Search

Low Earth orbit satellites are potentially impacted by atmospheric drag due to short-term enhancements in thermospheric density induced by solar irradiance and solar wind disturbances, affecting the design of launch missions to the safe landing of spacecraft on Earth. We utilize hourly solar and geomagnetic indices and thermospheric density as measured by Gravity Recovery and Climate Experiment Follow-On (GRACE-FO) satellites during the minimum and rising phases of solar activity from 2018 to 2022. Bidirectional Long Short-Term Memory (Bi-LSTM) and Gated Recurrent Unit (GRU) neural networks based on empirical mode decomposition (EMD) features are used to predict the thermospheric density. We found that the EMD of thermospheric density provided robust feature for the GRU and the Bi-LSTM models. The predictions are more effective in the rising phase when the thermospheric density is typically high which is of interested in satellite drags. The inputs of thermospheric density and its intrinsic mode functions (IMFs) with solar and geomagnetic indices improved prediction abilities for the rising phase, while only the IMFs of density or the geomagnetic indices is sufficient for the minimum phase. For categories based on disturbed and quiet geomagnetic conditions, the best prediction is for the coronal mass ejection (CME) event. The maximum values of R2 is in the stream interaction region-high speed solar wind event for both Bi-LSTM and GRU models with correlation coefficients 0.914 and 0.922, respectively. The Bi-LSTM is more suitable for predicting the thermospheric density during "SpaceX storm" of consecutive CME-CME geomagnetic storms, while the temporal-dependent variations in the density are accurately predicted by the GRU model. Predictions by both deep learning models are more accurate than by the NRLMSIS 2.0 model. This study reveals the main physical mechanisms driving the short-term variations in the thermospheric density. [ABSTRACT FROM AUTHOR]

This study aims to estimate P M 2.5 ​ concentrations in Delhi, addressing the escalating menace of air pollution that poses a critical challenge to environmental sustainability. The analysis utilizes time-series data of P M 2.5 ​ concentrations, encompassing historical air quality measurements in Delhi over the year 2019 to 2023. Leveraging deep learning methodologies, we employ a multifaceted evaluation framework that includes R-squared (R2), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Mean Squared Error (MSE) to assess model performance. Our novel hybrid approach amalgamates Bidirectional Long-Short-Term Memory (Bi-LSTM) and Gated Recurrent Unit (GRU) architectures to enhance estimation accuracy. The proposed Bi-LSTM-GRU model demonstrates superior performance in P M 2.5 ​ estimation, yielding an RMSE of 0.37, R2 of 0.75, MAE of 8.22, MAPE of 34.69, and MSE of 0.13. Statistical analyses, including the Friedman test, affirm the pre-eminence of our proposed model. The model's enhanced performance supports more effective air quality management strategies, contributing to healthier urban environments and providing a robust tool for policymakers to devise targeted interventions and regulations. [ABSTRACT FROM AUTHOR]

In human–human interactions, detecting emotions is often easy as it can be perceived through facial expressions, body gestures, or speech. However, in human–machine interactions, detecting human emotion can be a challenge. To improve this interaction, Speech Emotion Recognition (SER) has emerged, with the goal of recognizing emotions solely through vocal intonation. In this work, we propose a SER system based on deep learning approaches and two efficient data augmentation techniques such as noise addition and spectrogram shifting. To evaluate the proposed system, we used three different datasets: TESS, EmoDB, and RAVDESS. We employe several algorithms such as Mel Frequency Cepstral Coefficients (MFCC), Zero Crossing Rate (ZCR), Mel spectrograms, Root Mean Square Value (RMS), and chroma to select the most appropriate vocal features that represent speech emotions. Three different deep learning models were imployed, including MultiLayer Perceptron (MLP), Convolutional Neural Network (CNN), and a hybrid model that combines CNN with Bidirectional Long-Short Term Memory (Bi-LSTM). By exploring these different approaches, we were able to identify the most effective model for accurately identifying emotional states from speech signals in real-time situation. Overall, our work demonstrates the effectiveness of the proposed deep learning model, specifically based on CNN+BiLSTM enhanced with data augmentation for the proposed real-time speech emotion recognition. [ABSTRACT FROM AUTHOR]

