Search

We show expression rates and stability in Sobolev norms of deep feedforward ReLU neural networks (NNs) in terms of the number of parameters defining the NN for continuous, piecewise polynomial functions, on arbitrary, finite partitions T of a bounded interval (a , b). Novel constructions of ReLU NN surrogates encoding function approximations in terms of Chebyšev polynomial expansion coefficients are developed which require fewer neurons than previous constructions. Chebyšev coefficients can be computed easily from the values of the function in the Clenshaw–Curtis points using the inverse fast Fourier transform. Bounds on expression rates and stability are obtained that are superior to those of constructions based on ReLU NN emulations of monomials as considered in [24,22]. All emulation bounds are explicit in terms of the (arbitrary) partition of the interval, the target emulation accuracy and the polynomial degree in each element of the partition. ReLU NN emulation error estimates are provided for various classes of functions and norms, commonly encountered in numerical analysis. In particular, we show exponential ReLU emulation rate bounds for analytic functions with point singularities and develop an interface between Chebfun approximations and constructive ReLU NN emulations. [ABSTRACT FROM AUTHOR]

Recently, one artificial intelligence technique, known as artificial neural network (ANN), has brought advanced development to the arena of mathematical research. It competes effectively with other traditional methods in providing accurate solutions for fractional differential equations (FDEs). This work aims to implement a feedforward ANN with two hidden layers to solve nonlinear systems based on the fractional Riccati differential equation (FRDE). The network parameters are trained using the Adam optimization method with the aid of automatic differentiation. A vectorization algorithm is designated for the selected step to make the computation process more efficient. Two different initial value problems in integer-order derivatives and fractional-order derivatives are discussed. Numerical results demonstrate that the proposed method not only closely matches the exact solutions and reference solutions but also is more accurate than other existing methods. [ABSTRACT FROM AUTHOR]

A novel framework is proposed in this research based on multilevel method to solve the optimal control problem. In recent dacades, the mathematical theory of optimal control has rapidly developed into an important and separate field of applied mathematics. The solution of nonlinear partial differential equations is considerably difficult, and the theory of their optimal control is still an open field in many respects. These optimization problems have found diverse applications in various sciences including electrical engineering, mechanical engineering, and aerospace. Current methods for solving this class of optimal control problems usually fall into two classes: discrete-then-optimization or optimization-then-discrete approaches. The proposed approach, however, does not require discretization as it involves rewriting the optimal control problem as a multi-objective optimization problem followed by its solution with a feedforward single-layer artificial neural network based on learning through by the multi-level Levenberg–Marquardt method. Moreover, the convergence of the approach was discussed and some numerical results are presented. [ABSTRACT FROM AUTHOR]

This article examines whether the brain is organized based on specialized mechanisms for processing specific types of stimuli or if it is organized around mechanisms that can be applied to different types of stimuli. The authors conducted a study using deep neural networks to investigate the face inversion effect, which is a difficulty in recognizing faces when they are upside-down. They found that similar inversion effects exist for other types of stimuli, but these effects do not extend to unseen stimuli. The results suggest that the face inversion effect may be due to specialized processing mechanisms rather than general processing mechanisms. The article also discusses the potential of deep learning as a computational model for understanding human vision and the need for further research on the development of processing systems in the brain. [Extracted from the article]

The structured sand presents significant interparticle bonding and anisotropy, resulting in significant differences in the physical and mechanical properties from the pure sand. This study proposes a new surrogate model based on the concept of multi‐fidelity residual neural network (MR‐NN) as an alternative to DEM simulation for predicting the mechanical behaviours of structured sand with different initial anisotropy and saving largely computational costs. The model is initially trained using low‐fidelity (LF) data to focus on capturing the main underpinning correlations between macroscopic mechanical parameters with inter‐particle properties and anisotropic state variables (i.e., microscopic fabric and tilting angle of sand), where the LF data are generated from previously proposed anisotropic macro‐micro quantitative correlation. Subsequent training on sparser high‐fidelity (HF) data is used to calibrate and refine the model with HF data generated from DEM simulations for anisotropic structured sand. Feedforward neural network (FNN) is adopted as the baseline algorithm for training models. The macroscopic anisotropic parameters of structured sand predicted by the surrogate model are compared with DEM simulations to examine its feasibility and generalization ability. Furthermore, the robustness of the surrogate model is examined by discussing the effect of LF data on the performance of MR‐NN. The superiority of the MR‐NN is further verified by comparing the performance of the trained MR‐NN with the one‐shot trained FNN based on the same HF data. All results demonstrate that the proposed surrogate model can provide a fast and accurate simulation of the anisotropic parameters of structured sand. [ABSTRACT FROM AUTHOR]

This paper reports the prediction results of the ionospheric irregularities represented by the ROTI index close to the equatorial ionization anomaly's Northern peak. A feedforward backpropagation Neural Network (NN) approach is implemented using a time series Nonlinear Autoregressive approach with Exogenous inputs (NARX). We used the data from a dual frequency GPS-SCINDA station located at Helwan (geographic coordinates of 29.86°N, 31.32°E, and MLAT of 29.94°N) in Egypt. To allow the model to be independently tested during varying levels of solar activities, we collected a 5-minute resolution of ROTI data for the solar cycle 24 from 2009 to 2017. The factors that influence the development of ionospheric irregularities are involved in the established model representing the diurnal and seasonal variations as well as solar and geomagnetic activity parameters. To support learning, we included IRI-foF2 and IRI-hmF2 parameters in the input layer neurons to improve the model learning about the behavior of the ionospheric F layer. The results show that the NN-ROTI values precisely match the GPS-ROTI values with an RMSE of 0.106 TECU/min and a prediction efficiency of 95%. The predicted values are highly correlated with the originals, with a regression of 0.89. Furthermore, the irregularities were more prevalent during the equinox months than during the solstice months. It is also observed that predicted NN-ROTI values in all seasons have RMSEs less than 0.03 and 0.05 TECU/min for the low and high solar activity, respectively. The results also showed a clear correlation between solar activity and the occurrence percentage of ionospheric irregularities indicating a solar cycle dependency. The prediction values of the NN-ROTI occurrence % demonstrated a remarkable correlation with the observed GPS-ROTI occurrence % across SC24. [ABSTRACT FROM AUTHOR]

