Search

Millimeter-wave and Terahertz communications consist of complex analog and mixed-signal transceivers where overall system performance is often limited by the weakest performing subsystem. While analog and mixed-signal integrated circuits have significantly advanced, the future of 5G and 6G transceiver design could be accelerated by including artificial intelligence. In this combination, analog integrated circuit design and operation would harness machine learning to identify, characterize, and act upon variations and anomalies in system performance. Focusing on 5G and 6G, this paper investigates solutions for a unified intelligent integrated transceiver: a conceptual combination of a traditional analog subsystem, a supporting digital subsystem that enables artificial intelligence, and dedicated feedback circuitry or sensors that monitor performance, efficiency, or reliability. Active and passive components and propagation channels are reviewed based on their merits of introducing intelligence. Holistically and for broader applicability, the paper conceptualizes and coins the notion of an “intelligent integrated system (IIS)”, which brings forward a novel unified vision and approach toward context-aware subsystems that dynamically interact with ambient and varying operating conditions. To demonstrate viability, the paper concatenates a select set of measurement results.

Intelligent design is a trend in geotechnical engineering. However, a middle link between demand identification and intelligent design is missing in anchor engineering, resulting in lower efficiency in terms of the interaction between the anchor components, geological body and modeling information. To address this challenge, this paper proposes a method for modeling a three-dimensional anchor engineering information model based on sparse boreholes. This method includes three parts: modeling the geological body, integrated modeling of the anchor components and the geological body, and multiple information assignments based on the anchor model. The feasibility was validated in actual engineering scenarios by using neural networks and rendering engines. The results showed that the prediction accuracy of the strata improved by 17.69%, the modeling efficiency increased by double, and engineering information was fit to the model well. This paper has the application potential to achieve intelligent design in anchor engineering.

Medical time series data often exhibit intricate and dynamic patterns. With the rapid advancement of medical digitization, deep learning-based time series anomaly detection techniques have found extensive applications in the healthcare field, such as detecting irregular heart rhythms and monitoring patients’ vital signs. To fully leverage digitized medical records to identify anomalies in healthcare and address key challenges in precise anomaly detection, this paper provides a comprehensive review of deep learning-based anomaly detection techniques applied to medical time series data. By reviewing and summarizing the relevant research, this paper explores the deep learning-based time series anomaly detection techniques within the medical and health domain, analyzing the strengths and limitations of different deep learning architectures and algorithms in tackling specific medical tasks. Lastly, we discuss the challenges faced by this field and outline future research directions. By reviewing and summarizing advanced deep learning methods for time series anomaly detection in medical applications in recent years, this study contributes to the advancement of healthcare analytics, aiming to enhance patient treatment outcomes.

It is important for humans to remain hydrated, particularly for older adults who are at a greater risk of dehydration and may forget to drink. Monitoring liquid intake and getting reminders to drink throughout the day is a useful solution to increase hydration levels. The objective of this paper is to automatically detect drink events from multiple containers in a simulated home environment using a vision-based approach. The proposed work compares the use of depth and RGB (red, green, blue) cameras for this task. In this paper, we compared 2D and 3D Convolutional Neural Networks (CNN) using RGB and depth cameras. We collected data from nine participants performing drinking, eating and other Activities of Daily Living (ADL) in a simulated home environment. We found that for the 3D models, the RGB and depth camera inputs provided very similar F1-scores for both 10-Fold (94.3% vs 93.9%, respectively) and Leave-One-Subject-Out (LOSO) cross validation (84.2% vs 86.2%, respectively). This is a promising result as depth cameras also mitigate the challenges to privacy of RGB-based models. The 3D CNN models outperformed the 2D models, thereby creating a more robust system. Depth cameras are a useful alternative to RGB cameras with equal performance in identifying drinking events.

Non-linearity of wireless transceivers, specifically power amplifier (PA) non-linearity, could pose major limitations towards having high throughput, and cost and energy efficient wireless communication systems. Such limitations from the PA is typically compensated in the transmitter, e.g. by applying power back-off or performing digital-pre-distortion (DPD) aiming to linearize the transmitter. However, applying PA power back-off leads to lower energy efficiency, and lower output power, and hence lower coverage; and performing DPD results in higher complexity of the transmitters. This paper presents an alternative approach based on a receiver method to perform signal detection in the presence of distortions due to PA non-linearity. We propose a receiver technique using artificial neural networks (ANN) to compensate for the PA non-linearity at the receiver side. The paper presents link-level simulation results using pre-trained neural network models based on synthesized training data. The simulation results confirm that the designed receiver can tolerate higher distortions, hence allow the PA output power back-off to be reduced, leading to higher output power improving coverage, spectral efficiency, energy efficiency, and throughput.

This paper presents the application of Artificial Intelligence (AI) to automate the intricate process of designing Sigma-Delta Analog-to-Digital Converters (SD-ADC). It explores the integration of Artificial Neural Networks (ANN) to augment the design, optimization, and operation of Sigma Delta Modulators (SDM). The core concept focuses on training ANN models to achieve the optimal design parameters for SD-ADCs. Traditional SD-ADC design approaches typically involve an iterative simulation process, wherein circuit design parameters are adjusted until the desired performance is achieved. However, this approach can be time-consuming and computationally demanding, especially when dealing with the complex SD-ADCs architectures. In response, this article proposes an alternative by replacing this manual process with the assistance of ANN models. The Proposed methodology comprises two phases. In Phase 1, the trained ANN model predicts the SD-ADC output parameters, while Phase 2 focuses on estimating the optimal input design parameters of the SD-ADC. Remarkably, the circuit’s desired parameters are derived from ANN model predictions rather than the traditional method at the architectural level. The comparative analysis between ANN, Machine Learning (ML), and traditional optimization models is performed to validate the effectiveness of the proposed model.

Artificial intelligence stands for an excellent alternative to be considered in development of new adaptive high-efficiency control design methodologies for uncertain modern complex engineering systems driven by electric motors. In this sense, artificial neural networks can be embedded within innovative nonlinear control design strategies to add capabilities to neutralize various kinds of dynamic perturbations or uncertainties that can significantly deteriorate the operating efficiency of nonlinear electric motor systems. This paper introduces a novel adaptive sliding mode nonlinear control method based on B-spline artificial neural networks for efficient tracking of optimal smooth operating reference trajectories in induction or asynchronous motors under substantially disturbed operational scenarios. Robustness regarding nonlinear theoretical mathematical modelling errors, parametric uncertainty and unknown external multiple-frequency oscillating disturbing influences is considered. Numerical experiments involving multiple disturbed operating case studies are presented to demonstrate the effective performance of the proposed robust adaptive artificial neural-network control scheme on large horsepower three-phase induction motors. Finally, a comparative evaluation is conducted, emphasizing system performance through the use of index performance criteria, and comparing the proposed adaptive robust B-spline approach with a nonlinear passivity-based controller. New insights to extend the introduced adaptive neural network robust control design strategy to other considerably perturbed practical nonlinear engineering systems are thus provided.