In this article, a neuroadaptive sliding mode control (NSMC) strategy based on recurrent neural network (RNN) for robustly and adaptively tracking the desired position and attitude of an uncertain tilting quadrotor unmanned aerial vehicle (TQUAV) with unknown controllers is presented. The main contribution of this article is the real‐time adjustment of unknown flight controllers using the approximation characteristics of RNN, in which the derived approximation errors of RNN are sufficiently estimated by adaptive control method that can reduce or eliminate the impact of error terms on the evolution of closed‐loop systems. Especially, Lyapunov stability analysis is greatly simplified compared to existing methods and does not require amplification or reduction. Finally, the superior performance of the NSMC strategy was verified by comparing simulation results. [ABSTRACT FROM AUTHOR]

Steel corrosion is a time-series problem where the current corrosion state is influenced by past corrosion and environmental factors. Most models incorporate time as a variable but fail to comprehensively consider past states and spatial-temporal factors. This study collects continuously monitored corrosion data from the literature and develops time-variant corrosion rate prediction models using traditional machine learning (TML) and recurrent neural networks (RNN). The TML model captures temporal variations, while the RNN model incorporates past and current corrosion factors, focusing on different lookback lengths. Empirical and electrochemical models are also compared to evaluate efficacy. The results indicate superior predictive capabilities for bothTML and RNN models. The TML model yields an RMSE of 0.26 μA/cm2 and a MAPE of 11.27%, while the RNN model achieves 0.22 μA/cm2 and 8.15%. In contrast, empirical and electrochemical models yield suboptimal RMSE (1.48 μA/cm2, 0.53 μA/cm2) and MAPE (77.31%, 35.05%). Analysis of Simple-RNN and LSTM models, reveals optimal performance when considering corrosion factors from the preceding two-time intervals. These results underscore the importance of incorporating past corrosion factors and demonstrate the enhanced efficacy of data-driven approaches, particularly LSTM, over traditional models. [ABSTRACT FROM AUTHOR]

Recurrent neural networks (RNNs) based on model neurons that communicate via continuous signals have been widely used to study how cortical neural circuits perform cognitive tasks. Training such networks to perform tasks that require information maintenance over a brief period (i.e., working memory tasks) remains a challenge. Inspired by the robust information maintenance observed in higher cortical areas such as the prefrontal cortex, despite substantial inherent noise, we investigated the effects of random noise on RNNs across different cognitive functions, including working memory. Our findings reveal that random noise not only speeds up training but also enhances the stability and performance of RNNs on working memory tasks. Importantly, this robust working memory performance induced by random noise during training is attributed to an increase in synaptic decay time constants of inhibitory units, resulting in slower decay of stimulus-specific activity critical for memory maintenance. Our study reveals the critical role of noise in shaping neural dynamics and cognitive functions, suggesting that inherent variability may be a fundamental feature driving the specialization of inhibitory neurons to support stable information processing in higher cortical regions. [ABSTRACT FROM AUTHOR]