This research introduces a new technique to control constrained nonlinear systems, named Lyapunov-based neural network model predictive control using a metaheuristic optimization approach. This controller utilizes a feedforward neural network model as a prediction model and employs the driving training based optimization algorithm to resolve the related constrained optimization problem. The proposed controller relies on the simplicity and accuracy of the feedforward neural network model and the convergence speed of the driving training based optimization algorithm. The closed-loop stability of the developed controller is ensured by including the Lyapunov function as a constraint in the cost function. The efficiency of the suggested controller is illustrated by controlling the angular speed of three-phase squirrel cage induction motor. The reached results are contrasted to those of other methods, specifically the fuzzy logic controller optimized by teaching learning-based optimization algorithm, the optimized PID with particle swarm optimization algorithm, the neural network model predictive controller based on particle swarm optimization algorithm, and the neural network model predictive controller using driving training based optimization algorithm. This comparative study showcase that the suggested controller provides good accuracy, quickness and robustness due to the obtained values of the mean absolute error, mean square error root mean square error, enhancement percentage, and computing time in the different simulation cases, and it can be efficiently utilized to control constrained nonlinear systems with fast dynamics. [ABSTRACT FROM AUTHOR]

Oxygen excess ratio (OER), which is influenced by the air flow input to the cathode, is one of the important factors affecting the performance of fuel cell system. Insufficient oxygen supply will result in a sharp drop in output voltage, while an excessive supply will bring about a large parasitic power load on the air compressor, reducing the net output power of the system and operating efficiency. Both insufficient and excess oxygen supply can adversely affect the operating life of the proton exchange membrane fuel cell (PEMFC). Due to the serious nonlinear, hysteresis, uncertainty and interference in the process of OER control, this paper proposes an OER feedforward control based on extended state observer (ESO) for a PEMFC auxiliary system with the ESO to predict the total disturbance to improve the dynamic response of the system with large time delay utilizing terminal sliding mode control (TSMC). The control strategy is divided into two stages. The first stage constructs an ESO to realize real-time observation of the state variables required for the internal operation of the auxiliary machine system and the PEMFC, and yields the air compressor voltage control signal through terminal sliding mode control. In the second stage, combined with V/f control for the compressor, the transmitted air flow is adjusted, ultimately achieving the goal of changing the OER of the PEMFC and improving operating efficiency. The simulation results show that the proposed control strategy has good anti-interference capabilities, a faster response, and less overshoot. The gas flow transmission in the gas supply system is more robust, the PEMFC operation is safer, and its lifespan is extended. • The control difficulty caused by time delay characteristics of gas transmission in hydrogen production system considering gas-electric coupling is overcome. • The additional ESO can well observe the state variables and external disturbance that cannot be directly measured in the PEMFC system. The closed-loop control based on TSMC of the system is realized more accurately. • This paper considers the situation that when the energy storage system cannot meet the energy demand, PEMFC still wants to provide continuous power output. An adjusting strategy for OER in PEMFC through the electromagnetic effect of air compressor to extend the life span of the PEM fuel cell is proposed. • The proposed OER feedforward control strategy demonstrated superior robustness and fast response characteristics for oxygen excess ratio optimization. [ABSTRACT FROM AUTHOR]

With the current advancements in artificial intelligence and machine learning, data has become a powerful tool for major improvements in various fields. In the field of aerodynamic design, most algorithms utilize an iterative method to reach their target function or geometry due to their robustness. Deep learning models enable us to exploit the data generated during those iterations to leverage the design algorithm. In this paper, design procedures and guidelines were presented for the use of multilayer feedforward neural network (MFNN) and long-short term memory (LSTM) network to predict the target geometry with early generated data of the design algorithm to reduce its computational cost. The impact of various parameters and hyperparameters on the quality of the target prediction was discussed and early results were presented for various representations of input data using the NACA-0011 airfoil aerodynamic design data. The results indicated that selecting the appropriate network and hyperparameters can yield a reliable estimate of the target geometry using only 20 % to 30 % of the available data. [ABSTRACT FROM AUTHOR]

Intermittent demand forecasting is an important yet challenging task in many organizations. While prior research has been focused on traditional methods such as Croston's method and its variants, limited research has been conducted using advanced machine learning or deep learning methods. In this study, we introduce Transformer, a recently developed deep learning approach, to forecast intermittent demand. Its effectiveness is empirically tested with a dataset of 925 intermittent demand items from an airline spare parts provider and compared with that of two traditional methods such as Croston's and the Syntetos–Boylan approximation as well as several popular neural network architectures including feedforward neural networks, recurrent neural networks, and long short-term memory. Our results based on six different forecasting performance measures show that Transformer performs very well against other methods in a variety of settings. We also examine how data sparsity impacts model performance and find that different models perform similarly when sparsity is low. Although the performance of all models generally gets worse as the sparsity level increases, the advantage of Transformer over other models increases with sparsity. [ABSTRACT FROM AUTHOR]

