Search

Among the various Ultra-wideband (UWB) ranging methods, the absence of uplink communication or centralized computation makes downlink time-difference-of-arrival (DL-TDOA) localization the most suitable for large-scale industrial deployments. However, temporary or permanent obstacles in the deployment region often lead to non-line-of-sight (NLOS) channel path and signal outage effects, which result in localization errors. Prior research has addressed this problem by increasing the ranging frequency, which leads to a heavy increase in the user device power consumption. It also does not contribute to any increase in localization accuracy under line-of-sight (LOS) conditions. In this paper, we propose and implement a novel low-power channel-aware dynamic frequency DL-TDOA ranging algorithm, termed DynaTrack. It comprises a NLOS probability predictor based on a convolutional neural network (CNN), a dynamic ranging frequency control module, and an IMU sensor-based ranging filter. DynaTrack is implemented on Samsung Galaxy Note20 Ultra and Qorvo UWB board to show the feasibility and real-time applicability. Based on the conducted experiments, we show that DynaTrack achieves 41% higher accuracy in NLOS conditions while having 46% lower power consumption in LOS conditions compared to existing methods from prior research.

Resume information extraction technologies are crucial for automating the shortlisting and evaluation of resumes, benefiting both enterprises and job seekers. Resume block classification plays a pivotal role in the process of resume information extraction, as it significantly impacts the accuracy of the subsequent intra-block information extraction. However, as reported in the existing research, the classification of resume blocks often uses traditional and general-purpose text classification models, failing to consider the intrinsic contextual order relationships amongst resume blocks. Moreover, there are few studies that consider the transferability of classification models. Therefore, to address these limitations, we propose a series of methods to enhance the performance of resume block classification. Our approach focuses on three key aspects. Firstly, we introduce a novel sequence encoder that effectively extracts the sequence features among blocks of the same resume, leveraging the intrinsic contextual order relationships to improve classification accuracy. Secondly, considering the large classification variance of the existing models in different scenarios, we enhance the sequence encoder with a feature fusion strategy, which combines multiple features to improve the model’s robustness and transferability. Thirdly, from the perspective of ensemble learning, we propose a dynamic weighted hybrid model that dynamically generates weighting for each participating sub-model, enabling adaptive integration of different classification models. Finally, to alleviate the huge workload of cross-domain relabeling, we develop a transfer learning model specifically designed for the resume block classification task, facilitating the application of our approach across different domains. To evaluate the effectiveness of our proposed methods, we release three Chinese datasets that include 4,500 Chinese resumes on https://github.com/xqqhelloword/resume-block-classification. Experimental results show that our hybrid model achieves 97.6% accuracy, outperforming existing methods and establishing a new state-of-the-art in this field.

Infrared imaging has broad and important applications. However, the infrared detector manufacture technique limits the detector resolution and the resolution of infrared images. In this work, we design a Recurrent Large Kernel Attention Neural Network (RLKA-Net) for single infrared image super-resolution(SR), and then demonstrate its superior performance. Compared to other SR networks, RLKA-Net is a lightweight network capable of extracting spatial and temporal features from infrared images. To extract spatial features, we use multiple stacked Recurrent Learning Units (RLUs) to expand the network’s receptive field, while the large kernel attention mechanism in RLUs is used to obtain attention maps at various granularity. To extract temporal features, RLKA-Net uses the recurrent learning strategy to keep persistent memory of extracted features, which contribute to more precise reconstruction results. Moreover, RLKA-Net employs an Attention Gate (AG) to reduce the number of parameters and expedite the training process. We demonstrate the efficacy of the Recurrent Learning Stages (RLS), Large Kernel Attention Block (LKAB), and Attention Gate mechanisms through ablation studies. We test RLKA-Net on several infrared image datasets. The experimental results demonstrate that RLKA-Net presents state-of-the-art performance compared to existing SR models. The code and models are available at https://github.com/ZedFm/ RLKA-Net.

In the context of Enhanced Remote Area Communications (ERAC), Radio over Fiber (RoF) technology plays a crucial role in extending reliable connectivity to underserved and remote areas. This paper explores the significance of fifth-generation (5G) Digital Predistortion (DPD) role in mitigating non-linearities in Radio over Fiber (RoF) systems for enhancing communication capabilities in remote regions. The seamless integration of RoF and 5G technologies requires robust linearization techniques to ensure high-quality signal transmission. In this paper, we propose and exhibit the effectiveness of a machine learning (ML)-based DPD method for linearizing next-generation Analog Radio over Fiber (A-RoF) links within the 5G landscape. The study investigates the use of an optimized recurrent neural network (ORNN) based DPD experimentally on a multiband 5G new radio (NR) A-RoF system while maintaining low complexity. The ORNN model is evaluated using flexible-waveform signals at 2.14 GHz and 5G NR signals at 10 GHz transmitted over a 10 km fiber length. The proposed ORNN-based machine learning approach is optimized and is compared with conventional generalized memory polynomial (GMP) model and canonical piecewise linearization (CPWL) methods in terms of Adjacent Channel Power Ratio (ACPR), Error Vector Magnitude (EVM), and in terms of computation complexity including, storage, time and memory consumption. The findings demonstrate that the proposed ORNN model reduces EVM to below 2% as compared to 12% for non-compensated cases while ACPR is reduced by 18 dBc, meeting 3GPP limits.

An intrusion attack on the Internet of Things (IoT) is any malicious activity or unauthorized access that jeopardizes the integrity and security of IoT systems, networks, or devices. Regarding IoT, intrusions can result in severe problems, including service disruption, data theft, privacy violations, and even bodily injury. One of the intrusion attacks is a keylogging attack, sometimes referred to as keystroke logging or keyboard capture, which is a type of cyberattack in which the attacker secretly observes and records keystrokes made on a device’s keyboard. In the context of IoT, where connected objects communicate and exchange data, this assault may be especially concerning. Keylogging attacks can have severe repercussions in the IoT ecosystem since they can compromise sensitive information, including login passwords, personal information, financial information, or confidential communications. This paper explored the possibility of using an ensemble classifier to detect keylogging attacks in IoT networks. We built an ensemble classifier consisting of three classifiers: a convolutional neural network (CNN), a recurrent neural network (RNN), and a long-short memory network (LSTM). A proposed model uses the BoT-IoT dataset to detect a keylogging attack. Results show that the ensemble model can improve the model’s performance. The ensemble model had excellent accuracy and a low false positive rate. It also had significantly improved detection rates for keylogging attacks than other classifiers.

This work aims to investigate the use of the Recurrent Neural Network (RNN) in automated English grading. In order to achieve this, this work first constructs an automated English grading system based on the Internet of Things (IoT). Next, based on the variant of RNN called Gated Recurrent Unit (GRU), it introduces a self-attention mechanism into bidirectional GRU to form the Bidirectional-GRU_self-attention (Bi-GRU_Att) model. Simultaneously, an attention pooling (AP) mechanism is introduced into bidirectional GRU to form the Bidirectional-GRU_AP (Bi-GRU_AP) model. Comparative experiments are conducted using Chinese and English corpora to compare the performance of these two models. The results indicate that the Bi-GRU_AP model performs well on both Chinese and English datasets. On the Chinese dataset, compared to Bi-GRU_Att, Bi-GRU, and GRU, its accuracy is improved by 1.3%, 9.9%, and 19%, respectively. On the English dataset, compared to Bi-GRU_Att, Bi-GRU, and GRU, its accuracy is improved by 2.2%, 9.8%, and 19.2%, respectively. This suggests that introducing the AP module enables the model to better capture sentence information, thereby enhancing model performance. Additionally, after 20 iterations, the Bi-GRU_AP model exhibits good convergence and stability. The findings provide new insights for the development of automated English subjective grading systems based on IoT and deep learning.