Electrical engineering models often rely on complex circuit configurations that facilitate the dynamic flow of electrically charged particles within a closed conductive network. These circuits serve as essential tools for simulating and analyzing diverse electrical systems and components. This paper introduces a study on nonlinear fractional circuits modeling through the development of a stochastic neuro-computational artificial intelligent-based solver to address mathematical models governing the Fractional order Caputo–Fabrizio stiff electric circuit model (FO-CFSECM) by manipulating the knacks of layered recurrent neural networks (LRNNs) trained with Gradient-based local search algorithm (GLA). In fractional calculus, the Caputo–Fabrizio (CF) fractional order derivative (FOD) emerges as a powerful instrument, binding its capabilities to deliver remarkably accurate solutions for fractional stiff systems. The objective of this work is to exploit the numerical treatment comprehensively for the dynamics of fractal Resistor–Capacitor (RC) and fractal Resistor–Inductor (RL) circuit models by introducing the CF fractional operator. Through the application of artificial intelligence-based soft computing and advanced back-propagative deep neural networks, a deeper understanding of the behavior and distinctive characteristics inherent in these models is sought. The Levenberg–Marquardt optimizer serves as an efficient training GLA tool for learning of LRNNs weights of fractal RL/RC circuit models. The comparative studies on variants of FO-CFSECM demonstrate that LRNNs achieve an impressive mean square error (MSE) ranging from 10 − 9 to 10 − 1 9 and absolute error (AE) within 10 − 6 to 10 − 8 . The accuracy, reliability, and efficiency of LRNNs for solving the FO-CFSECM were further validated through MSE, AE, controlling parameters of state transitions, error histograms, and correlation measures. [ABSTRACT FROM AUTHOR]

The temporal dominance of sensations (TDS) method captures assessors' real-time sensory experiences during food tasting, while the temporal liking (TL) method evaluates dynamic changes in food preferences or perceived deliciousness. These sensory evaluation tools are essential for understanding consumer preferences but are also resource-intensive processes in the food development cycle. In this study, we used reservoir computing, a machine learning technique well-suited for time-series data, to predict temporal changes in liking based on the temporal evolution of dominant sensations. While previous studies developed reservoir models for specific food brands, achieving cross-brand prediction—predicting the temporal liking of one brand using a model trained on other brands—is a critical step toward replacing human assessors. We applied this approach to coffee products, predicting temporal liking for a given brand from its TDS data using a model trained on three other brands. The average prediction error across all brands was approximately 10% of the maximum instantaneous liking scores, and the mean correlation coefficients between the observed and predicted temporal scores ranged from 0.79 to 0.85 across the four brands, demonstrating the model's potential for cross-brand prediction. This approach offers a promising technique for reducing the costs of sensory evaluation and enhancing product development in the food industry. [ABSTRACT FROM AUTHOR]

Artificial neural networks (ANNs) are one of the most promising tools in the quest to develop general artificial intelligence. Their design was inspired by how neurons in natural brains connect and process, the only other substrate to harbor intelligence. Compared to biological brains that are sparsely connected and that form sparsely distributed representations, ANNs instead process information by connecting all nodes of one layer to all nodes of the next. In addition, modern ANNs are trained with backpropagation, while their natural counterparts have been optimized by natural evolution over eons. We study whether the training method influences how information propagates through the brain by measuring the transfer entropy, that is, the information that is transferred from one group of neurons to another. We find that while the distribution of connection weights in optimized networks is largely unaffected by the training method, neuroevolution leads to networks in which information transfer is significantly more focused on small groups of neurons (compared to those trained by backpropagation) while also being more robust to perturbations of the weights. We conclude that the specific attributes of a training method (local vs. global) can significantly affect how information is processed and relayed through the brain, even when the overall performance is similar. [ABSTRACT FROM AUTHOR]

The rapid expansion of IoT networks, combined with the flexibility of Software-Defined Networking (SDN), has significantly increased the complexity of traffic management, requiring accurate classification to ensure optimal quality of service (QoS). Existing traffic classification techniques often rely on manual feature selection, limiting adaptability and efficiency in dynamic environments. This paper presents a novel traffic classification framework for SDN-based IoT networks, introducing a Two-Level Fused Network integrated with a self-adaptive Manta Ray Foraging Optimization (SMRFO) algorithm. The framework automatically selects optimal features and fuses multi-level network insights to enhance classification accuracy. Network traffic is classified into four key categories—delay-sensitive, loss-sensitive, bandwidth-sensitive, and best-effort—tailoring QoS to meet the specific requirements of each class. The proposed model is evaluated using publicly available datasets (CIC-Darknet and ISCX-ToR), achieving superior performance with over 99% accuracy. The results demonstrate the effectiveness of the Two-Level Fused Network and SMRFO in outperforming state-of-the-art classification methods, providing a scalable solution for SDN-based IoT traffic management. [ABSTRACT FROM AUTHOR]