Cell trajectory inference is very important to understand the details of tissue cell development, state differentiation and gene dynamic regulation. However, due to the high noise and heterogeneity of the single-cell data, it is challenging to infer cell trajectory in complex biological processes. Here, we proposed a new trajectory inference method, called Metric Learning Bhattacharyya Kernel Feature Decomposition (MLBKFD). In MLBKFD, a statistical model was used to infer cell trajectory by calculating the Markov transition matrix and Bhattacharyya kernel matrix between cells. Before that, to expedite the matrix calculation in the statistical model, a deep feedforward neural network was used to perform dimensionality reduction on single-cell data. The MLBKFD was evaluated on four typical datasets as well as seven recent human fetal lung datasets. Comparisons with the two outstanding methods (i.e., DTFLOW and MARGARET) demonstrate that the MLBKFD is capable of accurately inferring cell development and differentiation trajectories from single-cell data with different sizes and sources. Notably, MLBKFD exhibits nearly twice the speed of DTFLOW while maintaining high precision, particularly when dealing with large datasets. MLBKFD provides accurate and efficient trajectory inference, empowering researchers to gain deeper insights into the complex dynamics of cell development and differentiation. 1. MLBKFD is a powerful probabilistic model to effectively infer cell pseudo trajectories from single-cell data by calculating the Markov transition matrix and Bhattacharyya kernel matrix between cells. 2. MLBKFD exhibits fast speed and excellent visualization effects on cell trajectory inference, by utilizing a deep feedforward neural networks for dimensionality reduction of single-cell data. 3. By iteratively refining cell labels in the deep neural networks, MLBKFD can achieve excellent cell clustering results, enabling it to infer cell trajectories without relying on true cell labels. [ABSTRACT FROM AUTHOR]

Selecting or generating ground motions (GMs) that elicit seismic responses matching specific standards or expected benchmarks for nonlinear time‐history analysis (NLTHA) is crucial for ensuring the rationality of structural seismic design and analysis. Typical GM inputs for NLTHA, either natural or artificial, are normally spectrum‐compatible, which may produce significant variations in analysis results, even using multiple GMs. This paper introduces a response‐compatible ground motion generation (RCGMG) method for generating GMs that are tailored to be response‐compatible. NLTHA results using only a few of these artificial GMs can closely approximate the mean responses from a large set of natural spectrum‐compatible GMs or target responses. The RCGMG method adopts the response diagram in the time domain (RDTD) to characterize the nonstationary features of GMs and their impacts on structural dynamic responses. A physics‐guided conditional generative adversarial network is developed to produce artificial RDTDs with features and impacts of RDTDs of natural GMs. These generated RDTDs are then mapped into response‐compatible GMs through a feedforward neural network. To verify the effectiveness of RCGMG, NLTHA of different structure models under various site conditions and target spectra is conducted. Seismic responses of NLTHA using RCGMG‐generated GMs are compared with responses from spectrum‐compatible natural GMs. The results demonstrate that responses from RCGMG GMs are closer to the target responses, with fewer GM inputs and robust generalization performance. [ABSTRACT FROM AUTHOR]

Autonomous pollination robots have been widely discussed in recent years. However, the accurate estimation of flower poses in complex agricultural environments remains a challenge. To this end, this work proposes the implementation of a transformer-based architecture to learn the translational and rotational errors between the pollination robot's end effector and the target object with the aim of enhancing robotic pollination efficiency in cross-breeding tasks. The contributions are as follows: (1) We have developed a transformer architecture model, equipped with two feedforward neural networks that directly regress the translational and rotational errors between the robot's end effector and the pollination target. (2) Additionally, we have designed a regression loss function that is guided by the translational and rotational errors between the robot's end effector and the pollination targets. This enables the robot arm to rapidly and accurately identify the pollination target from the current position. (3) Furthermore, we have designed a strategy to readily acquire a substantial number of training samples from eye-in-hand observation, which can be utilized as inputs for the model. Meanwhile, the translational and rotational errors identified in the end-manipulator Cartesian coordinate system are designated as loss targets simultaneously. This helps to optimize the training of the model. We conducted experiments on a realistic robotic pollination system. The results demonstrate that the proposed method outperforms the state-of-the-art method, in terms of both accuracy and efficiency. [ABSTRACT FROM AUTHOR]

This study aims to find the most powerful algorithm between LM and GD, applying them to the multilayer neural network (MLP) to predict the wind speed of the city of Tetouan. To achieve this we will use the meteorological data of this city from 31/07/2017 to 31/08/2022. The MLP adopted for our study is composed of two hidden layers, 30 neurons in the first layer and 15 in the second, 7 inputs and one output. The data is divided into 80% for training and 20% for testing. The results obtained showed that the Levenberg-Marquardt (LM) algorithm is more efficient than the gradient descent (GD) algorithm with a correlation coefficient R = 0.988102 and a mean square error MSE = 0. 0458. These results will allow us to accurately predict the wind speed of August for the year 2022 in this city. [ABSTRACT FROM AUTHOR]

In contemporary power transmission systems, substation monitoring stands as a vital but challenging task. While robotics offers promise in this regard, its potential is still nascent, struggling to replicate human intelligence. This article's core aim was to optimize robot path planning (RPP). Employing the enhanced red deer algorithm (ERDA), we sought to bolster RPP for more efficient substation inspections. The key methods used seem to be modeling, experimentation, comparative analysis, and some elements of data benchmarking to systematically evaluate and validate their proposed technique and models both in simulation and the real world. Research aims to enhance substation inspection effectiveness and bolster the safety of power usage in society. Proposed hybrid approach, combining proportional–integral–derivative (PID) with ERDA (PID–ERDA), underpins an Intelligent Intelligent RPP framework tailored to substation inspections. Examining the PID–ERDA model's performance, it significantly improved path length by 18%–29% and reduced response times by 14%–26% compared with PID or ERDA alone. PID–ERDA consistently achieved optimal solutions in 40–60 trials out of 85, while PID and ERDA managed 20–40 trials with inconsistent optimization. Additionally, it reduced average response times to 17–20 s from 21 to 27 s observed when using PID and ERDA separately. PID–ERDA also demonstrated superior path accuracy, surpassing methods like improved adaptive control algorithm‐feedforward neural network, enhanced unified algorithm‐susceptible‐infected‐removed, and bounded behavior‐particle swarm optimization by 7%–13%. The study affirms that the PID–ERDA model significantly enhances path planning for substation inspections, representing a milestone in RPP for power station inspections within modern power transmission systems. The primary contribution of this research is the significant improvement it brings to RPP for power station inspections, especially in substation monitoring within modern power transmission systems. [ABSTRACT FROM AUTHOR]

