence. Automatic tools for accurate earthquake detection are under emerging and intense development. This paper first proposes a new windowing procedure of seismic traces that highly facilitates earthquake detection. This procedure applies regular trace filtering and normalization, but also performs a strict window alignment to P wave onset. These event-aligned windows represent the input data to our P wave detection networks, with relatively small or moderate number of layers. We then develop Feed-Forward (FFNN) and Convolutional (CNN) neural networks and explore multiple architecture configurations to find relevant hyperparameter patterns for better detection. To assess network performance, we adopt the widely used metrics of accuracy (ACC) and the area under the curve (AUC) of the Receiver Operating Characteristic function. In terms of ACC, the best FFNN and CNN reach performances of 91% and 98%, respectively. On the other hand, the best FFNN and CNN in terms of AUC achieve performances of 96% and 99%, respectively. Thus, our novel trace windowing procedure allows developing networks with few hyperparameters, for correct earthquake detection under low computational costs. Finally, we use the CNN with best AUC performance as an effective trace filtering with the purpose of P wave arrival time estimation. This work is partially supported by the Generalitat de Catalunya under grant 2017-SGR-962 and the DRAC project (001-P-001723). The research leading to these results has received funding from the European Union’s Horizon 2020 research and innovation programme under the ChEESE project, grant agreement No. 823844. This project has also received funding from the European Union’s Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie grant agreement No. 777778 MATHROCKS.

The paper presents recent advances in the design of controllable, highly accurate, and multi–band Raman gain profiles. The ultra–wideband programmable gain profiles are implemented using a machine learning approach based on the mapping between gain profiles and pump powers.

The tremendous rise of electrical energy demand worldwide has led to many problems related to efficient use of electrical energy, consequently posing difficult challenges to electricity consumers of all levels—from households to large companies’ facilities. Most of these challenges could be overcome by the accurate prediction of electricity demand. Additionally, balance responsibility includes the penalty-based financial mechanism causing extra expense for badly estimated consumption, above the allowed imbalance limits. In this paper, a method for electricity consumption prediction based on artificial neural networks is proposed. The electricity consumption dataset is obtained from a cold storage facility, which generates data in hourly intervals. The data obtained are measured for a period of over 2 years and then separated to four seasons, so different models are developed for each season. Five different network structures (ordinary RNN, LSTM, GRU, bidirectional LSTM, bidirectional GRU) for five different values of horizon, i.e., input data (one day, two days, four days, one week, two weeks) are examined. Performance indices, such as mean absolute percentage error (MAPE), root mean square error (RMSE), mean absolute error (MAE) and mean square error (MSE), are used in order to obtain qualitative and quantitative comparisons among the obtained models. The results show that the modifications of recurrent neural networks perform much better than ordinary recurrent neural networks. GRU and LSTMB structures with horizons of 168h and 336h are found to have the best performances.

Most plant viral infections are vector-borne. There is a latent period of disease inside the vector after obtaining the virus from the infected plant. Thus, after interacting with an infected vector, the plant demonstrates an incubation time before becoming diseased. This paper analyzes a mathematical model for persistent vector-borne viral plant disease dynamics. The backpropagated neural network based on the Levenberg—Marquardt algorithm (NN-BLMA) is used to study approximate solutions for fluctuations in natural plant mortality and vector mortality rates. A state-of-the-art numerical technique is utilized to generate reference data for obtaining surrogate solutions for multiple cases through NN-BLMA. Curve fitting, regression analysis, error histograms, and convergence analysis are used to assess accuracy of the calculated solutions. It is evident from our simulations that NN-BLMA is accurate and reliable.

The aim of this paper is to present a novel approach to energy use forecasting. We propose a nested fuzzy cognitive map in which each concept at a higher level can be decomposed into another fuzzy cognitive map, multilayer perceptron artificial neural network or long short-term memory network. Historical data related to energy consumption are used to construct a nested fuzzy cognitive map in order to better understand energy use behavior. Through the experiments, the usefulness of the nested structure in energy demand prediction is demonstrated, by calculating three popular metrics: Mean Square Error, Mean Absolute Error and the correlation coefficient. A comparative analysis is performed, applying classic multilayer perceptron artificial neural networks, long short-term memory networks and fuzzy cognitive maps. The results confirmed that the proposed approach outperforms the classic methods in terms of prediction accuracy. Moreover, the advantage of the proposed approach is the ability to present complex time series in the form of a clear nested structure presenting the main concepts influencing energy consumption on the first level. The second level allows for more detailed problem analysis and lower forecast errors.

