Improve Model Accuracy Research Articles

Can Denoising Enhance Prediction Accuracy of Learning Models? A Case of Wavelet Decomposition Approach

Denoising is an integral part of the data pre-processing pipeline that often works in conjunction with model development for enhancing the quality of data, improving model accuracy, preventing overfitting, and contributing to the overall robustness of predictive models. Algorithms based on a combination of wavelet with deep learning, machine learning, and stochastic model have been proposed. The denoised series are fitted with various benchmark models, including long short-term memory (LSTM), support vector regression (SVR), artificial neural network (ANN), and autoregressive integrated moving average (ARIMA) models. The effectiveness of a wavelet-based denoising approach was investigated on monthly wholesale price data for three major spices (turmeric, coriander, and cumin) for various markets in India. The predictive performance of these models is assessed using root mean square error (RMSE), mean absolute percentage error (MAPE), and mean absolute error (MAE). The wavelet LSTM model with Haar filter at level 6 emerged as a robust choice for accurate price predictions across all spices. It was found that the wavelet LSTM model had a significant gain in accuracy than the LSTM model by more than 30% across all accuracy metrics. The results clearly highlighted the efficacy of a wavelet-based denoising approach in enhancing the accuracy of price forecasting.

SHAP values accurately explain the difference in modeling accuracy of convolution neural network between soil full-spectrum and feature-spectrum

Acquiring soil nutrient content quickly and accurately through remote sensing is the key to advance precision agriculture. The development of deep learning has provided new technical means for soil hyperspectral modeling. However, the problem of poor interpretability of deep learning models limits its development. Although SHapley Additive exPlanations (SHAP) values based on game theory have been successfully applied to the interpretation of deep learning soil spectral modeling, whether they can accurately explain the differences in deep learning model accuracy remains to be verified. Based on this, we explored whether SHAP values can accurately explain the differences in convolutional neural network (CNN) modeling accuracy. We collected soil samples from agricultural land in the Liangshui River Basin in the southern mountainous and hilly areas of China, and measured the soil total nitrogen (STN) content and soil spectral data in the laboratory. We compared the effects of full-spectrum and feature-spectrum on the accuracy of deep learning models, and obtained the contribution of wavelengths in the CNN modeling process by calculating SHAP values. The results showed that combining different spectral pre-processing methods can play their respective advantages and help improve modeling accuracy. Among them, the CNN model obtained the highest prediction accuracy under the first-derivative Savitzky-Golay smoothing combination standard normal variate (SG1-SNV) spectral pre-processing in full-spectrum modeling. Compared with the feature-spectrum selected for modeling by Mutual information (MI) and competitive adaptive reweighted sampling (CARS), the CNN model achieved higher accuracy in most pre-processed spectra in full-spectrum modeling, and SHAP values accurately explained this reason. This is because the contribution is usually higher at most wavelengths with a high correlation with STN content. The feature-spectrum selected by CARS is more widely distributed but lacks continuity, and some wavelengths with high correlation and high contribution will also be missed. Meanwhile, some wavelengths with low correlation also have high contributions, which are usually not involved in the feature spectrum modeling of MI, thus affecting the modeling accuracy. Therefore, the deep learning model is more suitable for full-spectrum modeling due to its strong feature extraction and self-learning capabilities, and SHAP can obtain the wavelength contribution of the CNN model in soil spectral modeling, and then explain the differences in modeling accuracy. This study further proves the interpretability of deep learning, provides an important basis for the application of deep learning in soil hyperspectral modeling.