All fields of science depend on mathematical models. Occam's razor refers to the principle that good models should exclude parameters beyond those minimally required to describe the systems they represent. This is because redundancy can lead to incorrect estimates of model parameters from data, and thus inaccurate or ambiguous conclusions. Here, we show how deep learning can be powerfully leveraged to apply Occam's razor to model parameters. Our method, FixFit, uses a feedforward deep neural network with a bottleneck layer to characterize and predict the behavior of a given model from its input parameters. FixFit has three major benefits. First, it provides a metric to quantify the original model's degree of complexity. Second, it allows for the unique fitting of data. Third, it provides an unbiased way to discriminate between experimental hypotheses that add value versus those that do not. In three use cases, we demonstrate the broad applicability of this method across scientific domains. To validate the method using a known system, we apply FixFit to recover known composite parameters for the Kepler orbit model and a dynamic model of blood glucose regulation. In the latter, we demonstrate the ability to fit the latent parameters to real data. To illustrate how the method can be applied to less well-established fields, we use it to identify parameters for a multi-scale brain model and reduce the search space for viable candidate mechanisms. Author summary: Mathematical modeling is a pillar of scientific inquiry, bridging the gap between theory and experimental observations. However, in complex systems such as those pervasive in biology (e.g., gene regulatory networks, multi-scale brain interactions, and drug pharmacokinetics), different mechanisms can yield equally plausible explanations for the data. This ambiguity is not due to data limitations but rather to the equations that govern these systems. Both the interpretation and estimation of the parameters of these models are hindered by these intrinsic degeneracies. For this reason, we present a general tool that harnesses the power of deep learning to automatically identify and rectify these ambiguities, allowing a broader range of models to be precisely determined by experimental data. We show, for example, that our method provides novel insights into the features of multi-scale brain dynamics that can be learned from functional neuroimaging. This represents a crucial initial step toward characterizing the mechanistic insights that different types of experiments can provide. [ABSTRACT FROM AUTHOR]

The cost for covering active power losses makes a significant item in transmission system operators (TSO) annual budgets, and still it received limited attention in the existing literature. The focus of accurate power loss forecasting and procurement is of high increase during the past 2 years due to spikes in electricity prices, making the cost of covering the active power losses a dominant factor of TSO operational costs. This paper presents practical aspects of the highly accurate models for transmission loss forecast in the day ahead time frame for the Croatian transmission system. The contributions are two‐fold: 1) Practical insights into usable TSO data are provided, filling a critical research gap and a foundational literature review is established on transmission loss forecasting. 2) A novel method utilizing only electricity transit data as input which outperforms existing practices is presented. For this, several algorithms such as gradient boosted decision tree model (XGB), support vector regressors, multiple linear regression and fully connected feedforward artificial neural networks are developed, and implemented and validated on data obtained from the Croatian TSO. The results show that the XGB model outperforms current TSO model by 32% for 4 months of comparison and TSCNET's commercial solution by 25% during a year‐long testing period. The developed XGB model is also implemented as a software tool and put into everyday operation with the Croatian TSO. [ABSTRACT FROM AUTHOR]

Purpose Methods Results Conclusions The purpose of this study is to compare the reconstructed corneal power (RCP) by working backwards from the post‐implantation spectacle refraction and toric intraocular lens power and to develop the models for mapping preoperative keratometry and total corneal power to RCP.Retrospective single‐centre study involving 442 eyes treated with a monofocal and trifocal toric IOL (Zeiss TORBI and LISA). Keratometry and total corneal power were measured preoperatively and postoperatively using IOLMaster 700. Feedforward neural network and multilinear regression models were derived to map keratometry and total corneal power vector components (equivalent power EQ and astigmatism components C0 and C45) to the respective RCP components.Mean preoperative/postoperative C0 for keratometry and total corneal power was −0.14/−0.08 dioptres and −0.30/−0.24 dioptres. All mean C45 components ranged between −0.11 and −0.20 dioptres. With crossvalidation, the neural network and regression models showed comparable results on the test data with a mean squared prediction error of 0.20/0.18 and 0.22/0.22 dioptres2 and on the training data the neural network models outperformed the regression models with 0.11/0.12 and 0.22/0.22 dioptres2 for predicting RCP from preoperative keratometry/total corneal power.Based on our dataset, both the feedforward neural network and multilinear regression models showed good precision in predicting the power vector components of RCP from preoperative keratometry or total corneal power. With a similar performance in crossvalidation and a simple implementation in consumer software, we recommend implementation of regression models in clinical practice. [ABSTRACT FROM AUTHOR]

Aerodynamic forces hold significant promise for controlling small-satellite formations in low Earth orbit. However, significant challenges still exist regarding complex input constraints and their application to multi-satellite formations. This paper proposes a constraint tightening-based control scheme for small-satellite formations, utilizing differential drag and lift as actuating forces. Two attitude adjustment strategies, namely the reorientation strategy and the single-radial rotation strategy, are compared in terms of their control effectiveness and applicability. Relative aerodynamic coefficients are employed as control inputs in view of their solid reachable region, followed by the development of a decoupling algorithm for determining pointing angles. Through a method that tightens the reachable region using liner models, the input constraints are linearized and can be explicitly expressed. Subsequently, an observer-based Model Predictive Control (MPC) algorithm is formulated to meet these linearized input constraints and provide feedforward compensation for system disturbances. This discrete algorithm enables practical onboard implementations owing to its low computational complexity. Additionally, a concept of overall computation for all satellites' attitudes is proposed to adapt the control scheme for multi-satellite formations. Hardware-in-the-loop simulations are conducted to evaluate the control scheme's performance in along-track and circular formations, accounting for uncertainties in aerodynamic forces and attitude dynamics. [ABSTRACT FROM AUTHOR]