There is a tendency for object detection systems using off-the-shelf algorithms to fail when deployed in complex scenes. The present work describes a case for detecting facial expression in post-surgical neonates (newborns) as a modality for predicting and classifying severe pain in the Neonatal Intensive Care Unit (NICU). Our initial testing showed that both an off-the-shelf face detector and a machine learning algorithm trained on adult faces failed to detect facial expression of neonates in the NICU. We improved accuracy in this complex scene by training a state-of-the-art “You-Only-Look-Once” (YOLO) face detection model using the USF-MNPAD-I dataset of neonate faces. At run-time our trained YOLO model showed a difference of 8.6% mean Average Precision (mAP) and 21.2% Area under the ROC Curve (AUC) for automatic classification of neonatal pain compared with manual pain scoring by NICU nurses. Given the challenges, time and effort associated with collecting ground truth from the faces of post-surgical neonates, here we share the weights from training our YOLO model with these facial expression data. These weights can facilitate the further development of accurate strategies for detecting facial expression, which can be used to predict the time to pain onset in combination with other sensory modalities (body movements, crying frequency, vital signs). Reliable predictions of time to pain onset in turn create a therapeutic window of time wherein NICU nurses and providers can implement safe and effective strategies to mitigate severe pain in this vulnerable patient population.

In the epoch of the fifth generation (5G) of wireless communication, this survey article illuminates the nuanced landscape of 5G Channel Measurements and Models. It meticulously navigates the challenges, methodologies, and practical implications that define the implementation of this groundbreaking technology. Embarking on the exploration, the article unravels the Challenges in 5G Implementation, addressing technological hurdles and standardized protocols pivotal for seamless integration. Unveiling the Existing Methods and Technologies, it highlights Massive MIMO, millimeter-wave communications, and the integration of artificial intelligence (AI) as transformative elements shaping high-capacity wireless communication. Beyond the technical intricacies, the article delves into the Diverse Applications and Use Cases of 5G, promising a paradigm shift from the Internet of Things (IoT) to mission-critical communications. Frequency Considerations and Channel Bands take center stage, dissecting the critical spectrum domain and exploring high-frequency millimeter-wave bands for enhanced data rates. Addressing concerns, the survey scrutinizes Health Concerns and Regulatory Considerations, presenting a balanced perspective on the impact of 5G networks on health and regulatory landscapes. The Benefits Over 4G and LTE drive the transition, with enhanced data rates, lower latency, and massive device connectivity distinguishing 5G. The environmental impact is considered in Energy Efficiency in 5G Networks, exploring strategies and innovations for making 5G networks more energy-efficient. Global Standards and Interoperability become imperative for 5G’s global potential, examining ongoing efforts by standardization bodies for a unified framework. The exploration extends to the realm of 5G Channel Measurements and Models, dissecting Massive MIMO Channel Measurements, Millimeter-Wave Channel Measurements, Beamforming, Antenna Array Configurations, Time-Variant, and Time-Invariant Channel Measurements, Channel Modeling, and Measurement Tools and Techniques. Critically addressing Challenges in 5G Channel Measurements and Models, the article explores limitations and proposes avenues for future research. This survey article serves as a definitive compendium on 5G Channel Measurements and Models, unraveling the complexities, challenges, and opportunities inherent in the deployment of this transformative technology. Its holistic examination provides a valuable resource for researchers, practitioners, and industry stakeholders navigating the dynamic landscape of 5G wireless communication.

Video motion recognition plays a crucial role in advanced sports analysis. With video motion recognition, sports analytics has become more data-driven and result-oriented, significantly enhancing the professionalism and efficiency in the sports domain. Over the years, the accuracy of skelecton-based motion recognition algorithms using Recurrent Neural Networks (RNNs) and Convolutional Neural Networks (ConvNets) has plateaued on relevant datasets, struggling to achieve further breakthroughs. This is partly because RNNs lack sufficient capability to model spatial structural features, and while ConvNets can alleviate difficulties in modeling spatial structures, the conversion of skelecton sequences into RGB pseudo-images inherently leads to some information loss. Moreover, ConvNets are not particularly adept at modeling temporal features. Since the associations between articulations in bodily skelectons are better represented using a graph structure, Graph Convolution Network (GCN)-based skelecton motion recognition methods have gained more attention. Recent advancements in Shift Graph Convolutional Network (S-GCN) have enhanced the expressiveness of spatial graphs while improving the versatility of spatial and temporal graph receptive fields. To further enhance this versatility, we propose the Dynamic Motion-Aware Shift Graph Convolutional Network (DMA-SGCN) for video motion recognition. Specifically, we introduce a data-aware driven method to represent associations between articulations. By analyzing the attributes of different motions and combining them with the natural associations of the bodily skelecton, we compute articulation affinities through representation learning. This approach not only improves the accuracy in defining articulation associations but also enhances the awareness of related articulation associations during motion behaviours. Furthermore, we dynamically use the adjacency matrix of skeletal data to guide the feature transfer between articulations. This topological method allows for more effective shift transformations based on articulation associations, addressing the issue of rigid receptive fields in previous GCNs for motion recognition. Contrast and ablation experiments on the largest 3D motion recognition dataset demonstrate that starting from the skeletal data and the motions themselves enables more accurate excavation of dynamic associations of skeletal articulations in the individual motion recognition.

Forecasting accuracy of electricity prices is crucial to the optimal operation of the electricity market, as improper forecasting can lead to inefficiencies, increased costs, and market instability. Thus, it is highly desired to develop a robust electricity price forecasting framework. The development of an optimal forecasting model depends on the proper choice of exogenous variables, and as the impact/characteristics of the input variables may change over time, thus the choice of appropriate external variables should be a dynamic task. Therefore, it is necessary to develop an online adaptive forecasting model, which will not only continuously forecast but also learn automatically by sensing the changes in the relationship of the variables. To sense the changes and to develop a parsimonious model proper feature engineering is required. Multi-level correlation with multicollinearity has been considered as the feature engineering tool for online training to create an accurate forecasting model. After analyzing existing studies and analyzing the gaps, an approach is proposed, utilizing a General Regression Neural Network (GRNN) with advanced feature engineering and simultaneous adaptive learning, that can outperform traditional models like ANN, RNN, and LSTM in terms of forecasting accuracy.