Traditional transfer diagnosis models for internal combustion engines show a decrease in generalization ability due to the multisource features aliasing in vibration signals and the effect of variable operating conditions. To address this problem, this paper proposes a transfer diagnosis model based on the deep subdomain adaptive network framework. To address feature aliasing, based on minimizing amplitude moment and reconstruction loss, a new adaptive decomposition layer is designed and embedded into the framework to decompose complex signals into single-impact components in time domain. To alleviate the effect of operating conditions, a new constraint for minimizing signal feature variance loss is designed and introduced into the framework’s loss function. This constraint calculates the variance of the sample features of the same fault label under variable operating conditions, aiming to excavate invariant features of operating conditions and complete feature mapping of domain adaptation. Validation with experimental data yields an accuracy of 94.81%.

In power systems with a significant hydroelectric component, instances of the Unit Commitment (UC) problem may be much more computationally intensive due to the longer decision horizons and the additional hydro constraints. Therefore, this paper presents a methodology to reduce the solution space to accelerate 168-hour-ahead UC formulated as a Mixed-Integer Linear Program (MILP). First, an offline model maps environment observations to actions in a Multi-Agent Deep Reinforcement Learning (MADRL) model. This mapping uses historical power system operation data to determine the on/off status of specific generation units. Then, the online model uses the binary variable solutions obtained by the offline model to solve a UC problem with a reduced solution space. The Multi-Agent approach allows each agent, based on Artificial Neural Networks (ANN) with a Temporal Convolutional Network (TCN) architecture, to group units that are located in the same region. A shared cumulative reward function is used to adjust simultaneously the different ANN weights during the learning phase. The effectiveness of our method is demonstrated using real operational data of the Chilean National Electricity System, achieving statistically significant lower computation times and a negligible error that is within the integrality gap of the solver.

An exponential growth and complexity in diverse distribution systems have contributed to protection coordination challenges. Initially, protection coordination schemes were achieved by means of conventional techniques; however, the utilisation of such methods is based on trial-and-error principles and laborious. Consequently, current studies have adopted the utilisation of particle swarm optimization, artificial intelligence models, and genetic algorithms to optimise overcurrent relay selectivity and operational speed. Particle swarm optimization, artificial intelligence, and genetic algorithms are optimization techniques that at times converges prematurely due to poor selection of control parameters and lack of optimal values, which results in increased computational time. Therefore, this paper presents a comprehensive review of recent developments in terms of parametric sensitivity analysis, selection of artificial intelligence models based on data availability, and the likelihood of solving overcurrent relay coordination problems. The reviewed literature shows that particle swarm optimization performance is greatly influenced by inertia weight and swarm size, while the number of iterations has insignificant effect. The findings also indicate that crossover rate, mutation probability, and population size affect genetic algorithms behaviour. Artificial intelligence models lack sensitivity study for parametric tuning, that is, number of hidden layers, membership functions, epsilon in support vector machine, and number of fuzzy rules affects the models’ performance.

This paper presents an in-depth analysis of state-of-the-art semantic segmentation algorithms applied to spacecraft safe planetary landing via hazard detection and avoidance. Several architectures are trained from binary safety maps and the rich dataset of the High-Resolution Imaging Science Experiment (HiRISE) embedded on Mars Reconnaissance Orbiter for realistic purposes. The study incorporates several metrics comparisons such as recognition accuracy, computational complexity, model complexity, and inference time. The proposed performance indices and combinations are analyzed and discussed. The experiments were performed using a Raspberry Pi 4B, which is a relevant commercial-of-the-shelf microcontroller surrogate of NASA’s High-Performance Spaceflight Computer (HPSC) that will thrive within the next decades in space exploration. This paper allows researchers to know what has been tested on the subject and serves as a catalog for users to pick the most relevant architecture for their own application.

This paper presents a novel Transformer-based facial landmark localization network named Localization Transformer (LOTR). The proposed framework is a direct coordinate regression approach leveraging a Transformer network to better utilize the spatial information in a feature map. An LOTR model consists of three main modules: 1) a visual backbone that converts an input image into a feature map, 2) a Transformer module that improves the feature representation from the visual backbone, and 3) a landmark prediction head that directly predicts landmark coordinates from the Transformer’s representation. Given cropped-and-aligned face images, the proposed LOTR can be trained end-to-end without requiring any post-processing steps. This paper also introduces a loss function named smooth-Wing loss, which addresses the gradient discontinuity of the Wing loss, leading to better convergence than standard loss functions such as L1, L2, and Wing loss. Experimental results on the JD landmark dataset provided by the First Grand Challenge of 106-Point Facial Landmark Localization indicate the superiority of LOTR over the existing methods on the leaderboard and two recent heatmap-based approaches. On the WFLW dataset, the proposed LOTR framework demonstrates promising results compared with several state-of-the-art methods. Additionally, we report an improvement in the performance of state-of-the-art face recognition systems when using our proposed LOTRs for face alignment.

The sixth generation (6G) wireless communication network presents itself as a promising technique that can be utilized to provide a fully data-driven network evaluating and optimizing the end-to-end behavior and big volumes of a real-time network within a data rate of Tb/s. In addition, 6G adopts an average of 1000+ massive number of connections per person in one decade (2030 virtually instantaneously). The data-driven network is a novel service paradigm that offers a new application for the future of 6G wireless communication and network architecture. It enables ultra-reliable and low latency communication (URLLC) enhancing information transmission up to around 1 Tb/s data rate while achieving a 0.1 millisecond transmission latency. The main limitation of this technique is the computational power available for distributing with big data and greatly designed artificial neural networks. The work carried out in this paper aims to highlight improvements to the multi-level architecture by enabling artificial intelligence (AI) in URLLC providing a new technique in designing wireless networks. This is done through the application of learning, predicting, and decision-making to manage the stream of individuals trained by big data. The secondary aim of this research paper is to improve a multi-level architecture. This enables user level for device intelligence, cell level for edge intelligence, and cloud intelligence for URLLC. The improvement mainly depends on using the training process in unsupervised learning by developing data-driven resource management. In addition, improving a multi-level architecture for URLLC through deep learning (DL) would facilitate the creation of a data-driven AI system, 6G networks for intelligent devices, and technologies based on an effective learning capability. These investigational problems are essential in addressing the requirements in the creation of future smart networks. Moreover, this work provides further ideas on several research gaps between DL and 6G that are up-to-date unknown.