Innovative compressive strength prediction for recycled aggregate/concrete using K-nearest neighbors and meta-heuristic optimization approaches

This paper presents a groundbreaking method for predicting the compressive strength (Fc) of recycled aggregate concrete (RAC) through the application of K-nearest neighbors (KNN) analysis. The task of designing mixture proportions to achieve the desired Fc can be remarkably intricate, owing to the intricate interplay among the components involved. Machine learning (ML) algorithms have exhibited considerable promise in tackling this complexity effectively. In pursuit of enhanced prediction accuracy, this research introduces a semi-empirical approach that seamlessly integrates strategies, including optimization techniques. This study incorporates two meta-heuristic methods, the Fire Hawk optimizer (FHO) and Runge–Kutta optimization (RUK) to enhance model accuracy. The research results reveal three separate models: KNFH, KNRK, and a single KNN model, each providing valuable insights for precise Fc prediction. Remarkably, the KNFH model stands out as a top performer, boasting an impressive R2 value of 0.994 and a meager RMSE value of 1.122. These findings not only validate the accuracy and reliability of the KNFH model but also highlight its effectiveness in predicting Fc outcomes. This approach holds great promise for precise Fc forecasting in the construction industry. Integrating meta-heuristic algorithms significantly improves model accuracy, leading to more reliable forecasts with profound implications for construction projects and their outcomes. This research marks a significant advancement in predicting Fc using ML, offering valuable tools for engineers and builders.

On the influence of vertical mixing, boundary layer schemes, and temporal emission profiles on tropospheric NO2 in WRF-Chem – comparisons to in situ, satellite, and MAX-DOAS observations

Abstract. We present WRF-Chem simulations over central Europe with a spatial resolution of 3 km × 3 km and focus on nitrogen dioxide (NO2). A regional emission inventory issued by the German Environmental Agency, with a spatial resolution of 1 km × 1 km, is used as input. We demonstrate by comparison of five different model setups that significant improvements in model accuracy can be achieved by choosing the appropriate boundary layer scheme, increasing vertical mixing strength, and/or tuning the temporal modulation of the emission data (“temporal profiles”) driving the model. The model setup with improved vertical mixing is shown to produce the best results. Simulated NO2 surface concentrations are compared to measurements from a total of 275 in situ measurement stations in Germany, where the model was able to reproduce average noontime NO2 concentrations with a bias of ca. −3 % and R=0.74. The best agreement is achieved when correcting for the presumed NOy cross sensitivity of the molybdenum-based in situ measurements by computing an NOy correction factor from modelled peroxyacetyl nitrate (PAN) and nitric acid (HNO3) mixing ratios. A comparison between modelled NO2 vertical column densities (VCDs) and satellite observations from TROPOMI (TROPOspheric Monitoring Instrument) is conducted with averaging kernels taken into account. Simulations and satellite observations are shown to agree with a bias of +5.5 % and R=0.87 for monthly means. Lastly, simulated NO2 concentration profiles are compared to noontime NO2 profiles obtained from multi-axis differential optical absorption spectroscopy (MAX-DOAS) measurements at five locations in Europe. For stations within Germany, average biases of −25.3 % to +12.0 % were obtained. Outside of Germany, where lower-resolution emission data were used, biases of up to +50.7 % were observed. Overall, the study demonstrates the high sensitivity of modelled NO2 to the mixing processes in the boundary layer and the diurnal distribution of emissions.

A continuous digital elevation representation model for DEM super-resolution

The surface of the Earth is continuous, and obtaining precise elevation data at arbitrary query positions is essential for many applications and analyses. However, existing digital elevation models suffer from a precision gap caused by discretization. Therefore, we develop a new continuous representation model (CDEM) that allows height values to be obtained at any arbitrary query position. Inspired by recent research on the implicit neural representation model, we train an encoder–decoder network to learn CDEM from discrete elevation data for DEM super-resolution tasks. The encoder targets generating latent codes from discrete elevation data, while the decoder composes these latent codes with query positions to predict corresponding elevation values. Such a learning pipeline is well-suited for DEM super-resolution tasks. To improve model accuracy, we also propose predicting the bias of elevation values between the query position and its closest known position. Real-world terrain surfaces exhibit inherent roughness with numerous small variations in localized regions, resulting in high-frequency targets that are difficult for neural networks to fit. In order to facilitate the network’s ability to model data with high-frequency variations, we introduce positional encoding to map query positions into a higher-dimensional space. Compared to the Bicubic interpolation method and state-of-the-art TfaSR model, our method is demonstrated to obtain more accurate elevation values and preserve more details of terrain structure on the TFASR30, Pyrenees, and Tyrol datasets. Specifically, our EBCF-CDEM model demonstrates performance improvements over the TfaSR model, with reductions in the RMSE of elevation, slope, and aspect by 7.03%, 4.81%, and 3.07% respectively at ×4 super-resolution scale on the TFASR30 dataset, by 9.92%, 7.06%, and 6.15% at ×8 super-resolution scale on the Pyrenees dataset. Extensive experiments further validate the generalizability of our EBCF-CDEM, compared to the TfaSR model, our results in terms of RMSE of elevation, slope, and aspect are reduced by 14.78%, 12.12%, and 8.99% respectively at ×8 super-resolution scale on the Tyrol dataset, and 26.92%, 7.14% and 5.36% on the TIFASR30to10 dataset. We release our source code (including the datasets) at https://github.com/AlcibiadesTophetScipio/EBCF-CDEM to reproduce our results and encourage future research.