This study proposed a novel method for lung Four-Dimensional Computed Tomography (4D-CT) deformable image registration (DIR) called convolutional long-short term memory network (C-LSTMNet). It allowed us to capture temporal and spatial features between current and previous phases, as well as perform groupwise registration for multiple image pairs in 4D-CT datasets. The proposed registration framework utilized a recurrent neural network (RNN) to accurately predict displacement vector fields (DVFs) between multiple image pairs in an end-to-end manner, taking into account the spatiotemporal relationship between phase images. The input consisted of a sequence of Three-Dimensional Computed Tomography (3D-CT) images captured from inspiratory phase to expiratory phase within a complete breathing cycle. In this sequence, the first phase image was defined as the target image, while the other phase images were considered as moving images. Multiple C-LSTM units were stacked together to capture temporal clues between these images. The proposed C-LSTMNet was trained using 85 collected 4D-CT datasets from lung cancer patients without supervision, and its efficiency was evaluated by employing an open-source 4D-CT dataset from dir-lab. Target registration error (TRE) was measured and compared between C-LSTMNet and 4 recently published registration methods in a control group. The mean and standard deviation of TRE in C-LSTMNet were $1.30\pm 0.87$ mm, which outperformed other existing deep-learning methods in the control group in this study, and the calculation time for each forward prediction was about 0.45 seconds. The preliminary results on oncologic patients demonstrated that the proposed C-LSTMNet had the potential to accurately and quickly synchronize lung 4D-CT.

Chronic wounds, known for their prolonged healing and the associated financial strain on patients, highlight the growing need for effective monitoring methods. One approach to handling chronic wounds involves monitoring biomarkers, with pH emerging as a promising option. Halochromic sensors, designed to detect color changes in response to pH fluctuations, can be employed for monitoring pH variations in wounds. Previous studies have demonstrated the utility of halochromic-based materials in the field of wound management. Nevertheless, it is worth noting that data derived from such sensors may be prone to misinterpretation, particularly when subjected to human visual processing. Hence, this paper aims to develop a pH prediction system based on the fabricated polymer’s extracted values of different color channels. As a feature selection, principal component analysis (PCA) was performed to determine the number of significant features, and neighborhood component analysis (NCA) was conducted to eliminate the unrelated elements. Subsequently, an RNN model was developed and enhanced through the application of three distinct metaheuristic optimization algorithms: chernobyl disaster optimizer (CDO), chaos game optimizer (CGO), and coronavirus herd immunity optimization algorithm (CHIO). Following the simulation and evaluation of each optimized RNN model, the CGO-RNN model exhibited the lowest mean squared error (MSE) value of 0.1583, indicating that the RNN with chaos game optimizer stands out as the most effective model for this particular application.

In Awake Neurosurgery (AN), the assessment of the patient’s capabilities is of paramount importance to minimize the risk of post-operative cognitive, language, and motor deficits. To retain fine hand motion capabilities, in the current clinical practice, a neuropsychologist assigns hand motion tasks to the patient during AN, and these tasks are evaluated by visual observation. This study introduces an innovative and non-invasive task evaluation method that employs an infrared stereo camera for precise hand pose acquisition. The primary contribution of this work is to offer objective, quantitative and reproducible task evaluations, addressing a critical aspect of patient care during AN procedures. The proposed method focuses on dynamic hand gestures recognition and utilizes unidirectional and bidirectional Long Short-Term Memory networks to assess common motor tasks during AN. To evaluate the tasks, we employ state-of-the-art features, both distance-based and angle-based, formed by the finger bones of the human hands. These feature vectors demonstrate a promising accuracy and inter-patient portability for the tasks under consideration, with mean accuracy exceeding 80% when tested on new patients, separate from those used in training. This approach provides an efficient solution for the identification and assessment of tasks, also eliminating the need for task-specific labeling. This, in turn, enhances usability and reduces the potential for human error. The proposed method has the potential to improve clinical decision-making in AN and offers reliable classification tools for neuropsychologists.

In this paper, an intelligent Model Reference Adaptive Control (MRAC) based on a neural network is proposed for robust tracking control of quadrotor UAV under external disturbances and parameter variations. First, the singularity-free dynamic model of the quadrotor is developed using Newton-Quaternion formalism. Then, conventional MRAC is designed to generate training data. With the generated data, the Feed Forward Neural Network (FFNN) and Recurrent Neural Network (RNN) are trained offline to get an initial set of network parameters for position controller parameters estimation and attitude control of the quadrotor, respectively, and an online learning algorithm is developed to update those network parameters in real-time. Finally, the performance of the designed Neural network-based MRAC has been evaluated using a numerical simulation in a nominal scenario and by introducing parametric variation and external disturbances as matched and unmatched uncertainties into the system. The simulation results show that the proposed controller has a better tracking performance and disturbance rejection capability compared with the Linear Quadratic Regulator (LQR) and conventional MRAC. Furthermore, the utilized control efforts are minimal and smooth proving functional safety and economical use of the controller. Therefore, the suggested controller is feasible for real-time implementation of the quadrotor UAV.

One of the main causes of death from cardiovascular diseases is Myocardial Infarction (MI), which is brought on by coronary artery problems. Myocardial infarction is a pathological condition resulting from an anatomical issue with the Left Ventricle (LV). MI is a potentially fatal cardiac disease for which prompt medical attention can lower the fatality risk. This paper proposes a Deep Learning (DL) based approach to robustly detect binary and multiclass myocardial infarction (MI) in two environments, i.e., with and without data balancing. We employed a Myocardial infarction dataset that contains 1700 MI patients’ medical records. The data preprocessing step is performed, during which data balancing and normalization are carried out. In many real-world medical datasets, class imbalance is a serious problem since it often causes the proposed algorithms to predict the majority class. We apply a class imbalance handling technique to solve the imbalance issue and create a balanced and trustworthy prediction approach. To ensure the generalizability and performance comparison, we employ various deep learning algorithms such as Recurrent Neural Network (RNN), Convolutional Neural Network (CNN), Deep Neural Network (DNN), Long Short-Term Memory (LSTM) without data balancing first, and then after that, with data balancing to enhance model performance. Experiments reveal that the DNN model outperformed others by applying the class imbalance handling technique compared to the method without balancing data. The DNN model attained a maximum test accuracy of 99.39% and a test loss of 0.0252 for the binary class and achieved a maximum test accuracy of 99.74% and a test loss of 0.0115 for multiclass categorization.

Decoding continuous human motion from surface electromyography (sEMG) in advance is crucial for improving the intelligence of exoskeleton robots. However, incomplete sEMG signals are prevalent on account of unstable data transmission, sensor malfunction, and electrode sheet detachment. These non-ideal factors severely compromise the accuracy of continuous motion recognition and the reliability of clinical applications. To tackle this challenge, this paper develops a multi-task parallel learning framework for continuous motion estimation with incomplete sEMG signals. Concretely, a residual network is incorporated into a recurrent neural network to integrate the information flow of hidden states and reconstruct random and consecutive missing sEMG signals. The attention mechanism is applied for redistributing the distribution of weights. A jointly optimized loss function is devised to enable training the model for simultaneously dealing with signal anomalies/absences and multi-joint continuous motion estimation. The proposed model is implemented for estimating hip, knee, and ankle joint angles of physically competent individuals and patients during diverse exercises. Experimental results indicate that the estimation root-mean-square errors with 60% missing sEMG signals steadily converges to below 5 degrees. Even with multi-channel electrode sheet shedding, our model still demonstrates cutting-edge estimation performance, errors only marginally increase 1 degree.