The clinical treatment of degenerative and developmental lumbar spinal stenosis (LSS) is different. Computed tomography (CT) is helpful in distinguishing degenerative and developmental LSS due to its advantage in imaging osseous and calcified tissues. However, boundaries of the vertebral body, spinal canal and dural sac have low contrast and are hard to identify in a CT image, so the diagnosis depends heavily on the knowledge of expert surgeons and radiologists. In this paper, we develop an automatic lumbar spinal CT image segmentation method to assist LSS diagnosis. The main contributions of this paper are as follows: 1) a new lumbar spinal CT image dataset is constructed that contains 2393 axial CT images collected from 279 patients, with the ground truth of pixel-level segmentation labels; 2) a dual densely connected U-shaped neural network (DDU-Net) is used to segment the spinal canal, dural sac and vertebral body in an end-to-end manner; 3) DDU-Net is capable of segmenting tissues with large scale-variant, inconspicuous edges (e.g., spinal canal) and extremely small size (e.g., dural sac); and 4) DDU-Net is practical, requiring no image preprocessing such as contrast enhancement, registration and denoising, and the running time reaches 12 FPS. In the experiment, we achieved state-of-the-art performance on the lumbar spinal image segmentation task. We expect that the technique will increase both radiology workflow efficiency and the perceived value of radiology reports for referring clinicians and patients.

Smart farming is a new concept that makes agriculture more efficient and effective by using advanced information technologies. The latest advancements in connectivity, automation, and artificial intelligence enable farmers better to monitor all procedures and apply precise treatments determined by machines with superhuman accuracy. Farmers, data scientists and, engineers continue to work on techniques that allow optimizing the human labor required in farming. With valuable information resources improving day by day, smart farming turns into a learning system and becomes even smarter. Deep learning is a type of machine learning method, using artificial neural network principles. The main feature by which deep learning networks are distinguished from neural networks is their depth and that feature makes them capable of discovering latent structures within unlabeled, unstructured data. Deep learning networks that do not need human intervention while performing automatic feature extraction have a significant advantage over previous algorithms. The focus of this study is to explore the advantages of using deep learning in agricultural applications. This bibliography reviews the potential of using deep learning techniques in agricultural industries. The bibliography contains 120 papers from the database of the Science Citation Index on the subject that were published between 2016 and 2019. These studies have been retrieved from 39 scientific journals. The papers are classified into the following categories as disease detection, plant classification, land cover identification, precision livestock farming, pest recognition, object recognition, smart irrigation, phenotyping, and weed detection.

This paper aims at visualizing deep convolutional neural network interpretations for aerial imagery and understanding how these interpretations change across datasets or when network weights are damaged. Our visualization results offer insights on the generalization power and resilience of commonly used networks, such as VGG16, ResNet50, and DenseNet121. Our experiments on the AID and the UCM aerial datasets demonstrate the emergence of object and texture detectors in convolutional networks commonly used for classification. We further analyze these interpretations when the network is trained on one dataset and tested on another to demonstrate the robustness of feature learning across aerial datasets. We also explore the shift in interpretations when performing transfer learning from an aerial dataset (AID) to a generic object dataset (MS-COCO). These results illustrate how transfer learning benefits the network's internal representations. For analyzing the effects of damages on activation maps, our work proposes to simulate damages by randomly zeroing network weights at different levels of the network. The paper carries out experiments with retraining the network to check if it can recover the lost interpretations. Visualizing changes in the neural network's interpretation when the undamaged weights are updated, allows us to assess the resilience of a network visually. Finally, we propose a new metric for the quantitative assessment of network resilience.

Bus arrival prediction has important implications for public travel, urban dispatch, and mitigation of traffic congestion. The factors affecting urban traffic conditions are complex and changeable. As the predicted distance increases, the difficulty of traffic prediction becomes more difficult. Forecast based on historical data responds quite slowly for changes under the short-term conditions, and vehicle prediction based on real-time speed is not sufficient to predict under long-term conditions. Therefore, an arrival prediction method based on temporal vector and another arrival prediction method based on spatial vector is proposed to solve the problems of remote dependence of bus arrival and road incidents, respectively. In this paper, combining the advantages of the two prediction models, this paper proposes a long short-term memory (LSTM) and Artificial neural networks (ANN) comprehensive prediction model based on spatial-temporal features vectors. The long-distance arrival-to-station prediction is realized from the dimension of time feature, and the short-distance arrival-to-station prediction is realized from the dimension of spatial feature, thereby realizing the bus-to-station prediction. Besides, experiments were conducted and tested based on the entity dataset, and the result shows that the proposed method has high accuracy among bus arrival prediction problems.

This paper proposes an artificial neural network (ANN)-based synthesis method for nonuniform linear arrays with mutual coupling effects. The proposed method can simultaneously optimize the location distributions and excitations of the elements. As is well known, for nonuniform linear arrays, the mutual coupling effects on the active element patterns (AEPs) and passive S parameters vary with the location distribution of the elements. However, few papers have focused on how to describe the relationship between the effects of mutual coupling and the location distribution of the elements. In this paper, we use several parallel and independent ANNs to characterize the effects of mutual coupling on the AEPs and passive S parameters in nonuniform linear arrays with variable location distributions. At the same time, to relieve the curse of dimensionality, we make use of the idea of subarrays, which makes it possible for the proposed method to model nonuniform linear arrays with a large number of elements. After building accurate ANN models, we make use of the adaptive differential evolution algorithm JADE to search for the optimal location distribution and the excitations of the elements, to satisfy the requirements for the radiation pattern and the active S parameters of the antenna array. In view of the fact that the mutual coupling effects are accounted for in the synthesis process, the proposed method probably exhibits more practical value in antenna array designs than in other synthesis methods. The validity and efficiency of the proposed method are confirmed based on three examples.

The concept of smart grids has enabled the addition of renewable energy resources in the utility grids. This integration is possible after energy conversion and energy conversion causes the degradation in the quality of electric power. The power quality can be enhanced by reducing the total harmonic distortion (THD) in energy conversion. In this paper, we have presented an improved artificial neural network (ANN) based approach that can be useful to reduce the THD levels in a cascaded H-bridge (CHB) multilevel inverter (MLI). The proposed ANN architecture consists of only one hidden layer with ten neurons and can generate accurate results in both overfitting and underfitting conditions. Due to a smaller number of neurons, the proposed architecture is less complex and can produce results relatively faster as compared to other ANN architectures available in the literature. The proposed ANN architecture is tested on an asymmetrical CHB MLI, and the results are compared with the recent state-of-the-art (SOTA) techniques. The CHB MLI under test had three DC voltage sources with 1:2:4 and inductive loads. The simulations are performed in MATLAB and Simulink environment and the switching angles are optimized by using the ANN architecture. The results have shown a 71.95% improvement in current THDs and a 13.91% improvement in voltage THDs as compared to the SOTA technique. The proposed configuration is 89.6% efficient on inductive load and uses a smaller number of switches. Finally, an experimental setup is created, and the simulation results are validated through various experiments.