Research on Sentiment Analysis of E-commerce Live Comments based on Text Mining

With the rapid rise of live e-commerce in the field of e-commerce, the rich emotional information generated by users in live comments has become an important research object. This study takes e-commerce live comments as samples, uses text mining technology combined with machine learning methods, builds sentiment dictionary, trains sentiment analysis model and uses SnowNLP library for sentiment analysis, and deeply digs users' emotional tendency towards goods, services and shopping experience. For negative comments, word cloud is generated through word segmentation and word frequency calculation to visually show users' dissatisfaction and provide substantial support for updating the emotional dictionary. Through this comprehensive analysis, it provides a deeper user insight for e-commerce enterprises, and provides strong support for operational decisions and service optimization. Future research can focus on the application of deep learning techniques in sentiment analysis to further improve model accuracy and adaptability.

Improving Network Security: An Intrusion Detection System (IDS) Dataset from Higher Learning Institutions, Mbeya University of Science and Technology (MUST), Tanzania

Nowadays, Internet-driven culture securing computer networks in Higher Learning Institutions (HLIs) has become a major responsibility. Intrusion Detection Systems (IDS) are crucial for protecting networks from unauthorized activity and cyber threats. This paper examines the process of improving network security by creating a comprehensive IDS dataset using real traffic from HLIs, highlighting the importance of accurate and representative data in improving the system's ability to identify and mitigate future cyber-attacks. The IDS model was created using a variety of machine learning (ML) techniques. Metrics like accuracy, precision, recall, and F1-score were used to assess the performance of each model. The dataset used for training and testing was real-world network traffic data obtained from the institution's computer network. The results showed that the developed IDS obtained exceptional accuracy rates, with Random Forest, Gradient Boosting, and XGBoost models all achieving an accuracy of around 93%. Precision and recall values were likewise quite high across all algorithms. Furthermore, the study discovered that data quality has a substantial impact on IDS performance. Proper data preparation, feature engineering, and noise removal were found to be helpful in improving model accuracy and reducing false positives. While the IDS models performed well throughout validation and testing, implementing such systems in a production setting necessitates careful thought. As a result, the essay also examined the procedures for testing and deploying the IDS models in a real-world scenario. It underlined the significance of ongoing monitoring and maintenance in order to keep the model effective in identifying intrusions. The research aids in the progress of network security in HLI. Educational institutions can better protect their precious assets and sensitive information from cyberattacks by understanding the impact of data quality on IDS performance and implementing effective deployment techniques

Enhancing The Accuracy of Image Classification Using Deep Learning and Preprocessing Methods

Deep learning is one of many methods in Artificial Intelligence (AI) that computers can use to process information like text, images, and audio. This manuscript will be focusing on image preprocessing, one of the many different techniques that are used to modify the neural network model training process, and how it affects the training speed and accuracy of the neural network. Six different image preprocessing techniques were picked for use in this study: Grayscale, Smoothing, Unmask Sharpening, Laplacian and Equalization, and Random Cropping and Rotation all of which were implemented using Python and the libraries NumPy, OpenCV, and PyTorch. For the dataset, a batch of 10000 images from the CIFAR10 dataset were used to train the model. This study explored the impact of preprocessing techniques on a deep learning model, employing the RESNET50 architecture. Notable improvements in model accuracy were observed, particularly with normalization and random cropping accompanied by rotation. The efficiency gains attributed to preprocessing were highlighted, leading to a more rapid training process and significant resource savings. This research underscores the importance of thoughtful preprocessing in enhancing the performance of deep learning models, offering valuable insights for practitioners in imageclassification tasks.