Ground-based cloud image features high-spatiotemporal resolution, presenting detailed local cloud structures and valuable weather information, which are crucial for meteorological forecasting. However, the inherent fuzziness and dynamism of ground-based clouds have hindered the development of effective prediction algorithms, resulting in low accuracy. This paper presents CloudPredRNN++, a novel method for predicting ground-based cloud dynamics, leveraging a deep spatiotemporal sequence prediction network enhanced with a self-attention mechanism. Initially, a Cascaded Causal LSTM (CCLSTM) with a dual-memory group decoupling structure is designed to enhance the representation of short-term cloud changes. Next, self-attention memory units are incorporated to capture the long-term dependencies and emphasize the non-stationary characteristics of cloud movements. These components are integrated into cloud dynamic feature mining units, which concurrently extract spatiotemporal features to strengthen unified spatiotemporal modeling. Finally, by embedding gradient highway units and adding skip connection, CloudPredRNN++ is constructed into a hierarchical recursive structure, mitigating the gradient vanishing and enhancing the uniform modeling of temporal–spatial features. Experiments on the sequence ground-based cloud dataset demonstrate that CloudPredRNN++ can predict the future cloud state more accurately and quickly. Compared with other spatiotemporal sequence prediction models, CloudPredRNN++ shows significant improvements in evaluation metrics, improving the accuracy of cloud dynamics forecasting and alleviating long-term dependency decay, thus confirming the effectiveness in ground-based cloud prediction tasks. [ABSTRACT FROM AUTHOR]

Detection of anomalies in video surveillance plays a key role in ensuring the safety and security of public spaces. The number of surveillance cameras is growing, making it harder to monitor them manually. So, automated systems are needed. This change increases the demand for automated systems that detect abnormal events or anomalies, such as road accidents, fighting, snatching, car fires, and explosions in real-time. These systems improve detection accuracy, minimize human error, and make security operations more efficient. In this study, we proposed the Composite Recurrent Bi-Attention (CRBA) model for detecting anomalies in surveillance videos. The CRBA model combines DenseNet201 for robust spatial feature extraction with BiLSTM networks that capture temporal dependencies across video frames. A multi-attention mechanism was also incorporated to direct the model's focus to critical spatiotemporal regions. This improves the system's ability to distinguish between normal and abnormal behaviors. By integrating these methodologies, the CRBA model improves the detection and classification of anomalies in surveillance videos, effectively addressing both spatial and temporal challenges. Experimental assessments demonstrate that the CRBA model achieves high accuracy on both the University of Central Florida (UCF) and the newly developed Road Anomaly Dataset (RAD). This model enhances detection accuracy while also improving resource efficiency and minimizing response times in critical situations. These advantages make it an invaluable tool for public safety and security operations, where rapid and accurate responses are needed for maintaining safety. [ABSTRACT FROM AUTHOR]

Forecasting of traffic in cellular network is a significant service for management of available resources strategically in an efficient way. Valuable resources such as link bandwidth and energy are increasing exponentially with increase in usage of cellular data. In this paper, we implement the design of Neural Network for identification of recurrent patterns in different metrics that can be then applied in forecasting of traffic in cellular networks. As this Neural Network design is based on memory and custom architecture, it is able to handle task of prediction in a precise and faster mode in real-time applications such as cellular network traffic forecasting. This work involves a Long Short Term Memory design of Recurrent Neural Network for traffic forecasting in cellular networks. It enhances the performance of the cellular network thereby providing a solution for the service providers as the available resources are utilized in an effective way. Same data set is involved for multiple prediction to analyze the performance of the design and found to be robust than the existing algorithms. [ABSTRACT FROM AUTHOR]