As demand for more hydrological data has been increasing, there is a need for the development of more accurate and descriptive models. A pending issue regarding the input data of said models is the missing data from observation stations in the field. In this paper, a methodology utilizing ensembles of artificial neural networks is developed with the goal of estimating missing precipitation data in the extended region of Chania, Greece on a daily timestep. In the investigated stations, there have been multiple missing data events, as well as missing data prior to their installation. The methodology presented aims to generate precipitation time series based on observed data from neighboring stations and its results have been compared with a Multiple Linear Regression model as the basis for improvements to standard practice. For each combination of stations missing daily data, an ensemble has been developed. According to the statistical indexes that were calculated, ANN ensembles resulted in increased accuracy compared to the Multiple Linear Regression model. Despite this, the training time of the ensembles was quite long compared to that of the Multiple Linear Regression model, which suggests that increased accuracy comes at the cost of calculation time and processing power. In conclusion, when dealing with missing data in precipitation time series, ANNs yield more accurate results compared to MLR methods but require more time for producing them. The urgency of the required data in essence dictates which method should be used.

This paper deals with the motion trajectory tracking control problem based on output feedback and artificial neural networks for anthropomorphic manipulator robots under disturbed operating scenarios. This class of manipulator robots constitutes nonlinear dynamic systems subjected to disturbance torques induced mainly by work payload. Parametric uncertainty and possible dynamic modeling errors stand for other kind of disturbances that can deteriorate the efficiency and robustness of the tracking of controlled nonlinear robotic system trajectories. In fact, the presence of unknown dynamic disturbances is unavoidable in industrial robotic engineering systems. Therefore, for high-precision applications, such as laser cutting, marking, or welding, effective control schemes should be designed to guarantee adequate motion profile tracking planned on this class of disturbed nonlinear robotic system. In this context, a new adaptive robust motion trajectory tracking control scheme based on output feedback and artificial neural networks of anthropomorphic manipulator robots is presented. Three-layer B-spline artificial neural networks and time-series modeling are properly exploited in the design of novel adaptive robust motion tracking controllers for robotic applications of laser manufacturing. In this way, dependency on detailed nonlinear mathematical modeling of robotic systems is considerably reduced, and real-time estimation of uncertain dynamic disturbances is not required. Furthermore, several cases studies to demonstrate the motion planning tracking control robustness for a class of MIMO nonlinear robotic systems are described. blue Insights for the extension of the introduced output-feedback adaptive neural control design approach for other architecture of nonlinear robotic systems are depicted.

Quantum entanglement is a unique phenomenon of quantum mechanics, which has no classical counterpart and gives quantum systems their advantage in computing, communication, sensing, and metrology. In quantum sensing and metrology, utilizing an entangled probe state enhances the achievable precision more than its classical counterpart. Noise in the probe state preparation step can cause the system to output unentangled states, which might not be resourceful. Hence, an effective method for the detection and classification of tripartite entanglement is required at that step. However, current mathematical methods cannot robustly classify multiclass entanglement in tripartite quantum systems, especially in the case of mixed states. In this paper, we explore the utility of artificial neural networks for classifying the entanglement of tripartite quantum states into fully separable, biseparable, and fully entangled states. We employed Bell’s inequality for the dataset of tripartite quantum states and train the deep neural network for multiclass classification. This entanglement classification method is computationally efficient due to using a small number of measurements. At the same time, it also maintains generalization by covering a large Hilbert space of tripartite quantum states.

Organic food consumption has become a significant trend in consumer behaviour, determined by various motives, among which the price does not play a major role, thus reflecting the Lancaster approach to the microeconomic consumer theory. Additionally, artificial neural networks (ANNs) have proven to have significant potential in providing accurate and efficient models for predicting consumer behaviour. Considering these two trends, this study aims to deploy the Lancaster approach in the emerging area of artificial intelligence. The paper aims to develop the ANN-based predictive model to investigate the relationship between organic food consumption, demographic characteristics, and health awareness attitudes. Survey research has been conducted on a sample of Croatian inhabitants, and ANN models have been used to assess the importance of various determinants for organic food consumption. A Three-layer Multilayer Perceptron Neural Networks (MLPNN) structure has been constructed and validated to select the optimal number of neurons and transfer functions. One layer is used as the first input, while the other two are hidden layers (the first covers the radially symmetrical input, sigmoid function; the second covers the output, softmax function). Three versions of the testing, training, and holdout data structures were used to develop ANNs. The highest accuracy was achieved with a 7-2-1 partition. The best ANN model was determined as the model that was showing the smallest percent of incorrect predictions in the holdout stage, the second-lowest cross-entropy error, the correct classification rate, and the area under the ROC curve. The research results show that the availability of healthy food shops and consumer awareness of these shops strongly impacts organic food consumption. Using the ANN methodology, this analysis confirmed the validity of the Lancaster approach, stating that the characteristics or attributes of goods are defined by the consumer and not by the product itself.