The permanent scatterer interferometric aperture radar (PS-InSAR) technique is used to measure and monitor displacements of the Earth’s surface over time. While the approach is promising for large-scale deformation, the density of the received PS points is insufficient for localized deformation analysis. In this first work, we aim to improve the technique by increasing the point density of high-precision deformation monitoring in PS-InSAR data by developing a convolutional long short-term memory (ConvLSTM) model that predicts PS points on different land covers, such as forest, urban, natural, water, and combinations among them. The proposed architecture, PS-ConvLSTM, was trained on a temporary dataset with interferograms to classify stable and unstable PS pixels from over 200,000 site images obtained from the city of Barcelona, Spain. The result showed that the trained PS-ConvLSTM model is highly compatible with the method currently used, which requires a large manual effort by an expert (accuracy: 99%). In addition, the proposed approach increased the point density by 15%, indicating that ConvLSTM is a promising approach for increasing the point density in PS-InSAR data and thus improving localized deformation analysis.

The paper introduces a Fast Model Predictive Control (FMPC) approach for vehicle path tracking, addressing the challenge of real-time performance in highly nonlinear systems. Utilizing a recurrent neural network with symmetric saturating linear transfer functions (SSL-RNN), our method efficiently constructs an SSL-RNN model for the vehicle. By transforming the MPC optimal control problem into a mixed integer linear programming problem, a swift online solution is achieved. Through simulations on a CarSim/Simulink platform, our FMPC outperforms RNN-based nonlinear MPC and long-short-term memory network-based MPC, demonstrating superior accuracy in vehicle path tracking and enhanced controller solution efficiency.

Reservoir computing (RC) is a bio-inspired framework suited for temporal data processing. Here we propose an all optical coherent RC for the equalization of the impairments of fiber optic (FO) system based on coherent detection. We compare the performances of the RC systems based on semiconductor saturable absorber mirror (SESAM) and highly nonlinear waveguide (HNLW). For a dispersion managed long haul FO system in which the nonlinear penalty is dominant, we find that the SESAM outperforms the HNLW due to its larger nonlinear memory. In contrast, for the equalization of dispersive impairments in a short haul FO system such as data center networks, we find that the HNLW has a much superior performance due to its larger linear memory.

Automated detection of epileptic seizures from scalp Electroencephalogram (EEG) is crucial for improving epilepsy diagnosis and management. This paper presents an automated inter-patient epileptic seizure detection method using multichannel EEG signals. The proposed method uses a scale mixture-based stochastic EEG model for feature extraction and a recurrent neural network for seizure detection. Specifically, the stochastic model, which accounts for uncertainties in EEG amplitude, is fitted to a specific frequency band to extract relevant seizure features. Then, a recurrent neural network-based recognition architecture learns the temporal evolution of these features. We evaluated our method using EEG data from 20 patients with focal epilepsy and conducted comprehensive assessments, including ablation studies on classifiers and features. Our results demonstrate that our approach outperforms static classifiers and existing feature sets, achieving high sensitivity while maintaining acceptable specificity. Furthermore, our feature set showed efficacy both independently and as a complement to existing features, indicating its robustness in seizure detection tasks. These findings reveal that learning the temporal evolution of the stochastic fluctuation and amplitude information of EEG extracted using a stochastic model enables highly accurate seizure detection, potentially advancing automated epilepsy diagnosis in clinical settings.

Decoding continuous human motion from surface electromyography (sEMG) in advance is crucial for improving the intelligence of exoskeleton robots. However, incomplete sEMG signals are prevalent on account of unstable data transmission, sensor malfunction, and electrode sheet detachment. These non-ideal factors severely compromise the accuracy of continuous motion recognition and the reliability of clinical applications. To tackle this challenge, this paper develops a multi-task parallel learning framework for continuous motion estimation with incomplete sEMG signals. Concretely, a residual network is incorporated into a recurrent neural network to integrate the information flow of hidden states and reconstruct random and consecutive missing sEMG signals. The attention mechanism is applied for redistributing the distribution of weights. A jointly optimized loss function is devised to enable training the model for simultaneously dealing with signal anomalies/absences and multi-joint continuous motion estimation. The proposed model is implemented for estimating hip, knee, and ankle joint angles of physically competent individuals and patients during diverse exercises. Experimental results indicate that the estimation root-mean-square errors with 60% missing sEMG signals steadily converges to below 5 degrees. Even with multi-channel electrode sheet shedding, our model still demonstrates cutting-edge estimation performance, errors only marginally increase 1 degree. [ABSTRACT FROM AUTHOR]

Obesity and laziness are some of the common issues in the majority of the youth today. This has led to the development of a proposed exergaming solution where users can play first-person physical games. This research study not only proposes a solution for physical fitness in the form of a game using wearable sensors but also proposes a multi-purpose system that provides different applications when trained for the domain-specific dataset. Critical tasks of gesture recognition and depiction in virtual reality can be applied to many applications in the domains of crime detection, fitness, healthcare, online learning, and sports. In particular, the proposed system enables a user to perform, detect, and depict different gestures in the virtual reality game. First, the system pre-processes input data by applying a median filter to overcome the anomalies. Then, features are extracted through a convolutional neural network, power spectral density, skewness, and kurtosis methods. Further, the system optimizes different features by using the grey wolf optimization. Lastly, the feature set which is optimized is fed to a recurrent neural network for classification. When Compared to the traditional methods, the suggested system gives better results while being easier to use. The IMSporting behaviors (IMSB) dataset includes badminton and other physical activities, the WISDM dataset includes common locomotor motions, and the ERICA dataset which includes a variety of exercises, were used in the experimentation. According to experimental findings, the suggested approach outperformed current methods, which showed detection accuracies of 85.01%, 88.46%, and 93.18% over the IMSB, WISDM, and ERICA datasets, respectively.

Positive and unlabelled (PU) learning for multi-variate time series classification refers to build a binary classification model when only a small set of positive and a large set of unlabelled samples are accessible at training stage. Different from binary semi-supervised scenario in which the training set contains labelled samples from both positive and negative classes, in the PU learning setting, only positive samples are labelled due to cost-restriction or issues related to defining what belongs to the negative class. With the objective to deal with this challenging task, here, we propose a new deep learning framework, referred as DMTS-PUL. Our method has two different steps: firstly, it selects a set of reliable negative samples from the set of unlabelled data and, successively, it iteratively enriches the training data by selecting pseudo-labels to train a binary classification model via self-training. Experimental evaluations on several benchmarks have highlighted the quality of DMTS-PUL w.r.t. competing approaches and the obtained findings have pointed out the suitability of our proposal when only small amounts of positive labelled samples are available.

Artificial intelligence is becoming increasingly important in the air combat domain. Most air combat research now assumes that all aircraft information is known. In practical applications, however, some aircraft information, such as their position, attitude, velocity, etc., can be incorrect or impossible to obtain due to realistic limitations and sensor errors. In this paper, we propose a deep reinforcement learning-based framework for developing a model capable of performing within visual range (WVR) air-to-air combat under the conditions of a partially observable Markov decision process (POMDP) with insufficient information. To deal robustly with such a situation, we use recurrent neural networks and apply a soft actor-critic (SAC) algorithm to cope effectively with realistic limitations and sensor errors. Additionally, to raise the efficiency and effectiveness of learning, we apply the curriculum learning technique to restrict the scope of exploration in state space. Finally, simulations and experiments show that the proposed techniques can deal with practical problems caused by sensor limitations and errors in a noisy environment while also being efficient and effective in reducing the training time for learning.