Groundwater models are valuable tools to quantify the response of groundwater level to hydrological stresses induced by climate variability and groundwater extraction. These models strive for sustainable groundwater management by balancing recharge, discharge, and natural processes, with groundwater level serving as a critical response variable. While traditional numerical models are labour-intensive, machine learning and deep learning offer a data-driven alternative, learning from historical data to predict groundwater level variations. The groundwater level in wells is typically recorded as continuous groundwater level time series data and is essential for implementing managed aquifer recharge within a particular region. Machine learning and deep learning are essential tools to generate a data-driven approach to modeling an area, and there is a need to understand if they are the most suitable tools to improve model prediction. To address this objective, the study evaluates two machine learning algorithms - Random Forest (RF) and Support Vector Machine (SVM); and two deep learning algorithms - Simple Recurrent Neural Network (SimpleRNN) and Long Short-Term Memory (LSTM) for modeling groundwater level changes in the West Coast Aquifer System of South Africa. Analysis of regression error metrics on the test dataset revealed that SVM outperformed the other models in terms of the root mean square error, whereas random forest had the best performance in terms of the MAE. In the accuracy analysis of predicted groundwater levels, SVM achieved the highest accuracy with an MAE of 0.356 m and an RMSE of 0.372 m. The study concludes that machine learning and deep learning are effective tools for improved modeling and prediction of groundwater level. Further research can incorporate more detailed geologic information of the study area for enhanced interpretation. [ABSTRACT FROM AUTHOR]

Air pollution, particularly ambient particulate matter with aerodynamic diameter less than 2.5 μm (PM2.5), has emerged as a significant global concern due to its adverse impact on public health and the environment. Rapid urbanization, industrialization, and the increased number of automobiles in the cities have led to a significant enhancement in the PM2.5 concentrations to their hazardous level, which indicates the requirement for early warning systems to reduce exposure. Artificial Intelligence and Machine Learning (AI/ML) have come forth as highly sought-after tools widely utilized for air quality (AQ) forecasting. A deep learning based Recurrent Neural Network (RNN) models are highly being used due to their performance in predicting the AQ from the time series data. The present study evaluated three types of RNNs, namely SimpleRNN, Gradient Recurrent Units (GRU) and Long Short-Term Memory (LSTM) to forecast the PM2.5 in the four major Indian metropolitan cities. This research utilizes the daily in-situ PM2.5 data from national AQ monitoring agency in India, known as Central Pollution Control Board (CPCB) for the period 2018 to 2023. Various atmospheric gases and dispersion factors were employed to train model for the prediction of PM2.5 over the cities of Chennai, Delhi, Hyderabad and Kolkata. The ability of the each RNN model is evaluated and compared with observed data using various statistical parameters such as root mean squared error, mean absolute error, and mean absolute percentage error, coefficient of determination and correlation coefficient. Our findings indicate that all three neural networks can capture future PM2.5 trends consistently, albeit with some uncertainty. GRU was the most proficient in estimating PM2.5 levels in all the cities, followed by LSTM and SimpleRNN. The highest accuracy score was observed over Hyderabad followed by Kolkata, Chennai and Delhi. [ABSTRACT FROM AUTHOR]

The cold start problem in serverless computing leads to increased latency when functions are invoked after being idle. This paper proposes a predictive pre-warming strategy that leverages machine learning and historical data analysis to mitigate the cold start problem. By using a Recurrent Neural Network (RNN) to predict future invocations and a pre-warming scheduler to determine the number of instances to pre-warm, our approach aims to optimize resource utilization and reduce latency. Results show that the proposed method adapts idle-container times efficiently, reducing cold starts and idle periods. The proposed method outperforms OpenWhisk by executing more invocations, demonstrating a 49.52% improvement and enhancing container resource allocation optimization. [ABSTRACT FROM AUTHOR]