In this study, the effect of high‐power bath and horn ultrasound at different powers on specific surface area (SBET), total pore volume (Vtotal), and average pore volume (Dave) of bleaching clay was examined. After subjecting the bleaching clay to ultrasonication treatment, the SBET values demonstrated an escalation from 31.4 ± 2.7 m2 g−1 to 59.8 ± 3.1 m2 g−1 for HU200BC, 143.8 ± 3.9 m2 g−1 for HU400BC, 54.4 ± 3.6 m2 g−1 for BU400BC, and 137.5 ± 2.8 m2 g−1 for BU800BC. The mean pore diameter (Dave) declined from 29.7 ± 0.14 nm in bleaching clay to 11.3 ± 0.13 nm in HU200BC, 8.3 ± 0.12 nm in HU400BC, 16.7 ± 0.14 nm in BU400BC, and 9.6 ± 0.12 nm in BU800BC. Therefore, horn ultrasound‐treated bleaching clay significantly increased SBET and Vtotal, indicating improved adsorption capacity. Moreover, to establish the relationship between bleaching parameters, seven multi‐output ML regression models of Feedforward Neural Network (FNN), Random Forest (RF), Support Vector Regression (SVR), Multi‐Task Lasso, Ridge regression, Extreme Gradient Boosting (XGBoost), and Gradient Boosting are used, and compared with response surface methodology (RSM). ML has revolutionized the understanding of complex relationships between ultrasonic parameters, oil color, and pigment degradation, providing insights into how various factors such as temperature, ultrasonic power, and time can influence the bleaching process, ultimately enhancing the efficiency and precision of the treatment. The XGBoost model showed outstanding performance in predicting the target variables with a high R2‐train up to 1, R2‐test up to .983, and a minimum mean absolute error (MAE) of 0.498. The lower error between the predicted and experimental values implies the superiority of the XGBoost model to predict outcomes rather than RSM. It represents the suitability of bath ultrasound as a mild condition for low‐pigmented oil bleaching. Finally, the Bayesian optimization method in conjunction with XGBoost was used to optimize the amount of bleaching clay and energy consumption, and its performance was compared with RSM. It was observed that the consumption of bleaching clay was reduced by approximately 60% for sunflower oil and 30%–35% for soybean oil. [ABSTRACT FROM AUTHOR]

Disinformation in the medical field is a growing problem that carries a significant risk. Therefore, it is crucial to detect and combat it effectively. In this article, we provide three elements to aid in this fight: 1) a new framework that collects health-related articles from verification entities and facilitates their check-worthiness and fact-checking annotation at the sentence level; 2) a corpus generated using this framework, composed of 10335 sentences annotated in these two concepts and grouped into 327 articles, which we call KEANE (faKe nEws At seNtence lEvel); and 3) a new model for verifying fake news that combines specific identifiers of the medical domain with triplets subject-predicate-object, using Transformers and feedforward neural networks at the sentence level. This model predicts the fact-checking of sentences and evaluates the veracity of the entire article. After training this model on our corpus, we achieved remarkable results in the binary classification of sentences (check-worthiness F1: 0.749, fact-checking F1: 0.698) and in the final classification of complete articles (F1: 0.703). We also tested its performance against another public dataset and found that it performed better than most systems evaluated on that dataset. Moreover, the corpus we provide differs from other existing corpora in its duality of sentence-article annotation, which can provide an additional level of justification of the prediction of truth or untruth made by the model. [ABSTRACT FROM AUTHOR]

One‐hidden‐layer feedforward neural networks are described as functions having many real‐valued parameters. Approximation properties of neural networks are established (universal approximation property), and the approximation error is related to the number of parameters in the network. The essentially optimal order of approximation error bounds was already derived in 1996. We focused on the numerical experiment that indicates the neural networks whose parameters contain stochastic perturbations gain better performance than ordinary neural networks and explored the approximation property of neural networks with stochastic perturbations. In this paper, we derived the quantitative order of variance of stochastic perturbations to achieve the essentially optimal approximation order and verified the justifiability of our theory by numerical experiments. [ABSTRACT FROM AUTHOR]

Nowadays, in spite of advanced technology, there are still some sound problems on modern cars because of mechanical parts, oil lubrications, and electric motors. Due to these unwanted problems, it is necessary to design intelligent predictors such as artificial neural networks. In this investigation, a procedure of testing and evaluation on the sound quality of two types of cars are proposed and sound quality is analyzed through the cars road running test on the providing ground, which is carried out with varying running speed. To improve and predict the results of experimental approach analysis, a proposed neural network predictor is also designed to model of the system for possible experimental applications. The proposed neural network is a feedforward type network, which consists of multi hidden layers. Three different training algorithms are used for training the network. As basic factors for sound quality, only objective factors a considered such as loudness, sharpness, speech intelligibility, sound pressure level. The correlation between sound pressure level and other factors are discussed from a point of view of running speed dependency. Results of both computer simulations and experiments show that the neural predictor algorithm gives good results at accommodating different cases and provides superior prediction on two cars's sound analysis. [ABSTRACT FROM AUTHOR]

A hybrid optimization filter for weather and wave numerical models is proposed and tested in this study. Parametrized Artificial Neural Networks are utilized in conjunction with Extended Kalman Filters to provide a novel postprocess strategy for 10 m wind speed, 2 m air temperature, and significant wave height simulations. The innovation of the developed model is the implementation of Feedforward Neural Networks and Radial Basis Function Neural Networks as estimators of an exogenous parameter that adjusts the covariance matrices of the Extended Kalman Filter process. This hybrid system is evaluated through a time window process leading to promising results, thus enabling a decrease in systematic errors alongside the restriction of the error variability and the corresponding forecast uncertainty. The obtained results showed that the average reduction of the systematic error exceeded 75%, while the corresponding nonsystematic part of that error decreased by 35%. [ABSTRACT FROM AUTHOR]