One of the aspects that mainly characterize wireless networks is their apparent unpredictability. Although several attempts were made in the past years to define for them deterministic medium access techniques, for instance by having data exchanges scheduled by an access point, as a matter of fact they remain a partial solution and are unable to ensure the same behavior as wired infrastructures, since interference may also come from devices outside the network, which obey different rules. A possible way to cope with disturbance on air, both internal and external to the network, is to obtain some knowledge about it by analyzing what happened in the recent past. This information, usually expressed in terms of suitable metrics, is then employed to optimize network operation, for example by prioritizing time-sensitive traffic when needed. In the simplest approaches such metrics coincide with statistical indices evaluated on transmission outcomes, like the failure rate. In this paper we analyze a more sophisticated solution that relies on machine learning, and in particular on artificial neural networks, to predict the behavior of a Wi-Fi link in terms of its frame delivery ratio. Results confirm that more accurate predictions than simpler methods (e.g., moving average) are possible, even when training is partially independent from the specific conditions experienced on the different channels.

The Matrix converters (MCs) are widely used in a large number of applications such as Aircraft, Submarines, and AC drives. In these applications, because of the many chances of occurrence of a critical fault that leads to a halt of the system, and due to the failure of power electric switches, the MC operations can be compromised and there is a fear of burning of the components caused by the whole system shutdown. Therefore, a Fault-Tolerant Control System (FTCS) is extremely necessary for the continued operations of the Three-Phase Matrix Converter (TPMC) to improve the reliability and productivity of the system under faulty conditions. In this paper, an advanced FTCS is proposed based on detection, and dual hardware redundancy for TPMC. Two types of faults are injected into the system to observe the performance of the proposed system: internal open circuit faults on switches and external short circuit faults at the load side. A Fault Detection and Isolation (FDI) unit is used to detect and isolate the faulty switches using Artificial Neural Networks (ANN), and dual hardware redundancy in the switches has been proposed for fault tolerance. In case of external fault, a load-side fault detector is implemented using ANN. The simulation results in MATLAB/Simulink environment show the accurate and stable working of TPMC under faulty conditions, and hardware-in-the-loop is implemented with STM32-Nucleo-F103RB board to verify open circuit fault results. The proposed dual redundant FTC with FDI unit offers an excellent solution for the continued performance of TPMC which ultimately enhances the reliability of the system.

Neural-based sequence-to-sequence methods (Seq2Seq) have proven to be highly effective for Context-sensitive Thai spelling correction. However, they also inherit the drawbacks of Seq2Seq, such as a fixed vocabulary and large data requirements. However, dictionary-based methods and their typical applications are insufficiently robust to produce corrections with reduced error rates. These drawbacks inhibit the application of these methods in a broader range of use cases. In this paper, we provide a practical guide on how to build correction systems progressively and efficiently with three main contributions. First, we present a process for efficiently and progressively producing training data for both neural-based and dictionary-based methods. Our annotation process enables existing methods to be trained with only two percent of the data hand annotated. Second, we propose the Extendable Neural Contextual Corrector (XNCC), a novel text correction approach that decouples the dictionary from the neural model. This enables the dictionary to be extended post-training. Finally, we compare text correction systems with various configurations to demonstrate how these systems can be effectively used to produce corrections. Our experiments show that 1) minor changes to dictionary-based methods can significantly improve correction performance, 2) neural-based correction systems can be trained using a fraction of the data, and 3) XNCC can have the dictionary extended to generalize to new data without re-training. Lastly, we provide recommendations for progressively building text correction systems at multiple levels of implementation effort based on our findings.

Inverter-based microgrids essentially constitute an extensive communication layer that makes them vulnerable to cyber anomalies. The distributed cooperative controllers implemented at the secondary control level of such systems exchange information among physical nodes using the cyber layer to meet the control objectives. The cyber anomalies targeting the communication network may distort normal operation, therefore, an effective cyber anomaly mitigation technique using an Artificial Neural Network (ANN) is proposed in this paper. The intelligent anomaly mitigation control is modeled using a dynamic neural network that employs a nonlinear autoregressive network with exogenous inputs. The effects of false data injection on the distributed cooperative controller at the secondary control level are considered. The training data for designing the neural network are generated by multiple simulations of the designed microgrid under various operating conditions using MATLAB/Simulink. An explainable framework is employed to interpret the output generated by the trained neural network-based controller after the neural network has been trained offline and validated online in the simulated microgrid. The proposed technique is applied as secondary voltage and frequency control of distributed cooperative control-based microgrid to regulate the voltage under various operating conditions. The performance of the proposed control technique is verified by injecting various types of false data injection-based cyber anomalies. The proposed ANN-based secondary controller maintained the normal operation of the microgrid under various cyber anomalies as demonstrated on a real-time digital simulator.

The growing interest in renewable energy and the falling prices of solar panels place solar electricity in a favourable position for adoption. However, the high-rate adoption of intermittent renewable energy introduces challenges and the potential to create power instability between the available power generation and the load demand. Hence, accurate solar Photovoltaic (PV) power forecasting is essential to maintain system reliability and maximize renewable energy integration. The current solar PV power forecasting approaches are an essential tool to maintain system reliability and maximize renewable energy integration. This paper presents a comprehensive and comparative review of existing Machine Learning (ML) based approaches used in PV power forecasting, focusing on short-term horizons. We provide an overview of factors affecting solar PV power forecasting and an overview of existing PV power forecasting methods in the literature, with a specific focus on ML-based models. To further enhance the comparison and provide more insights into the advancement in the area, we simulate the performance of different ML methods used in solar PV power forecasting and, finally, a discussion on the results of the work.

The cogging torque of permanent magnet machines with a modular stator is affected by additional harmonic components due to the segmentation of the stator lamination. This paper proposes a novel approach based on the shaping of the stator tooth tips with sinusoidal profiles to minimize the cogging torque of such machines. A theoretical study and a design formula are proposed to determine the spatial frequency of the sinusoidal profiles, while an optimization procedure based on genetic algorithm and artificial neural networks is adopted to determine their amplitudes and phase shifts. The proposed method is validated through finite element analysis considering two different case studies. Also, a comparison with other approaches from the literature is presented to highlight the effectiveness of the proposed technique. Finally, an additional analysis is reported to demonstrate the effectiveness of the proposed method against manufacturing and assembling tolerances.

In this paper, an algorithm for the estimation of the linear inter-channel crosstalk in a dense-WDM polarization-multiplexed 16-QAM transmission scenario is proposed and demonstrated. The algorithm is based on the use of a feed-forward neural network (FFNN) inside the coherent digital receiver. Two types of FFNNs were considered, the first based on a regression algorithm and the second based on a classification algorithm. Both FFNN algorithms are applied to features extracted from the histograms of the in-phase and quadrature components of the equalized digital samples. After a simulative investigation, the performance of the channel spacing estimation algorithms was experimentally validated in a 3 × 52 Gbaud 16-QAM WDM system scenario.