As location-based social media has rapidly grown, trip recommendations have become increasingly important. Many recommender systems do not consider the valuable information contained in user reviews, which is a missed opportunity. By incorporating review text, recommendation performance can be improved, and the Cold Start issue could be alleviated. This study proposes a new personalized method for recommending Point of Interest (POI) trips based on user reviews. The proposed method reduces the time required to find POIs using two-level clustering based on the Manhattan interval. Furthermore, our method employs an LSTM network (Long Short-Term Memory) to find similar users based on their feedback, reducing data scarcity's impact and coping with the Cold Start issue. Moreover, it introduces multifaceted contextual information and represents a novel approach to determining user preferences. Finally, this neural hybrid framework identifies a list of the most efficient trip candidates by mining personalized POIs in a sequential pattern and incorporating them into the recommendation process. The proposed methodology was tested using datasets from Yelp, Gowalla, and Tripadvisor, and the results represented that it performed better than other methods in multiple metrics, including MAP, NDCG, RMSE, and F-Score. [ABSTRACT FROM AUTHOR]

Software Fault Prediction has a critical role in improving effectiveness and reliability of software systems by identifying potential faults early in the development cycle. A hybrid PSO optimized CNN-RNN model leverages the strengths of both RNN and CNN in capturing temporal and spatial data features, while PSO optimizes hyperparameters to enhance model performance in this research. The proposed PSO optimized CNN-RNN model is compared against existing hybrid machine learning models, where PSO optimized Genetic Algorithm (GA) was used for hyperparameter tuning and feature selection of Support Vector Machine (SVM). Our experiments are performed on publicly available software fault's datasets, providing a comprehensive comparison of model performance that is evaluated on the basis of various matrices of performance like F-measures, accuracy, recall, F1-score, SD and precision. The results demonstrate that while optimized Machine Learning algorithms perform well in some cases, the CNN-RNN-PSO model consistently outperforms them, offering superior fault prediction capabilities. The NASA MDP repository's benchmark datasets are used for the comparative analysis and the results demonstrated that the optimized hybrid machine learning model achieves competitive performance. The proposed PSO optimized CNNRNN model demonstrates superior accuracy and robustness due to its deep learning architecture and optimization capabilities. This research focus on the potential of a hybrid DL approach which improves the software reliability and suggests future directions for integrating intelligent models in SFP. [ABSTRACT FROM AUTHOR]

In the software engineering domain, the distinction between functional (FRs) and non-functional requirements (NFRs) is paramount, as it directly influences the design and development of software systems. However, several challenges, such as dealing with limited training data, domain-specific datasets, and high computational costs, have driven the need for innovative solutions, particularly those related to classifying functional and non-functional software requirements. The limited availability of labeled data for training deep learning models and their high computational costs have hindered progress. This study proposes a novel hierarchical transfer learning (HTL) approach to address the challenges of limited training data and high computational costs associated with deep learning models. The HTL model leverages transfer learning techniques, incorporating pre-trained models such as global vectors for word representation (GloVe) for text vectorization and a bidirectional long short-term memory (BiLSTM) architecture. By harnessing knowledge from large text corpora and capturing both high-level semantic relationships and detailed syntactic patterns, the HTL model demonstrates enhanced classification performance. We have evaluated the model's performance using precision, recall, F1-score, and the area under the receiver operating characteristic curve. For FRs classification, we have observed a 26% improvement in precision, a 9% improvement in recall, and an 18% in F1-score for small datasets. Similarly, for NFRs, classification achieves a 20% improvement in precision, a 38.8% improvement in recall, and a 31.8% improvement in F1-score. For large datasets, we have observed a 25% improvement in precision, a 7% improvement in recall, and a 15% improvement in F1-score for FRs classification. For NFRs classification, it achieves a 24% improvement in precision, a 39.8% improvement in recall, and a 41.8% improvement in F1-score. Our study presents a pioneering HTL approach for FRs and NFRs classification, demonstrating superior performance compared to traditional methods. Furthermore, we identify areas for future research, including improving model interpretability, handling data biases, and fine-tuning hyperparameters, which will further enhance the capabilities and applicability of the HTL model. [ABSTRACT FROM AUTHOR]

Copyright of Journal of Frontiers of Computer Science & Technology is the property of Beijing Journal of Computer Engineering & Applications Journal Co Ltd. and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