Water stress and in particular drought are some of the most significant factors affecting plant growth, food production, and thus food security. Furthermore, the possibility to predict and shape irrigation on real plant demands is priceless. The objective of this study is to characterize, classify, and forecast water stress in tomato plants by means of in vivo real time data obtained through a novel sensor, named bioristor, and of different artificial intelligence models. First of all, we have applied classification models, namely Decision Trees and Random Forest, to try to distinguish four different stress statuses of tomato plants. Then, we have predicted, through the help of recurrent neural networks, the future status of a plant when considering both a binary (water stressed and not water stressed) and a four-status scenario. The obtained results are very good in terms of accuracy, precision, recall, F-measure, and of the resulting confusion matrices, and they suggest that the considered novel data and features coming from the bioristor, together with the used machine and deep learning models, can be successfully applied to real-world on-the-field smart irrigation scenarios in the future.

Cloud computing is a promising platform for running massive workflow applications based on a pay-per-use model. In cloud computing, the reduction of energy consumption and providing security to workflow scheduling are the key research areas. The primary focus of the existing algorithms, viz., particle swarm optimization (PSO), crow Search optimization (CSO) and other non-metaheuristic algorithms like Round Robbin (RR), SJF, Min-Min, Min-Max etc., is based on the execution time and cost of the workflow applications as a budget constraint. However, these algorithms failed to adequately determine energy consumption, resource utilization, and security in workflow scheduling. To address this issue, a multi-objective scheduling framework is proposed. In this paper, the framework performs dynamic workflow scheduling using universal unique identification- Blake (UUID-Blake), Manhattan Distance-Partition around algorithm (MD-PAM), Linear Scaling-Crow Search Optimization (LS-CSO), Anova-Recurrent Neural Network. The implementation of this framework was achieved in three phases (Phase 1, Phase 2, and Phase 3). Phase 1 is about user registration and authentication using UUID-Blake, which enhances security by allowing legitimate users into the cloud environment. Phase 2 deals with clustering and resource monitoring using MD-PAM and A-RNN, to reduce makespan the similar tasks are clustered using task length and maximize the resource utilization by predicting the resource availability. Phase 3 deals with the scheduling of dynamic workflows using LS-CSO by selecting suitable virtual machines. We have considered the heterogeneous computing scheduling problem (HCSP) and grid workload archive (GWA)-T-12 Bitbrains datasets for comparing our proposed framework with existing works. Based on the result analysis, the proposed LS-SCO outperformed when compared with the algorithms CSO, PSO and RR has achieved better performance.

Protein structure prediction (PSP) is a vital challenge in bioinformatics, structural biology and drug discovery. Protein secondary structure (SS) prediction is critical since three-dimensional (3D) structures are primarily made up of secondary structures. With the advancement of deep learning approaches, SS classification accuracy has been significantly improved. Many existing methods use an ensemble of complex neural networks to improve SS prediction. Because of the high dimensionality of the hyperparameter space, deep neural networks with complex architectures are typically challenging to train effectively. Also, predicting secondary structures in the boundary regions between different types of SS is challenging. This study presents Multi-S3P, which employs bidirectional Long-Short-Term-Memory (BILSTM) and Convolutional Neural Networks (CNN) with a self-attention mechanism to improve the secondary structure prediction using an effective training strategy to capture the unique characteristics of each type of secondary structure and combine them more effectively. The ensemble of CNN and BILSTM can learn both contextual information and long-range interactions between the residues. In addition, using a self-attention mechanism allows the model to focus on the most important features for improving performance. We used the SPOT-1D dataset for the training and validation of our model using a set of four input features derived from amino acid sequences. Further, the model was tested on four popular independent test datasets and compared with various state-of-the-art predictors. The presented results show that Multi-S3P outperformed the other methods in terms of Q3, Q8 accuracy and other performance metrics, achieving the highest Q3 accuracy of 87.57% and a Q8 accuracy of 77.56% on the TEST2016 test set. More importantly, Multi-S3P demonstrates high performance in SS boundary regions. Our experiment also demonstrates that the combination of different input features and a multi-network-based training strategy significantly improved the performance.

Music is composed of a set of regular sound waves, which are usually ordered and have a large number of repetitive structures. Important notes, chords, and music fragments often appear repeatedly. Such repeated fragments (referred to as motifs) are usually the soul of a song. However, most music generated by existing music generation methods can not have distinct motifs like real music. This study proposes a novel multi- encoders model called Motif Transformer to generate music containing more motifs. The model is constructed using an encoder-decoder framework that includes an original encoder, a bidirectional long short term memory-attention encoder (abbreviated as bilstm-attention encoder), and a gated decoder. Where the original encoder is taken from the transformer’s encoder and the bilstm-attention encoder is constructed from the bidirectional long short-term memory network (BILSTM) and the attention mechanism; Both the original encoder and the bilstm-attention encoder encode the motifs and input the encoded information representations to the gated decoder; The gated decoder decodes the entire input of the music and the information passed by the encoders and enhances the model’s ability to capture motifs of the music in a gated manner to generate music with significantly repeated fragments. In addition, in order to better measure the model’s ability of generating motifs, this study proposes an evaluation metric called used motifs. Experiments on multiple music field metrics show that the model proposed in this study can generate smoother and more beautiful music, and the generated music contains more motifs.

High-resolution, multi-channel remote sensing images have complex spatial layout and channel structure. Traditional Convolutional Neural Networks (CNNs) methods do not make full use of the features of remote sensing images, ignoring contextual relevance and globality, resulting in insufficient features discrimination and limiting the accuracy of scene classification. To address these issues, we propose a discriminative network HCA-TSA based on head-tail global joint dual attention mechanism. Considering the characteristics between the spatial dimension and the channel dimension, the channel attention module Head-Channel Attention (HCA) and the spatial attention module Tail-Spatial Attention (TSA) are proposed. This network uses the context information discrimination ability of the Gated Recurrent Unit (GRU) module and the structure of the global layout to learn and output the importance weights of different channels and spatial features, pay more attention to the salient features in the remote sensing image, and ignore the insignificant features, thus improve the discriminative power of feature representation. The proposed network can be connected to any benchmark CNNs, and the whole network structure can be trained end-to-end. Through a comprehensive comparative experiment on three public datasets with large differences, AID, UC-Merced and HSRS-SC, the highest classification accuracies in the experimental results are 97.68%, 99.41% and 98.38%, respectively. The results show that the proposed method can effectively discriminate scenarios and obtain competitive classification results, which verifies the effectiveness of the proposed method.

The precise orientation estimation of moving objects in 3D space is crucial for the inertial and magnetic measurement unit (IMMU)-based motion capture applications. Disturbance components such as external acceleration and magnetic disturbance in the sensor signal deteriorate the estimation accuracy. While conventional filters such as Kalman filters and complementary filters successfully deal with these issues, there are still much room to improve the estimation performance. One alternative approach involves training an end-to-end neural network (NN) using raw IMMU datasets based on ground truth measurements. In this study, we propose an end-to-end NN to estimate a 3D orientation over time without an additional conventional filter. The architecture of the proposed NN comprises two separate recurrent NNs in a parallel configuration. The proposed parallel network can independently estimate two vectors, corresponding to the attitude and heading, and then combine the two vectors to form a direction cosine matrix (DCM) that represents the 3D orientation. The proposed DCM-based NN has been experimentally verified under various disturbance conditions. The orientation estimation accuracy of the proposed method was superior to those of the conventional filters as well as that of the quaternion-based NN.