This paper deals with model predictive control synthesis which take benefits from artificial neural networks to model (non-linear) dynamical system. More precisely, thanks to a systematic and rigorous methodology, it is shown that residual networks (ResNet) and PolyInception networks (PolyNet) neural network architectures, developed initially for image recognition, are very good candidate for i) identification of dynamical systems, ii) being used as embedded model in a model predictive control laws. Concretely, the widely used non-linear dynamical system quadruple tank process is used as a benchmark. The neural network architectures studied are i) feedforward networks as a reference point, and the two other linked to Euler integration method ii) residual networks and iii) PolyInception networks. Networks training is performed by mixing classical back-propagation algorithm and hyperparameters optimisation through heuristics. The identification results provided show that neural networks of types ii) and iii) perform better than the classical one i), with a better generalisation capability. Finally, model predictive controllers are synthesized based on the various networks trained. The simulation results obtained for controlling water levels of a 4 tanks system benchmark give interesting insights. They show that residual networks based model predictive control is better suited than feedforward networks and PolyInception networks based ones, both taking into account computation time and set point errors.

The increasing integration of Power Electronics (PE)-based renewable energy sources into the electric power system has significantly affected the traditional levels and characteristics of fault currents compared to the ones observed in power systems dominated by synchronous generating units. The secure operation of a renewable rich power system requires the proper estimation of fault currents with wide range of scenarios of the high share of renewables. Although the utilization of detailed and complex time-domain dynamic simulations allows for calculating the fault currents, the resulting modeling complexity and computational burden might not be adequate from the operational perspective. Thus, it is necessary to develop alternative quicker data-driven fault current estimation approaches to support the system operator. For this purpose, this paper utilizes an Artificial Neural Network (ANN)-based tool to estimate the characteristics of short circuit currents in power systems with high penetration of power electronics-based renewables. The short circuits against different penetration of renewables are produced offline using the DIgSILENT PowerFactory considering the control requirements for renewables (e.g., fault ride through requirement). The resulting dataset is utilized to train the ANN to provide the mapping between the penetration level and the characteristics of the short circuit currents. The application of the approach using the modified IEEE 9-bus test system demonstrates its effectiveness to estimate the components of short circuit currents (sub-transient current, transient current, and peak current) with high accuracy based only on the penetration of power electronics-based renewables.

It is common to find multiple metaheuristics to solve continuous optimization problems. However, choosing what optimizer may obtain the best results for a given task requires exhaustive evaluations that are highly application-dependent. Besides, it is necessary to find sufficiently good tuning parameters to achieve satisfactory performance with the selected approach. In this context, the automatic design of algorithms, particularly those based on heuristics, has been increasing in popularity in the previous years due to its undoubted relevance nowadays. This paper explores a novel approach based on hyper-heuristics to carefully select population-based search operators and their tuning parameters to generate metaheuristics capable of dealing with a given practical engineering problem. The proposed strategy is assessed using three highly relevant and illustrative problems: training Artificial Neural Networks, designing PID controllers, and modeling a calorimetric phenomenon based on fractional calculus. In addition, we implement three well-known optimization metaheuristics to compare achieved solutions via the proposed hyper-heuristic strategy, namely Particle Swarm Optimization, Genetic Algorithm, and Cuckoo Search. Results from extensive numerical tests prove that the customized metaheuristics are generally superior to the three well-known algorithms, taking only a few iterations to converge to an optimal solution. This is an excellent indicator of alleviating the effort and expertise required to choose the proper methodology when dealing with real-valued optimization problems.

This paper presents a review of the literature on network traffic prediction, while also serving as a tutorial to the topic. We examine works based on autoregressive moving average models, like ARMA, ARIMA and SARIMA, as well as works based on Artifical Neural Networks approaches, such as RNN, LSTM, GRU, and CNN. In all cases, we provide a complete and self-contained presentation of the mathematical foundations of each technique, which allows the reader to get a full understanding of the operation of the different proposed methods. Further, we perform numerical experiments based on real data sets, which allows comparing the various approaches directly in terms of fitting quality and computational costs. We make our code publicly available, so that readers can readily access a wide range of forecasting tools, and possibly use them as benchmarks for more advanced solutions.

In this paper, we consider a molecular communication system that is made of a 3D unbounded diffusion channel model without flow, a point transmitter, and a spherical absorbing receiver. In particular, we study the impact of inter-symbol interference and analyze the performance of different threshold-based receiver schemes. The aim of this paper is to analyze and optimize the receivers by using the conventional model-based approach, which relies on an accurate model of the system, and the emerging data-driven approach, which, on the other hand, does not need any apriori information about the system model and exploits deep learning tools. We develop a general analytical framework for analyzing the performance of threshold-based receiver schemes, which are suitable to optimize the detection threshold. In addition, we show that data-driven receiver designs yield the same performance as receivers that have perfect knowledge of the underlaying channel model.

This paper aims to exploit approximate computing units in image processing systems and artificial neural networks. For this purpose, a general design methodology is introduced, and approximation-oriented architectures are developed for different applications. This paper proposes a method to compromise power/area efficiency of circuit-level design with accuracy supervision of system-level design. The proposed method selects approximate computational units that minimize the total computation cost, yet maintaining the ultimate performance. This is accomplished by formulating a linear programming problem, which can be solved by conventional linear programming solvers. Approximate computing units, such as multipliers, neurons, and convolution kernels, which are proposed by this paper, are suitable for power/area reduction through accuracy scaling. The formulation is demonstrated on applications in image processing, digital filters, and artificial neural networks. This way, the proposed technique and architectures are tested with different approximate computing units, as well as system-level requirement metrics, such as PSNR and classification performance.

Currently, population growth is global and tends to concentrate in large cities, which increases the demand for illuminating public spaces for safety, visual orientation, aesthetic considerations, and quality of life. The undesirable side effects are increase in energy consumption and light pollution. The current tools used for designing public lighting systems are not suitable for optimizing multiple objectives in addition to energy savings, and these solutions could provide for a more sustainable environment. The application of evolutionary optimization techniques seems to be growing rapidly because of the nonlinearity of the model behavior and the nonproprietary nature of the algorithms, which are considered as black box systems. This paper develops a data model for these types of optimizers, analyzing the ability of different artificial neural network (ANN) architectures to simulate a simple public lighting design by measuring the performance with respect to the fitness function, training speed, and goodness of fit with a dataset generated with different conditions. The architectures selected in this paper are those with multilayer perceptrons (MLPs) with different hidden layer configurations using different numbers of neurons in each layer, which have been analyzed to determine the configuration that best fits the purpose of this work. The data for training the ANNs were generated with a recognized open-software platform, DIALux. The experiments were repeated and analyzed to determine the variance of the results obtained. In this way, it was possible to identify the most appropriate number of iterations required. The results show that better precision is obtained when using the Levenberg-Marquardt training algorithm, especially when the ANN architecture has fewer neurons in the hidden layer.