This paper develops a novel attention-based neural point process approach to model behavioral dynamics and predict future activities of consumers in digital content consumption. The consumption of digital content products (e.g., video games and live streaming) is often associated with multifaceted, dynamically interacting consumer behavior that is subject to influence from pertinent external events. Inspired by these characteristics, we develop a novel attention-based neural point process approach to holistically capture the richness and complexity of consumer behavioral dynamics in modern digital content consumption. Our model features a new multirepresentational, continuous-time attention mechanism that can flexibly model dynamic interactions between different types of behavior under external influence. Using learned representations as sufficient statistics of past events, we build a marked point process to efficiently characterize the occurrence time, behavior combination, and consumption quantity of consumers' future activities. We illustrate our model development and applications in the empirical context of a sports video game, showing its superior predictive performance over a wide range of baseline methods. Leveraging individual-level parameter estimates, we further demonstrate our model's utility for conducting segmentation analysis and evaluating the effects of past events on consumers' future engagement. Our model provides managers and practitioners with a powerful tool for developing more effective and targeted marketing strategies and gaining insights into consumer behavioral dynamics in digital content consumption. History: Yuxin Chen served as the senior editor. Funding: J. Yin was partly supported by the Adobe Digital Experience Research Award and the Amazon AWS Machine Learning Research Award. Y. (K.) Feng was supported by the Research Grants Council of Hong Kong [ECS Grant 25508819]. Y. Liu gratefully acknowledges the support from Bob Eckert, chairman of the board at Levi Struss and former CEO and chairman at Mattel, through the Robert A. Eckert endowed chair in marketing at the University of Arizona. Supplemental Material: The online appendices and data files are available at https://doi.org/10.1287/mksc.2020.0180. [ABSTRACT FROM AUTHOR]

In this study, novel approach of employing a recurrent neural network (RNN) model to predict the shear strength of aluminium-low carbon steel explosive clads is attempted. The explosive cladding process parameters, namely the explosive ratio (ranging from 0.6 to 1.4), standoff distance (ranging from 5.1 to 9.1 mm), were varied and detailed elsewhere. Due to the nonlinear relationship between these process parameters, predicting the shear strength through analytical techniques becomes challenging, making computational machine learning approaches more relevant. The RNN model was trained using experimental shear data and preliminary experiments in a Python environment. The model performance was then evaluated against the remaining test data. The RNN demonstrated high prediction accuracy, achieving a coefficient of determination (R2) of 0.9762, a mean squared error (MSE) of 0.7571, a root mean squared error (RMSE) of 0.87 and a mean absolute error (MAE) of 0.72. [ABSTRACT FROM AUTHOR]

Featured Application: A better discrimination in a prediction model does not imply a better interpretability. In healthcare, where transparency is crucial, both discriminability and interpretability should be checked before stating that a new model is better. Background: In predictive modelling, particularly in fields such as healthcare, the importance of understanding the model's behaviour rivals, if not surpasses, that of discriminability. To this end, attention mechanisms have been included in deep learning models for years. However, when comparing different models, the one with the best discriminability is usually chosen without considering the clinical plausibility of their predictions. Objective: In this work several attention-based deep learning architectures with increasing degrees of complexity were designed and compared aiming to study the balance between discriminability and plausibility with architecture complexity when working with longitudinal data from Electronic Health Records (EHRs). Methods: We developed four deep learning-based architectures with attention mechanisms that were progressively more complex to handle longitudinal data from EHRs. We evaluated their discriminability and resulting attention maps and compared them amongst architectures and different input processing approaches. We trained them on 10 years of data from EHRs from Catalonia (Spain) and evaluated them using a 5-fold cross-validation to predict 1-year all-cause mortality in a subsample of 500,000 people over 65 years of age. Results: Generally, the simplest architectures led to the best overall discriminability, slightly decreasing with complexity by up to 8.7%. However, the attention maps resulting from the simpler architectures were less informative and less clinically plausible compared to those from more complex architectures. Moreover, the latter could give attention weights both in the time and feature domains. Conclusions: Our results suggest that discriminability and more informative and clinically plausible attention maps do not always go together. Given the preferences within the healthcare field for enhanced explainability, establishing a balance with discriminability is imperative. [ABSTRACT FROM AUTHOR]