The traditional support vector machine (SVM) requires manual feature extraction to improve classification performance and relies on the expressive power of manually extracted features. However, this characteristic poses limitations in complex Industrial Internet of Things (IIoT) environments. Traditional manual feature extraction may fail to capture all relevant information, thereby restricting the application effectiveness of SVM in IIoT settings. CNN-RNN, as a deep learning network capable of simultaneously extracting spatial and temporal features, can alleviate researchers’ burden. In this paper, we propose a novel intrusion detection system (IDS) framework based on anomalies, called CRSF. The framework’s pre-training part employs a dimension transformation function to process input data into two-dimensional images. Two-dimensional convolutional kernels are then employed to extract spatial features, and the feature sequences are passed to an RNN to capture richer temporal features. After sufficient pre-training, SVM is used as a classifier to map the pre-training data from the feature space to a high-dimensional space and learn nonlinear decision boundaries, enabling the framework to accurately differentiate feature representations of different classes. Simulation experiments on the TON_IoT-Datasets demonstrate the effectiveness of the CRSF framework in intrusion detection. When using the “linear” kernel function in SVM, the framework achieves an accuracy, F1-score, and AUC of 0.9959, 0.9959, and 0.9977, respectively, indicating its capability and superiority in intrusion detection.

The importance of accurate prediction of power load variations for ensuring the reliability and rationality of power supply cannot be overstated. The GCES-adRNNe (Grey Relational Analysis Contextually Enhanced ES-adRNNe) is proposed as a load forecasting model that incorporates context augmentation on similar load days. Firstly, the selection of similar load days as representative sequences using the comprehensive grey correlation analysis method is enforced. Furthermore, context information is extracted from the representative sequences, and the individual sequences are dynamically adjusted to accommodate the main track prediction using XGBoost. Moreover, a stacked recurrent neural network (RNN) architecture with multiple layers is used. It incorporates attention-enhanced gated recurrent units (GRU) to capture various dependencies in the time series, such as short-term, long-term, and seasonal dependencies. The GRU also dynamically weight input information. Finally, the prediction results are post-processed. Experimental results indicate that the proposed model yields improvements in terms of RMSE, MAPE, StdPE, and other aspects compared to other load forecasting models. It is indicated that the model, when considering similar load days, is capable of capturing the trends and characteristics of electricity load changes more accurately, thereby providing strong support for ensuring the reliability and rationality of power supply.

Brain Epilepsy seizure is a critical disorder, which is an uncontrolled burst of electrical activity of brain. The early detection of brain seizure can save the life of humans. The electroencephalogram (EEG) signals may be used to automatically identify brain seizures, which is one of the most prominent solutions for this issue. However, the conventional methods are failed to classify the brain seizure effectively. So, this work implemented the Brain Epilepsy Seizure-Detection-Network (BESD-Net) using deep learning, recurrent learning properties. Initially, the dataset pre-processing is performed, which eliminates the noise, unwanted data from EEG dataset. Then, the deep learning based customized convolution neural network (CCNN) is trained on the pre-processed EEG data for precise extraction of disease correlated features. The machine learning based exhaustive random forest (ERF) feature selection is used to optimize the features obtained from the CCNN, which are highly correlated with disease dependent properties. In conclusion, the recurrent neural network (RNN) based bi-directional long short-term memory (BLSTM) is used in order to detect brain seizures from the chosen ERF features. Training and testing of suggested methodology had made use of CHB-MIT Scalp EEG Database. The aforementioned model has achieved the values of 98.36%, 97.54%, 97.91%, 98% and 95.08% respectively for precision, sensitivity, F1-Score, accuracy and specificity. The findings of the simulations demonstrate that the suggested BESD-Net led to superior performance when compared to the technologies that are already in use.

This paper presents a security control method of Grid energy storage based on neural network model. The clean energy consumption effect of hybrid ESS was studied through a load forecasting method based on improved RNN (Recurrent Neural Network). Based on the current mainstream deep learning architecture, deep RNNs with different ring kernels were established to optimize the hybrid ESS model. The research results indicate that the curve obtained by this method is smoother after peak shaving and valley filling. The planned variance of this method is 43.037, which is 7.37% lower than the load variance of the literature method. It improves the stability of the distribution network operation and the absorption of photovoltaic and wind energy, reducing the cost of exceeding the limit of battery losses. The optimized operation status of microgrids can reduce costs, improve the security of microgrid systems, and better meet the proposed optimization goals.

We propose a novel graph-oriented machine learning algorithm which we use for estimating the performance of a recurrent memory cell on a given task. Recurrent neural networks have been successfully used for solving numerous tasks and usually, for each new problem, generic architectures are used. Adapting the architecture could provide superior results, but would be time-consuming if it would not be automated. Neural architecture search algorithms aim at optimizing the architectures for each specific task, but without a fast performance estimation strategy it is difficult to discover high-quality architectures, as evaluating each candidate takes a long period of time. As a case study, we selected the task of sentiment analysis on tweets. Analyzing the sentiments expressed in posts on social networks offers important insights into what are the opinions on different topics and this has applications in numerous domains. We present the architecture of the estimation algorithm, discussing each component. Using this algorithm, we were able to evaluate one million recurrent memory cell architectures and we discovered novel designs that obtain good performances on sentiment analysis. We describe the discovered design that obtains the best performances. We also describe the methodology that we designed, such that it can be applied to other tasks.

Cloud computing is a technology for efficiently using computing infrastructures and a business model for selling computing resources and services. However, intruders find such complex and distributed infrastructures appealing targets for cyber-attacks. Cyber-attacks are severe threats that can jeopardize the quality of service provided to clients and compromise data integrity, confidentiality, and availability. Cyber-attacks are becoming more complex, making it more challenging to detect intrusions effectively. Due to the high traffic and increased malicious activities on the Internet, a single Intrusion Detection System (IDS) can be overwhelmed. Despite the various Deep Learning (DL) approaches that have been proposed as alternative solutions, there are still pertinent security issues to be addressed especially in federated cloud computing domains. This work proposes a Secure Federated Intrusion Detection Model Version 1 (SecFedIDM-V1) using blockchain technology and Bidirectional Long Short-Term Memory (BiLSTM) Recurrent Neural Network (RNN). The Cobourg Intrusion Detection Dataset (CIDDS) was acquired, pre-processed and split into 60:20:20, 70:15:15, and 80:10:10 for training, testing, and validation respectively to develop the proposed intrusion traffic classification component of the proposed model. The developed SecFedIDM-V1 was later deployed as a Python-based web application that captures network packets for classifying attacks into normal or an attack type. The attack packets are recorded in a Hyperledger Fabric (a private blockchain technology) to serve as a signature database to be used by other nodes in the network. From the evaluation results of the intrusion classifier, the 80:10:10 BiLSTM network performed better than GRU with a Precision of 0.99624, Recall of 0.99906, F1 Score of 0.99614, False Positive Rate (FPR) of 0.00094, False Negative Rate (FNR) of 0.00395 and True Positive Rate (TPR) of 0.99605. The SecFedIDM-V1 can be deployed alongside Firewalls in a federated cloud computing environment to reinforce the security of the infrastructure.