Approximately 2.5 quintillion bytes of data are emitted on a daily basis, and this has brought the world into the era of “big data.” Artificial neural networks (ANNs) are known for their effectiveness and efficiency for small datasets, and this era of big data has posed a challenge to the big data analytics using ANN. Recently, much research effort has been devoted to the application of the ANN in big data analytics and is still ongoing, although it is in it is early stages. The purpose of this paper is to summarize recent progress, challenges, and opportunities for future research. This paper presents a concise view of the state of the art, challenges, and future research opportunities regarding the applications of the ANN in big data analytics and reveals that progress has been made in this area. Our review points out the limitations of the previous approaches, the challenges in the ANN approaches in terms of their applications in big data analytics, and several ANN architecture that have not yet been explored in big data analytics and opportunities for future research. We believe that this paper can serve as a yardstick for future progress on the applications of the ANN in big data analytics as well as a starting point for new researchers with an interest in the exploration of the ANN in big data analytics.

Medical image analysis is currently experiencing a paradigm shift due to deep learning. This technology has recently attracted so much interest of the Medical Imaging Community that it led to a specialized conference in “Medical Imaging with Deep Learning” in the year 2018. This paper surveys the recent developments in this direction and provides a critical review of the related major aspects. We organize the reviewed literature according to the underlying pattern recognition tasks and further sub-categorize it following a taxonomy based on human anatomy. This paper does not assume prior knowledge of deep learning and makes a significant contribution in explaining the core deep learning concepts to the non-experts in the Medical Community. This paper provides a unique computer vision/machine learning perspective taken on the advances of deep learning in medical imaging. This enables us to single out “lack of appropriately annotated large-scale data sets” as the core challenge (among other challenges) in this research direction. We draw on the insights from the sister research fields of computer vision, pattern recognition, and machine learning, where the techniques of dealing with such challenges have already matured, to provide promising directions for the Medical Imaging Community to fully harness deep learning in the future.

Multiple parameters with large nonlinear characteristics must be considered simultaneously to design the coil dimensions of static inductive power transfer (SIPT) systems. The design of dynamic inductive power transfer (DIPT) systems is more challenging due to the large number of parameters needed to be considered. In the conventional artificial neural network (ANN)-based design approach, optimal coil dimensions are found using ANN that has learned the nonlinear characteristics between coil dimensions and magnetic characteristics using the finite element method (FEM). However, this approach requires a large amount of training data, and it is difficult to reach an optimum design if there are many design criteria. In order to overcome these challenges, this paper proposes a design optimization method using two approaches: improving the time efficiency of ANN training data collection by superposing the magnetic fields from the coils and improving the input value of ANN using a genetic algorithm. Design results predicted by the ANN are compared with FEM simulation, circuit simulations, and experimental results to verify the validity of the proposed algorithm. The FEM and circuit simulation results and the ANN prediction results match with errors of 10.2% or less for all design requirements. Experimental results are provided for a 3 kW DIPT system with four transmitter coils and an automated test rail. Comparison results between ANN predicted values and experimental values match with an error of less than 12.7%.

This paper proposes a simple neural network compact form model-free adaptive controller (NNCFMFAC) for a single thin McKibben muscle (TMM) system. The main contribution of this work is the simplification of the current neural network (NN) based compact form model-free adaptive controller (CFMFAC), which requires only two adaptive weights. This is achieved by designing a NN topology to specifically enhance the CFMFAC response. The prominent control parameters of the CFMFAC are combined and an adaptive weight is used for self-tuning, while the second adaptive weight is used to minimize the offset at each operating point. Hence the issues of redundant adaptive weights in complex neuro-based CFMFACs and slow response of the CFMFAC are significantly addressed. The idea is proven in three ways: analytically, simulation on a nonlinear system and experiments on a TMM platform. Experimental results demonstrating the superiority of the proposed method over the conventional CFMFAC is confirmed by a 76% improvement in convergence speed and a 60% reduction in root mean square error (RMSE). It is envisaged that the proposed controller can be very useful for TMM driven applications as it is model-independent, has fast response, high tracking accuracy, and minimal complexity.

An Explainable Recommendation system has become more prominent because it could convince the user to trust the prediction. However, the existing Explainable Recommendation systems face one of these two limitations. The first limitation is the utilization of a pre-defined feature set to represent the users/items’ preferences. By representing the preferences with a pre-defined feature set, the uniqueness of each user will be limited by the characteristics of the pre-defined set. For the other limitation, the user/item preference representation in existing models is constructed from the entire past interactions. This can mislead the model because the user’s preferences change over the interaction’s characteristics. Therefore, the entire past interactions might not be able to represent the user’s preferences for the target interaction. In this paper, we proposed an Explainable Recommendation system, which represents the user’s preferences with a personalized feature set and is able to adapt the target user/item preference representation to suit the target interaction’s characteristics. By representing the users/items’ preferences with a personalized feature set, the proposed method is able to give the score of importance in the rating contribution to each feature and uses these scores to assume the reasons for the user’s decision. To make the representation suitable to the target interaction’s characteristics, the proposed model utilizes only the target interaction’s information to represent the preferences of users/items. The proposed method also is not relying on the entire past interactions to avoid the bias of dominant features in the past interactions. For the experiment, the proposed method consistently outperforms all baselines on three benchmark datasets. We also analyze the explanation resulting from the proposed model to support our assumptions that the preferences of the user are unique and dynamically change over the interaction’s characteristics.

Recently, healthcare stakeholders have orchestrated steps to strengthen and curb the COVID-19 wave. There has been a surge in vaccinations to curb the virus wave, but it is crucial to strengthen our healthcare resources to fight COVID-19 and like pandemics. Recent researchers have suggested effective forecasting models for COVID-19 transmission rate, spread, and the number of positive cases, but the focus on healthcare resources to meet the current spread is not discussed. Motivated from the gap, in this paper, we propose a scheme, ABV-CoViD (Availibility of Beds and Ventilators for COVID-19 patients), that forms an ensemble forecasting model to predict the availability of beds and ventilators (ABV) for the COVID-19 patients. The scheme considers a region-wise demarcation for the allotment of beds and ventilators (BV), termed resources, based on region-wise ABV and COVID-19 positive patients (inside the hospitals occupying the BV resource). We consider an integration of artificial neural network (ANN) and auto-regressive integrated neural network (ARIMA) model to address both the linear and non-linear dependencies. We also consider the effective wave spread of COVID-19 on external patients (not occupying the BV resources) through a $\theta $ - ARNN model, which gives us long-term complex dependencies of BV resources in the future time window. We have considered the COVID-19 healthcare dataset on 3 USA regions (Illinois, Michigan, and Indiana) for testing our ensemble forecasting scheme from January 2021 to May 2022. We evaluated our scheme in terms of statistical performance metrics and validated that ensemble methods have higher accuracy. In simulation, for linear modelling, we considered the $ARIMA(1,0,12)$ model, and $N^{8-3-2}$ model for ANN modelling. We considered the $\theta -ARNN$ (12,6) forecasting. On a population of 2,93,90,897, the obtained mean absolute error (MAE) on average for 3 regions is 170.5514. The average root means square error (RMSE) of $\theta $ -ARNN is 333.18, with an accuracy of 98.876%, which shows the scheme’s efficacy in ABV measurement over conventional and manual resource allocation schemes.