The input layer, hidden layer, and output layer are three models of neural processors that comprise feedforward neural networks. In this paper, an enhanced tunicate swarm algorithm based on a differential sequencing alteration operator (ETSA) with symmetric cooperative swarms is presented to train feedforward neural networks. The objective is to accomplish minimum classification errors and the most appropriate neural network layout by regulating the layers' connection weights and neurons' deviation thresholds according to the transmission error between the anticipated input and the authentic output. The TSA mimics jet motorization and swarm scavenging to mitigate directional collisions and to maintain the greatest solution that is customized and regional. However, the TSA exhibits the disadvantages of low computational accuracy, a slow convergence speed, and easy search stagnation. The differential sequencing alteration operator has adaptable localized extraction and search screening to broaden the identification scope, enrich population creativity, expedite computation productivity, and avoid search stagnation. The ETSA integrates exploration and exploitation to mitigate search stagnation, which has sufficient stability and flexibility to acquire the finest solution. The ETSA was distinguished from the ETTAO, EPSA, SABO, SAO, EWWPA, YDSE, and TSA by monitoring seventeen alternative datasets. The experimental results confirm that the ETSA maintains profound sustainability and durability to avoid exaggerated convergence, locate the acceptable transmission error, and equalize extraction and prospection to yield a faster convergence speed, superior calculation accuracy, and greater categorization accuracy. [ABSTRACT FROM AUTHOR]

Multi-interest sequential recommendations leverage users' historical behavior to provide recommendations that match multiple interests. Most of these methods have not fully extracted higher-order information hidden in users' interactions and have overlooked the multiple features of items. To this end, this paper proposes a multi-interest model called "multi-interest sequential recommendation with simplified graph convolution and item multi-features (SGCMF)". Firstly, a simplified graph convolution module is designed based on bipartite graphs, which utilizes mean pooling to aggregate neighboring information and employs a feedforward neural network (FNN) for nonlinear transformations and combinations. This method reduces redundant information and captures higher-order relationships, thereby simplifying the complexity of modeling high-order interactions and improving prediction accuracy. Secondly, an item multi-feature extraction module is proposed, which represents item features with multiple vectors, and analyzes each feature from multiple perspectives while preserving important relationships between features. The model correlates multiple features of the item with user interests, thereby achieving a fine-grained analysis of user interests. Extensive experiments are conducted on five real-world scenarios, and the results are compared with state-of-the-art methods. The experimental results show that SGCMF outperforms other baselines. [ABSTRACT FROM AUTHOR]

Huge waves caused by typhoons often induce severe disasters along coastal areas, making the effective prediction of typhoon-induced waves a crucial research issue for researchers. In recent years, the development of the Internet of Underwater Things (IoUT) has rapidly increased the prediction of oceanic environmental disasters. Past studies have utilized meteorological data and feedforward neural networks (e.g., BPNN) with static network structures to establish short lead time (e.g., 1 h) typhoon wave prediction models for the coast of Taiwan. However, sufficient lead time for prediction remains essential for preparedness, early warning, and response to minimize the loss of lives and properties during typhoons. The aim of this research is to construct a novel long lead time typhoon-induced wave prediction model using Long Short-Term Memory (LSTM), which incorporates a dynamic network structure. LSTM can capture long-term information through its recurrent structure and selectively retain necessary signals using memory gates. Compared to earlier studies, this method extends the prediction lead time and significantly improves the learning and generalization capability, thereby enhancing prediction accuracy markedly. [ABSTRACT FROM AUTHOR]

Background: Biologically annotated neural networks (BANNs) are feedforward Bayesian neural network models that utilize partially connected architectures based on SNP-set annotations. As an interpretable neural network, BANNs model SNP and SNP-set effects in their input and hidden layers, respectively. Furthermore, the weights and connections of the network are regarded as random variables with prior distributions reflecting the manifestation of genetic effects at various genomic scales. However, its application in genomic prediction has yet to be explored. Results: This study extended the BANNs framework to the area of genomic selection and explored the optimal SNP-set partitioning strategies by using dairy cattle datasets. The SNP-sets were partitioned based on two strategies–gene annotations and 100 kb windows, denoted as BANN_gene and BANN_100kb, respectively. The BANNs model was compared with GBLUP, random forest (RF), BayesB and BayesCπ through five replicates of five-fold cross-validation using genotypic and phenotypic data on milk production traits, type traits, and one health trait of 6,558, 6,210 and 5,962 Chinese Holsteins, respectively. Results showed that the BANNs framework achieves higher genomic prediction accuracy compared to GBLUP, RF and Bayesian methods. Specifically, the BANN_100kb demonstrated superior accuracy and the BANN_gene exhibited generally suboptimal accuracy compared to GBLUP, RF, BayesB and BayesCπ across all traits. The average accuracy improvements of BANN_100kb over GBLUP, RF, BayesB and BayesCπ were 4.86%, 3.95%, 3.84% and 1.92%, and the accuracy of BANN_gene was improved by 3.75%, 2.86%, 2.73% and 0.85% compared to GBLUP, RF, BayesB and BayesCπ, respectively across all seven traits. Meanwhile, both BANN_100kb and BANN_gene yielded lower overall mean square error values than GBLUP, RF and Bayesian methods. Conclusion: Our findings demonstrated that the BANNs framework performed better than traditional genomic prediction methods in our tested scenarios, and might serve as a promising alternative approach for genomic prediction in dairy cattle. [ABSTRACT FROM AUTHOR]