The paint industry comprises an elaborate supply chain involving various activities such as raw material procurement, manufacturing, and distribution. In addition, the accuracy of demand prediction significantly impacts supply chain management. A recurrent neural network (RNN) is a powerful method that learns intricate patterns through vast amounts of historical data and provides a prediction, demonstrating excellent performance in demand prediction. However, standard RNN-based demand predictions are limited by many outliers that occur due to the characteristics of the paint industry. Unexpected events such as a factory fire cause rapid fluctuations in paint demand and are difficult to pre-observe. To overcome these limitations, we propose a novel approach that employs clustering to identify demand time series data with similar characteristics and applies statistical outlier adjustment to allow the prediction model to learn the complex patterns of actual demand. The prediction target is the sum of sales for 15 days in the future, and sales data for four paint products are used to evaluate the proposed approach. Experimental results demonstrate a satisfactory prediction accuracy improvement ranging from 100.9% to 152.4%. In addition, unlike the RNN-based models used for comparison, the proposed method is more robust against actual demand fluctuations.

Hydrogen refueling stations are pivotal for renewable energy and carbon neutrality; however, they encounter challenges owing to equipment malfunctions. This study addresses the use of time-series forecasting techniques to predict and diagnose critical equipment failure at these stations. An analysis of the station equipment was conducted to create scenarios for potential malfunctions in compression equipment. Techniques such as the Recurrent Neural Network (RNN), Long Short-Term Memory network (LSTM), and Gated Recurrent Unit (GRU) have been employed to forecast the conditions of high-pressure compression equipment. Deep neural networks were constructed to enhance prediction accuracy, typically achieving an error margin of 0.01. Multi-step predictions using autoregression were utilized to bolster equipment resilience against aging and progressive failures. Autoregressive prediction models, particularly those using LSTMs and GRUs, outperform RNNs. However, predictions may be subject to errors due to algorithmic limitations and environmental factors. This study introduces a stochastic forecasting approach that, utilizes Gaussian distributions to predict probability distributions, not single-point estimates. This method yielded a 95% prediction interval with a standard deviation of 1.96. The reliability of multi-time step forecasts is significantly improved by adopting stochastic autoregressive forecasting and establishing prediction intervals. The proposed model enhances not only the accuracy of equipment failure predictions but also proactive maintenance, thus reducing downtime and boosting the efficiency of the hydrogen fuel infrastructure, which contributes to the wider utilization of hydrogen as a clean energy source.

File fragment type identification is an important step in file carving and data recovery. Machine learning techniques, especially neural networks, have been utilized for this problem, some with very promising results. This paper presents a novel neural network architecture for identifying file fragment types using a combination of byte embeddings as well as recurrent and convolutional elements. The corresponding classification model, ByteRCNN, has been trained on the publicly available file fragment FiFTy dataset and evaluated in closed-set and open-set recognition settings using FiFTy and other available file fragment datasets. Evaluation results have demonstrated that ByteRCNN can compete with state-of-the-art models described in literature in terms of classification accuracy, with 71.1% average accuracy on 512-byte fragments and 83.9% average accuracy on 4,096-byte fragments from the FiFTy dataset. When evaluated on other publicly available datasets in closed-set and open-set recognition settings, ByteRCNN performs similarly or slightly better than the FiFTy classification model. Obtained results overall suggest that ByteRCNN is a competitive file fragment classification model, but they also reveal that there is still plenty of space for further improving file type identification methods using more complex datasets or in open-set recognition settings. ByteRCNN is publicly available at https://github.com/kristian-fer/ByteRCNN.

A rapid recurrent neural network (RNN)-based physical feasibility test algorithm for base parameters during the identification process of a robot dynamic model is proposed in this paper. Firstly, related physical constraints such as inertia tensor, drive chain inertia, and friction are combined into the formulation of linear matrix inequality (LMI) to examine the physical consistency of the base parameters. The optimization problem of LMI is then solved by a matrix-oriented gradient-type RNN. Since the network structure of this type of RNN is simple and the process for solving the optimization problem of LMI is parallel distributed, the physical feasibility test can be completed more quickly than when utilizing commonly used semi-definite programming techniques. By taking advantage of highly efficient computation capabilities, the proposed physical feasibility test algorithm is particularly suitable for identification approaches that require rapid feasibility assessment such as on-line identification methods or optimization-based identification methods. An evolutionary algorithm-based identification method is used as an illustrative example to assess the performance of the proposed approach. In addition, a stability proof for the proposed gradient-type RNN is also provided. Results of the experiments conducted on a 6-DOF industrial robot manipulator verify the effectiveness of the proposed approach.

Video quality assessment (VQA) has received remarkable attention recently. Most of the popular VQA models employ recurrent neural networks (RNNs) to capture the temporal quality variation of videos. However, each long-term video sequence is commonly labeled with a single quality score, with which RNNs might not be able to learn long-term quality variation well: What’s the real role of RNNs in learning the visual quality of videos? Does it learn spatio-temporal representation as expected or just aggregating spatial features redundantly? In this study, we conduct a comprehensive study by training a family of VQA models with carefully designed frame sampling strategies and spatio-temporal fusion methods. Our extensive experiments on four publicly available in- the-wild video quality datasets lead to two main findings. First, the plausible spatio-temporal modeling module (i. e., RNNs) does not facilitate quality-aware spatio-temporal feature learning. Second, sparsely sampled video frames are capable of obtaining the competitive performance against using all video frames as the input. In other words, spatial features play a vital role in capturing video quality variation for VQA. To our best knowledge, this is the first work to explore the issue of spatio-temporal modeling in VQA. [ABSTRACT FROM AUTHOR]

Enhancing the diversity of sentences to describe video contents is an important problem arising in recent video captioning research. In this paper, we explore this problem from a novel perspective of customizing video captions by imitating exemplar sentence syntaxes. Specifically, given a video and any syntax-valid exemplar sentence, we introduce a new task of Syntax Customized Video Captioning (SCVC) aiming to generate one caption which not only semantically describes the video contents but also syntactically imitates the given exemplar sentence. To tackle the SCVC task, we propose a novel video captioning model, where a hierarchical sentence syntax encoder is first designed to extract the syntactic structure of the exemplar sentence, then a syntax conditioned caption decoder is devised to generate the syntactically structured caption expressing video semantics. As there is no available syntax customized groundtruth video captions, we tackle such a challenge by proposing a new training strategy, which leverages the traditional pairwise video captioning data and our collected exemplar sentences to accomplish the model learning. Extensive experiments, in terms of semantic, syntactic, fluency, and diversity evaluations, clearly demonstrate our model capability to generate syntax-varied and semantics-coherent video captions that well imitate different exemplar sentences with enriched diversities. Code is available at https://github.com/yytzsy/Syntax-Customized-Video-Captioning. [ABSTRACT FROM AUTHOR]