Optimization problems are usually solved using heuristic algorithms such as swarm intelligence algorithms owing to their ability of providing near-optimal solutions in a feasible amount of time. An example of an optimization problem is the training of artificial neural networks to obtain the most optimal connection weights. Artificial Neural Network (ANN), being the most prominent machine learning algorithm, has a multitude of applications in a myriad of areas. Recently, the use of ANNs has risen exponentially owing to its effective ability of making conclusions based on certain inputs. This ability is primarily achieved during the training phase of the ANN, which is a vital process prior to being able to use the ANN. Gradient descent-based algorithms, which are usually used for the training process, often encounter the problem of local optima, thus being unable to obtain the optimal connection weights of the ANN. Metaheuristic algorithms, including swarm intelligence algorithms, have been found to be a better alternative to train ANNs. The Dragonfly Algorithm (DA) is a swarm intelligence algorithm that has been found to be more effective than multiple swarm intelligence algorithms. However, despite having a good performance, it still suffers from low exploitation. In this paper, we propose to further improve the performance of DA by using hill climbing as a local search technique so as to enhance its low exploitation. The optimized DA algorithm is then used for training artificial neural networks which are employed for classification problems. Based on the experimental results, the optimized DA algorithm has higher effectiveness than the original DA and some other swarm intelligence algorithms as the ANNs trained by the optimized DA have a lower root mean squared error and a higher classification accuracy.

In the optical communication systems, the optical spectrum (OS) provides useful informations for optical performance monitoring and optical link diagnosis. In this paper, we investigate the neural-network based OS analysis (OSA) techniques able to simultaneously recognize the OS and estimate the optical signal to noise ratio (OSNR), cascaded filtering distortion (CFD) and carrier wavelength drift (CWD). We demonstrate that, compared with the multi-task artificial neural network (MT-ANN) and convolutional neural network (MT-CNN), the proposed multi-task cascaded ANNs (CANN) and cascaded CNNs (CCNN) can greatly improve the OSA performance and accelerate the training process by exploiting specific features and loss functions for different tasks. The CCNN able to deal with the high-resolution OS (HOS) can further improve the OSA performance compared with the CANN. The averaged OS recognition accuracy and OSNR, CFD and CWD estimation errors obtained with the CCNN (CANN) for the 15 kinds of optical signals are 99.32% (97.07%), 0.36 (0.55) dB, 0.24 (0.32) and 0.04 (0.06) GHz, respectively, even when the various OS distortions are present. The proposed CANN and CCNN with a better versatility, higher OSA accuracy and faster convergence speed are promising enabling techniques for the future intelligent OSA systems.

Artificial Neural Networks (ANNs) have amassed unprecedented success in information processing ranging from image recognition to time series prediction. The success can largely be attributed to the availability of large datasets for training and the increased complexity of the models. Unfortunately, for some applications only a limited amount of samples is available for training. Fewer training samples increases the risk of over-fitting and poor generalization especially in high complexity models. Moreover, complex models with a large number of trainable parameters require more energy to train and optimize compared to simpler ones. In this paper, to the best of our knowledge, we propose the first use of ANNs for Early Stage Alzheimer Disease classification (ES-AD) from the handwriting (HW). We propose using a framework for building Recurrent Neural Networks (RNNs) known as Reservoir Computing (RC), both numerically and experimentally, that simplifies training by optimizing the output layer only. We also propose the Bidirectional Long Term Short Term (BiLSTM) and Convolutional Neural Network (CNN) methods for comparison. For a fairer comparison, we not only consider the accuracies but also the energy costs incurred to obtain the respective accuracies in order to assess the accuracy-efficiency trade-off. Our numerical and experimental results show that RC yields a classification accuracy of 85%, which is 3% worse than that of BiLSTM and 2% better than that of CNN, at a relatively lower training and significantly lower inference costs. We hope that our findings highlight the importance of examining the accuracy-efficiency trade-off of various models in the community in order to reduce the overall impact of ANNs training on the environment.

This paper introduces PlaneLoc2 - a novel indoor global localization system designed to harness the potential of stereo cameras. A need for robust global localization that does not produce incorrect results (false positives) is present in almost every life-long autonomy task. We show that planar segments extracted from stereo vision data by a neural network enable such robust localization. Planar segments are easier to discriminate than keypoint features and provide easy-to-use geometric constraints. We propose an architecture that exploits a single deep neural network (DNN) to detect planar segments, produce appearance descriptors, and estimate segment geometry. Moreover, we introduce a novel view-based segment map and a novel pose retrieval procedure that considers the uncertainty of features to efficiently use the geometric constraints provided by them. We also show that the new learned descriptor provides better discrimination than the hand-crafted one. Finally, we present experimental results that show that our solution outperforms other state-of-the-art global localization methods and does not produce incorrect agent poses. For both test scenes it recognizes at least 15% more poses than the second best method without incorrect recognitions.

Personalized medicine provides more safe and effective treatment by individualizing the choice of drug and dose based on an individual’s genetic profile. Cancer patients’ response to anti-cancer treatments (drugs) is one of the foremost challenges in personalized medicine that releases the target treatment. Both size and availability of drug sensitivity data have motivated researchers to develop Artificial Intelligence (AI), based models, for predicting drug response to advance cancer treatment. The concerned AI models include Machine Learning (ML) and the recently advanced Deep Learning (DL) based models. This paper introduces both; a data federation method and a DL-based model for predicting drug response. The fundamental goal is to generalize the predictor so it will be able to predict the response to different drugs accurately. As the data has a considerable effect on any AI model, the data federation is utilized to consolidate the data. The proposed consolidation process is carried out to make each cell line contains gene expression data, its mutation profile, and drug response data. ML models such as Support Vector Machine (SVM) and Linear Regression (LR) are used along with Principal Component Analysis (PCA) for feature reduction, and the AI models have been tested with and without data federation. The results show that data federation enhanced the accuracy and decreased the Mean Square Error (MSE) by almost 25%. The proposed DL model uses dimension reduction encoders. The encoder is a DL model that uses unsupervised learning. It is trained by integrating an encoder with a decoder to achieve equality between the input and output. The proposed model has achieved the best accuracy compared to some other recent models in terms of the Pearson correlation coefficient (PCC) as a performance measure. In addition, the results show that the Enhanced Deep Drug Response prediction (Enhanced Deep-DR) model has achieved the best PCC value even with the largest number of genes and drugs, which proves the high capacity and efficiency of the proposed model. Convolutional Neural Network (CNN) based-model is also implemented; it achieves higher accuracy in predicting the drug response than in some other DL-based models but less than the Enhanced Deep learning. The Enhanced Deep-DR achieves better accuracy within the range of 5% to 12% than other DL-models.