Estimating feed efficiency (FE) in dairy sheep is challenging due to the high cost of systems that measure individual feed intake. Identifying proxies that can serve as effective predictors of FE could make it possible to introduce FE into breeding programs. Here, 39 Assaf ewes in first lactation were evaluated regarding their FE by 2 metrics, residual feed intake (RFI) and feed conversion ratio (FCR). The ewes were classified into high, medium and low groups for each metric. Milk samples of the 39 ewes were subjected to untargeted metabolomics analysis. The complete milk metabolomic signature was used to discriminate the FE groups using partial least squares discriminant analysis. A total of 41 and 26 features were selected as the most relevant features for the discrimination of RFI and FCR groups, respectively. The predictive ability when utilizing the complete milk metabolomic signature and the reduced data sets were investigated using 4 machine learning (ML) algorithms and a multivariate regression method. The orthogonal partial least squares algorithm outperformed other ML algorithms for FCR prediction in the scenarios using the complete milk metabolite signature (R2 = 0.62 ± 0.06) and the 26 selected features (R2 = 0.62 ± 0.15). Regarding RFI predictions, the scenarios using the 41 selected features outperformed the scenario with the complete milk metabolite signature, where the multilayer feedforward artificial neural network (R2 = 0.18 ± 0.14) and extreme gradient boosting (R2 = 0.17 ± 0.15) outperformed other algorithms. The functionality of the selected metabolites implied that the metabolism of glucose, galactose, fructose, sphingolipids, amino acids, insulin, and thyroid hormones was at play. Compared with the use of traditional methods, practical applications of these biomarkers might simplify and reduce costs in selecting feed-efficient ewes. [ABSTRACT FROM AUTHOR]

This paper studies a multi-hydraulic system (MHS) synchronization control algorithm. Firstly, a general nonlinear asymmetric MHS state space entirety model is established and subsequently the model form is simplified by nonlinear feedback linearization. Secondly, an entirety model-type solution is proposed, integrating a nonlinear model predictive control (NMPC) algorithm with a cross-coupling control (CCC) algorithm. Furthermore, a novel disturbance compensator based on the system's inverse model is introduced to effectively handle disturbances, encompassing unmodeled errors and noise. The proposed innovative controller, known as nonlinear model predictive control—cross-coupling control with deep neural network feedforward (NMPC-CCC-DNNF), is designed to minimize synchronization errors and counteract the impact of disturbances. The stability of the control system is rigorously demonstrated. Finally, simulation results underscore the efficacy of the NMPC-CCC-DNNF controller, showcasing a remarkable 60.8% reduction in synchronization root mean square error (RMSE) compared to other controllers, reaching up to 91.1% in various simulations. These results affirm the superior control performance achieved by the NMPC-CCC-DNNF controller. • The nonlinear model of an asymmetric multi-hydraulic system is established. • The entirety model-type solution is proposed with nonlinear model predictive control and cross-coupling control algorithm. • A novel disturbance compensator based on the inverse deep neural network model is proposed. • Results showed the proposed controller can decrease 60.8% and up to 91.1% sysnchronization root mean square error. [ABSTRACT FROM AUTHOR]

To improve water quality, total daily loads must be established, and this requires determining the quality of the water in rivers, storage tanks, ponds, and coastal areas. Current methods to evaluate water quality involve the collection of water samples for subsequent laboratory analysis. Although these technologies offer precise measurements for a specific location and time, they are expensive, time-consuming, and do not provide the continuous, temporal, or spatial conditions of water quality that are required for managing, assessing, and monitoring water quality. In order to calculate the water quality, the water quality index is modeled using artificial neural network models that incorporate feedforward neural network backpropagation neural networks and radial neural networks. The water quality index of Malaysia’s Klang River was determined by training the artificial network using six major sub-quality parameters. Compared to the current method, the artificial neural network simplifies and expedites the computation of the water quality index. The artificial neural network method could provide a significant saving in terms of money and time while offering a robust assessment of water quality. The proposed method could also be used as an early warning system for pollution of water bodies. The best artificial neural network was the feedforward neural network with one hidden layer containing 5 neurons. Furthermore, conventional approaches for calculating the water quality index rely on empirical equations, often introducing a high degree of approximation and uncertainty into the results. Moreover, these equations cannot be applied when some parameters are not measured. In contrast, the artificial neural network methods and technique offer an efficient and straightforward process for estimating and creating prediction models for water quality index. [ABSTRACT FROM AUTHOR]

Soil potassium is a vital macronutrient and element for both crops and biogeochemical cycles. However, studies on variety of soil potassium and environmental factors in long term cultivated farmlands are still lack. In this study, both soil properties of topsoil (0-20 cm) and environmental factors of ten soil health monitoring station in China were investigated from 1990 to 2018. Seven learning methods were used to build soil total potassium (TK) prediction model. Results showed that the mean value of soil TK content in 0-20 cm was 15.64±0.09 g·kg-1 with the range of 3.17-31.04 g·kg-1. TK significantly increased with both the increasing of soil pH and elevation, while significantly decreased with the increasing of total organic carbon, organic matters, air temperature, amount of precipitation, humidity, and atmospheric pressure. The order of R2 of soil TK prediction models from low to high was Linear Regression < Support Vector Regression < Decision Tree < Light Gradient Boosting Machine < Random Forest < eXtreme Gradient Boosting < Feedforward neural networks with the highest R2 at 0.91 and the lowest values of RMSE and EF. Soil pH, air temperature, and precipitation could be important environmental factors affecting significantly soil total potassium. Machine learning methods generally showed better performances than that linear regression one. Based on long-term and in-situ monitor of soil properties and environmental factors, a sustainability management strategy for farmlands could come true. [ABSTRACT FROM AUTHOR]

Hardware-in-the-loop (HIL) testing is a popular control system testing method because it bridges the gap between modeling/simulation and experiments. Instead of designing a full hardware-based experiment to validate the results of a simulation, the plant hardware can be replaced with an emulator device that responds to exogenous inputs and effectively emulates the dynamic behavior of a system. This approach can be more cost-effective and modular, since the emulated plant system can be modeled in a simulation environment, implemented on a simplified piece of hardware and changed quickly without having to fabricate new parts. This paper develops the hardware and control scheme for a certain type of HIL device called a dynamic load emulator that consists of a 1-DOF linear hydraulic dynamometer equipped with in-line sensing to measure both its own position and the force exerted on it by a device-under-test. This measured force is passed to a real-time model of the emulated dynamic system. The model outputs the emulated system position, and a closed-loop controller is used to emulate this position. The emulator controller incorporates both model-based feedforward and standard feedback PI control. This paper characterizes the dynamometer-based dynamic load emulator and its controller, determining its hardware limitations and validating its capabilities when experiencing a force input from a linear spring with known parameters. Additionally, this paper demonstrates the ability of the emulator to represent the dynamics of a 1-DOF robotic joint when actuated by a pair of fluidic artificial muscles (FAMs). The primary contribution of this work is to allow for more comprehensive testing of FAM configurations, topologies, and controllers for a wide range of parameters, because the same hardware can be used to emulate multiple systems. As a result, this work will lead to more cost-effective, time-efficient, and energy-efficient designs of robotic systems and the FAMs used to actuate them. [ABSTRACT FROM AUTHOR]