In this article, a new idea of chaos synchronization and chaos-based secure communication is developed. First, the chaotic master system is used as a transmitter in chaos-based secure communication, then a drive signal is constructed, and the information message is encrypted into the drive signal to form a transmitted signal for secure communication. Second, in the receiver, a recurrent Takagi–Sugeno–Kang (TSK) fuzzy brain emotional learning cerebellar model articulation controller (RTFBECAC) is developed to control the slave system to follow the master system in the transmitter. Third, after descripting the chaotic signal, the embedded information message can be recovered. Besides, the stability problem is analyzed in detail based on the stability theory. Finally, two simulation examples, including audio signal and image, are introduced to illustrate the effectiveness and the advantages of the proposed method. [ABSTRACT FROM AUTHOR]

In this article, we present a novel decomposed vector rotation (DVR)-based recurrent neural network behavioral model for digital predistortion (DPD) of radio frequency (RF) power amplifiers (PAs) in wideband scenarios. By representing memory terms of DVR with recurrent states and redesigning the piecewise modeling, we propose a novel recurrent DVR scheme. To ensure stable operation and enhanced modeling accuracy, we integrate the recurrent DVR into the gated learning mechanism of the modified Just Another NETwork (JANET). Experimental results confirm that the proposed DVR-JANET model provides much improved linearization performance with significantly reduced model complexity compared with the recent existing models. [ABSTRACT FROM AUTHOR]

Recently, model-based predictive current control (MB-PCC) has been presented as a good alternative to classical control algorithms in terms of simplicity and performance reliability. However, MB-PCC suffers from the high dependence on system parameters, which may deteriorate its performance under parameters variations. On the other hand, synchronous reluctance motors (SynRMs) are susceptible to suffer from inductances variations due to the magnetic saturation. Accordingly, in this article a new model-free predictive current control of SynRMs based on a recurrent neural network (RNN-PCC) is developed and proposed. The proposed RNN-PCC relies on the identification of the SynRM currents without considering any parameters. Simulation and experimental results show that both RNN-PCC and MB-PCC have similarly excellent dynamics, while better control performance and tracking errors can be achieved thanks to the proposed RNN-PCC. [ABSTRACT FROM AUTHOR]

Due to the rapid development of technology, social media has become more and more common in human daily life. Social media is a platform for people to express their feelings, feedback, and opinions. To understand the sentiment context of the text, sentiment analysis plays the role to determine whether the sentiment of the text is positive, negative, neutral or any other personal feeling. Sentiment analysis is prominent from the perspective of business or politics where it highly impacts the strategic decision making. The challenges of sentiment analysis are attributable to the lexical diversity, imbalanced dataset and long-distance dependencies of the texts. In view of this, a data augmentation technique with GloVe word embedding is leveraged to synthesize more lexically diverse samples by similar word vector replacements. The data augmentation also focuses on the oversampling of the minority classes to mitigate the imbalanced dataset problems. Apart from that, the existing sentiment analysis mostly leverages sequence models to encode the long-distance dependencies. Nevertheless, the sequence models require a longer execution time as the processing is done sequentially. On the other hand, the Transformer models require less computation time with parallelized processing. To that end, this paper proposes a hybrid deep learning method that combines the strengths of sequence model and Transformer model while suppressing the limitations of sequence model. Specifically, the proposed model integrates Robustly optimized BERT approach and Long Short-Term Memory for sentiment analysis. The Robustly optimized BERT approach maps the words into a compact meaningful word embedding space while the Long Short-Term Memory model captures the long-distance contextual semantics effectively. The experimental results demonstrate that the proposed hybrid model outshines the state-of-the-art methods by achieving F1-scores of 93%, 91%, and 90% on IMDb dataset, Twitter US Airline Sentiment dataset, and Sentiment140 dataset, respectively.

Aircraft trajectory prediction is a challenging problem in air traffic control, especially for conflict detection. Traditional trajectory predictors require a variety of inputs such as flight-plans, aircraft performance models, meteorological forecasts, etc. Many of these data are subjected to environmental uncertainties. Further, limited information about such inputs, especially the lack of aircraft tactical intent, makes trajectory prediction a challenging task. In this work, we propose a deep learning model that performs trajectory prediction by modeling and incorporating aircraft tactical intent. The proposed model adopts the encoder-decoder architecture and makes use of the convolutional layer as well as Gated Recurrent Units (GRUs). The proposed model does not require explicit information about aircraft performance and wind data. Results demonstrate that the provision of enriched aircraft intent, together with appropriate model design, could improve the prediction error up to 30% at a prediction horizon of 10 minutes (from 4.9 nautical miles to 3.4 nautical miles). The model also guarantees the mean error growth rate with increasing look-ahead time to be lower than 0.2 nautical miles per minute. In addition, the model offers a very low variance in the prediction, which satisfies the variance-standard specified by EUROCONTROL (EU Organization for Safety and Navigation of Air Traffic) for trajectory predictors. The proposed model also outperforms the state-of-the-art trajectory prediction model, where the Root Mean Square Error (RMSE) is reduced from 0.0203 to 0.0018 for latitude prediction, and from 0.0482 to 0.0021 for longitude prediction in a single prediction step of 15 seconds look-ahead. We showed that the pre-trained model on ADS-B data maintains its high performance, in terms of cross-track and along-track errors, when being validated in the Bluesky Air Traffic Simulator. The proposed model would significantly improve the performance of conflict detection systems where such trajectory prediction models are needed.

Explaining dynamic relationships between input and output variables is one of the most important issues in time dependent domains such as economic, finance and so on. In this work, we propose a novel locally adaptive interpretable deep learning architecture that is augmented by recurrent neural networks to provide model explainability and high predictive accuracy for time-series data. The proposed model relies on two key aspects. First, the base model should be a simple interpretable model. In this step, we obtain our base model using a simple linear regression and statistical test. Second, we use recurrent neural networks to re-parameterize our base model to make the regression coefficients adaptable for each time step. Our experimental results on public benchmark datasets showed that our model not only achieves better predictive performance than the state-of-the-art baselines, but also discovers the dynamic relationship between input and output variables.

Automatic speech recognition (ASR) is one of the most demanding tasks in natural language processing owing to its complexity. Recently, deep learning approaches have been deployed for this task and have been proven to outperform traditional machine learning approaches such as Artificial Neural Network (ANN). In particular, deep-learning methods such as long short-term memory (LSTM) have achieved improved ASR performance. However, this method is limited to processing continuous input streams. Traditional LSTM requires four (4) linear layers (multilayer perceptron (MLP) layer) per cell with a large memory bandwidth for each sequence time step. LSTM cannot accommodate the many computational units required for processing continuous input streams because the system does not have sufficient memory bandwidth to feed the computational units. In this study, an enhanced deep learning LSTM recurrent neural network (RNN) model was proposed to resolve this shortcoming. In the proposed model, the RNN is incorporated as a “forget gate” to the memory block to allow the resetting of cell states at the beginning of the sub-sequences. This enables the system to process continuous input streams efficiently without necessarily increasing the required bandwidths. In the proposed model, the standard architecture of the LSTM network is modified to effectively use the model parameters. Some CNN-based and sequential models were used on the same dataset, and the models were compared with the proposed model. LSTM-RNN outperformed the other deep learning models with an accuracy of 99.36% on the well-established public benchmark spoken English digit dataset.