Supervised deep learning approaches for automated diagnosis support require datasets annotated by experts. Intra-annotator variability of a single annotator and inter-annotator variability between annotators can affect the quality of the diagnosis support. As medical experts will always differ in annotation details, quantitative studies concerning the annotation quality are of particular interest. A consistent and noise-free annotation of large-scale datasets by, for example, dermatologists or pathologists is a current challenge. Hence, methods are needed to automatically inspect annotations in datasets. In this paper, we categorize annotation noise in image segmentation tasks, present methods to simulate annotation noise, and examine the impact on the segmentation quality. Two novel automated methods to identify intra-annotator and inter-annotator inconsistencies based on uncertainty-aware deep neural networks are proposed. We demonstrate the benefits of our automated inspection methods such as focused re-inspection of noisy annotations or the detection of generally different annotation styles using the biomedical ISIC 2017 Melanoma image segmentation dataset.

This paper presents a novel self-adaptive approach for speech enhancement in the context of highly nonstationary noise. A two-stage deep neuroevolutionary technique for speech enhancement is proposed. The first stage is composed of a deep neural network (DNN) method for speech enhancement. Two DNN methods were tested at this stage, namely, both a deep complex convolution recurrent network (DCCRN) and a residual long short-term memory neural network (ResLSTM). The ResLSTM method was combined with a minimum mean-square error method to perform a preliminary enhancement. The ResLSTM network is used as an a priori signal-to-noise ratio (SNR) estimator. The second stage implements a self-adaptive multiband spectral subtraction enhancement method using tuning optimization based on a genetic algorithm. The proposed two-stage technique is evaluated using objective measures of speech quality and intelligibility. The experiments are carried out using the NOIZEUS noisy speech corpus using conditions of real-world stationary, colored, and nonstationary noise sources at multiple SNR levels. These experiments demonstrate the advantage of building a cooperative approach using evolutionary and deep learning-based techniques that are capable of achieving robust speech enhancement in adverse conditions. Indeed, the experimental tests show that the proposed two-stage technique outperformed a baseline implementation using a state-of-the-art deep learning approach by an average 13% and 6% improvement for six noise conditions at a −5 dB and a 0 dB input SNR, respectively.

Technological advances and scalability are leading Human-Computer Interaction (HCI) to evolve towards intuitive forms, such as through gesture recognition. Among the various interaction strategies, radar-based recognition is emerging as a touchless, privacy-secure, and versatile solution in different environmental conditions. Classical radar-based gesture HCI solutions involve deep learning but require training on large and varied datasets to achieve robust prediction. Innovative self-learning algorithms can help tackling this problem by recognizing patterns and adapt from similar contexts. Yet, such approaches are often computationally expensive and hardly integrable into hardware-constrained solutions. In this paper, we present a gesture recognition algorithm which is easily adaptable to new users and contexts. We exploit an optimization-based meta-learning approach to enable gesture recognition in learning sequences. This method targets at learning the best possible initialization of the model parameters, simplifying training on new contexts when small amounts of data are available. The reduction in computational cost is achieved by processing the radar sensed data of gestures in the form of time maps, to minimize the input data size. This approach enables the adaptation of simple convolutional neural network (CNN) to new hand poses, thus easing the integration of the model into a hardware-constrained platform. Moreover, the use of a Variational Autoencoders (VAE) to reduce the gestures’ dimensionality leads to a model size decrease of an order of magnitude and to half of the required adaptation time. The proposed framework, deployed on the Intel® Neural Compute Stick 2 (NCS 2), leads to an average accuracy of around 84% for unseen gestures when only one example per class is utilized at training time. The accuracy increases up to 92.6% and 94.2% when three and five samples per class are used.

Wireless communications systems are impacted by multi-path fading and Doppler shift in dynamic environments, where the channel becomes doubly-dispersive and its estimation becomes an arduous task. Only a few pilots are used for channel estimation in conventional approaches to preserve high data rate transmission. Consequently, such estimators experience a significant performance degradation in high mobility scenarios. Recently, deep learning has been employed for doubly-dispersive channel estimation due to its low-complexity, robustness, and good generalization ability. Against this backdrop, the current paper presents a comprehensive survey on channel estimation techniques based on deep learning by deeply investigating different methods. The study also provides extensive experimental simulations followed by a computational complexity analysis. After considering different parameters such as modulation order, mobility, frame length, and deep learning architecture, the performance of the studied estimators is evaluated in several mobility scenarios. In addition, the source codes are made available online in order to make the results reproducible.

In this paper, we propose the use of GANs as learned, data-driven knowledge database that can be queried for rapid synthesis of suitable antenna designs given a desired response. As an example, we consider the problem of designing the Log-Periodic Folded Dipole Array (LPFDA) antenna for two non-overlapping ranges of Q-factor values. By representing the antenna with the vector of its structural parameters and considering each desirable range of the Q-factor as a class, we transform our problem to that of generating new samples from a given class. We develop two alternative models, a Conditional Wasserstein GAN and a label-switched library of vanilla Wasserstein GANs and train them with a dataset of features and their associated labels (parameter vectors and Q-factor range). The main component of these models is a generator network that learns to map a normally distributed noise vector along with a binary label to the vector of parameters of candidate structures. We demonstrate that in inference mode, these models can be relied upon for fast generation of suitable designs.

This paper presents a multivariate time series prediction framework based on a transformer model consisting of convolutional neural networks (CNNs). The proposed model has a structure that extracts temporal features of input data through CNN and interprets correlations between variables through an attention mechanism. This framework solves the problem of the inability to simultaneously analyze the temporal features of the input data and the correlation between variables, which is a limitation of the forecasting models presented in existing studies. We designed a forecasting experiment using several time series datasets with various data characteristics to precisely evaluate the proposed model. In addition, comparative experiments were performed between the proposed model and several predictive models proposed in recent studies. Furthermore, we conducted ablation studies on the extent to which the proposed CNN structure in the prediction model affects the forecasting results by substituting a specific layer of the model. The results of the experiments showed that the proposed predictive model exhibited good performance in predicting time series data with a clear cycle and high correlation between variables, and improved the accuracy by approximately 3% to 5% compared with that of previous studies’ time series prediction models.