This study aims to provide precise predictions for the compression of reinforced polymers during the sheet Molding Compound (SMC) process, ensuring the attainment of a predefined structure while preventing material overflow during the process. The primary challenge revolves around identifying the optimal initial shape to prevent material rebound during the process. To confront this issue, a numerical model is utilized, faithfully simulating the SMC process and forming the foundation for our investigations. Furthermore, to optimize the pre-fill stage, a surrogate model is proposed to enhance modeling efficiency, and then an inverse analysis method is applied. This approach of minimizing material rebound during the SMC process results in a reliable metamodel to predict an initial mass shape accurately and at a low computational cost, thus ensuring the squeezed material fits the mold shape. [ABSTRACT FROM AUTHOR]

In this paper, an effective equivalent modeling technique has been proposed to describe small-signal characteristics of InP-based high electron mobility transistors (HEMTs) after proton radiation, which is composed of an artificial neural network and equivalent-circuit models. Small-signal intrinsic parameters of InP-based HEMTs are extracted from S-parameters before and after 2 MeV proton radiation as modeling objects. The deep learning model of a generative adversarial network has been explored to expand the measured finite data samples. Four feedforward neural networks are incorporated to equivalent-circuit topology to form the equivalent model, which are trained to accurately predict the radiation-induced variations of Cgs, Cgd, Rds, and gm, respectively. The prediction accuracy of the developed equivalent model has been well verified in terms of the broad-band S-parameters under radiation fluence of 1 × 1014 and 5 × 1013 H+/cm2. This equivalent modeling method with characterization of radiation damage effects could provide significant guidance for the aerospace monolithic millimeter-wave integrated circuit design. [ABSTRACT FROM AUTHOR]

Machine learning models that are used for the prediction and control of production can improve quality and yield. However, developing models that are highly accurate and reflective of real-world processes is challenging. We propose a feedforward neural network model specifically designed for continuous Multistage Manufacturing Processes (MMPs) without intermediate outputs. This model, which is termed "MMP Net," can accurately represent the control mechanism of continuous MMPs. Whereas existing studies on learning MMPs assume an intermediate output data, the MMP Net does not require such an unrealistic assumption. We use the MMP Net to develop prediction models for the lubricant base oil production process of a world-leading lubricant manufacturer. Evaluation results show that the MMP Net is superior to other deep neural network and machine learning models. Consequently, the MMP Net was actually implemented in a real factory in 2022 and is expected to save 900,000 dollars per year for each production line. We believe that our work can serve as a basis to develop customized machine learning solutions for improving continuous MMPs. [ABSTRACT FROM AUTHOR]

Many studies have demonstrated that incorporating lexical information into characters can effectively improve the performance of Chinese Named Entity Recognition (CNER). However, we argue that previous studies have not extensively explored the interactive relationship between characters and lexical information, and have only used the lexical information to enhance character-level representation. To address this limitation, we propose an interactive fusion approach that integrates characters and lexical information for CNER. Specifically, we first design graph attention networks to initially fuse character and lexical information within an interactive graph structure. Additionally, by introducing methods such as feedforward neural networks, residual connections, and layer normalization, the fusion effect of the graph attention network is further enhanced. Finally, concatenating and reducing dimensionality of character feature vectors and lexical feature vectors to achieve secondary fusion, thereby obtaining a more comprehensive feature representation. Experimental results on multiple datasets demonstrate that our proposed model outperforms other models that fuse lexical information. Particularly, on the CCKS2020 and Ontonotes datasets, our model achieves higher F1 scores than previous state-of-the-art models. The code is available via the link: https://github.com/wangye0523/The-interactive-fusion-of-characters-and-lexical-information-for-Chinese-named-entity-recognition. [ABSTRACT FROM AUTHOR]

Histone H3 lysine-9 methylation (H3K9me) is a hallmark of the condensed and transcriptionally silent heterochromatin. It remains unclear how H3K9me controls transcription silencing and how cells delimit H3K9me domains to avoid silencing essential genes. Here, using Arabidopsis genetic systems that induce H3K9me2 in genes and transposons de novo, we show that H3K9me2 accumulation paradoxically also causes the deposition of the euchromatic mark H3K- 36me3 by a SET domain methyltransferase, ASHH3. ASHH3-induced H3K36me3 confers anti-silencing by preventing the demethylation of H3K4me1 by LDL2, which mediates transcriptional silencing downstream of H3K9me2. These results demonstrate that H3K9me2 not only facilitates but orchestrates silencing by actuating antagonistic silencing and anti-silencing pathways, providing insights into the molecular basis underlying proper partitioning of chromatin domains and the creation of metastable epigenetic variation. [ABSTRACT FROM AUTHOR]

This article presents a capacitor‐less low dropout (LDO) regulator with a high power supply rejection (PSR) for system‐on‐chip (SoC) applications. With the proposed adaptive bulk‐driven feedforward (ABDFF) technique, in which the ripples are injected adaptively to the bulk of power transistor, the LDO maintains a steady PSR enhancement in wide load conditions, and a lower quiescent current and faster load transient response are realized. Post‐layout simulation results show that the proposed LDO regulator achieves −75‐dB PSR at 1 MHz from 1 to 100 mA and 60‐fs figure of merit (FoM) with a 100‐pF load capacitor. [ABSTRACT FROM AUTHOR]