Motion control of modern and advanced aircraft has to be provided under conditions of incomplete and inaccurate knowledge of their parameters and characteristics, possible flight modes, and environmental influences. In addition, various abnormal situations may occur during flight, in particular, equipment failures and structural damage. These circumstances cause the problem of a rapid adjustment of the used control laws so that the control system can adapt to the mentioned changes. However, most adaptive control schemes have a model of the control object, which plays a crucial role in adjusting the control law. That is, it is required to solve also the identification problem for dynamical systems. We propose an approach to solving the above-mentioned problems based on artificial neural networks (ANNs) and hybrid technologies. In the class of traditional neural network technologies, we use recurrent neural networks of the NARX type, which allow us to obtain black-box models for controlled dynamical systems. It is shown that in a number of cases, in particular, for control objects with complicated dynamic properties, this approach turns out to be inefficient. One of the possible alternatives to this approach, investigated in the paper, consists of the transition to hybrid neural network models of the gray box type. These are semi-empirical models that combine in the resulting network structure both empirical data on the behavior of an object and theoretical knowledge about its nature. They allow solving with high accuracy the problems inaccessible by the level of complexity for ANN models of the black-box type. However, the process of forming such models requires a very large consumption of computational resources. For this reason, the paper considers another variant of the hybrid ANN model. In it, the hybrid model consists not of the combination of empirical and theoretical elements, resulting in a recurrent network of a special kind, but of the combination of elements of feedforward networks and recurrent networks. Such a variant opens up the possibility of involving deep learning technology in the construction of motion models for controlled systems. As a result of this study, data were obtained that allow us to evaluate the effectiveness of two variants of hybrid neural networks, which can be used to solve problems of modeling, identification, and control of aircraft. The capabilities and limitations of these variants are demonstrated on several examples. Namely, on the example of the problem of aircraft longitudinal angular motion, the possibilities of modeling the motion using the NARX network as applied to a supersonic transport aircraft (SST) are first considered. It is shown that under complicated operating conditions this network does not always provide acceptable modeling accuracy. Further, the same problem, but applied to a maneuverable aircraft, as a more complex object of modeling and identification, is solved using both a NARX network (black box) and a semi-empirical model (gray box). The significant advantage of the gray box model over the black box one is shown. The capabilities of the hybrid model realizing deep learning technologies are demonstrated by forming a model of the control object (SST) and neurocontroller on the example of the MRAC adaptive control scheme. The efficiency of the obtained solution is illustrated by comparing the response of the control object with a failure situation (a decrease in the efficiency of longitudinal control by 50%) with and without adaptation. [ABSTRACT FROM AUTHOR]

Through the rapid advancement of network technology, Network Intrusion Detection Systems (IDS) have become a vital constituent of network security. However, many existing traditional models struggle to identify key features due to the complexity of the feature dimensions, which diminishes classification accuracy. Hence, this research proposes the Di-strategy-based Beluga Whale Optimization (DS-BWO) approach for performing the feature selection in intrusion detections. The DS-BWO approach excels in selecting the most relevant features from high-dimensional datasets, effectively reducing dimensionality and improving model performance while maintaining important information. A Hybrid Recurrent Neural Network with Bidirectional Long Short-Term Memory (RNN-BiLSTM) approach is used for the classification of intrusions into binary classes like normal and malignant. By processing data in both forward and backward directions, BiLSTM captures dependencies in sequences more effectively. The proposed DS-BWO approach is implemented using three standard datasets: CIC-IDS2018, NSL-KDD, and UNSW-NB15. The implementation results demonstrate that the proposed DS-BWO approach reaches a superior accuracy of 0.999 on the CIC-IDS2018 dataset in comparison to the existing algorithms like Multi-Agent Feature Selection IDS namely (MAFSIDS) and Extreme Gradient Boosting. [ABSTRACT FROM AUTHOR]