Spherical Harmonics for the 1D Radiative Transfer Equation. II. Thermal Emission

Approximate methods for radiative transfer equations that are fast, reliable, and accurate are essential for the understanding of atmospheres of exoplanets and brown dwarfs. The simplest and most popular choice is the “two-stream method,” which is often used to produce simple yet effective models for radiative transfer in scattering and absorbing media. Toon et al. (hereafter, Toon89) outlined a two-stream method for computing reflected light and thermal spectra that was later implemented in the open-source radiative transfer model PICASO. In Part I of this series, we developed an analytical spherical harmonics method for solving the radiative transfer equation for reflected solar radiation that was implemented in PICASO to increase the accuracy of the code by offering a higher-order approximation. This work is an extension of this spherical harmonics derivation, to study thermal emission spectroscopy. We highlight the model differences in the approach for thermal emission and benchmark the four-term method (SH4) against Toon89 and a high-stream discrete-ordinates method, CDISORT. By comparing the spectra produced by each model, we demonstrate that the SH4 method provides a significant increase in accuracy, compared to Toon89, which can be attributed to the increased order of approximation and to the choice of phase function. We also explore the trade-off between computational time and model accuracy. We find that our four-term method is twice as slow as our two-term method, but is up to five times more accurate, when compared with CDISORT. Therefore, SH4 provides excellent improvement in model accuracy with minimal sacrifice in numerical expense.

Tree-PointNet: A Novel Neural Network for Dormant Apple Trees Using Semi-Circle-Based Point Cloud for Robotic Pruning

HighlightsTree-PointNet, a novel neural network, integrates geometric down-sampling for efficient pruning.A Semi-Circle-Based point cloud dataset of apple trees was collected to aid in automated pruning.Enhanced algorithm performance was investigated, showing significant improvements in model accuracy. Abstract. Pruning during the dormant season is one of the most costly and labor-intensive operations in the production of specialty crops. In winter, many trained seasonal workers have to carefully remove branches from hundreds of trees according to pre-defined rules. The aim of automated pruning is to reduce this reliance on the large amount of labor currently required for the operation. The first step in the automatic application of pruning rules is the acquisition of point cloud data from trees. To match the optical vision system of an ideal automatic pruning robot, a brand new and challenging apple tree dataset was collected in which the point clouds were all based on semicircular shapes, rather than the carefully crafted point clouds of complete apple trees. 1000 Semi-Circle-Based point clouds of apple trees were used. For the brand-new structure of point cloud part segmentation, a neural network model (namely Tree-PointNet) was proposed. A Semi-Circle-Based point cloud of apple tree was segmented by the Tree-PointNet with the segmentation result consisting of a trunk and primary branches. With a weight size of only 7.9% of PointStack, the best OA of 95.6% was obtained by the Tree-PointNet network proposed in this paper. The GDS was also used, which could obtain more key points in the Branch-to-Trunk connection region as input to the proposed network. On the other hand, the OA and mIoU of the model improved by 10.2% and 7.98%, respectively, compared to the PointNet++. Keywords: Dormant apple trees, Part segmentation, PointNet++, Semi-circle-based point cloud.

Research on incentive mechanisms for anti-heterogeneous federated learning based on reputation and contribution

<abstract> <p>An optimization algorithm for federated learning, equipped with an incentive mechanism, is introduced to tackle the challenges of excessive iterations, prolonged training durations, and suboptimal efficiency encountered during model training within the federated learning framework. Initially, the algorithm establishes reputation values that are tied to both time and model loss metrics. This foundation enables the creation of incentive mechanisms aimed at rewarding honest nodes while penalizing malicious ones. Subsequently, a bidirectional selection mechanism anchored in blockchain technology is developed, allowing smart contracts to enroll nodes with high reputations in training sessions, thus filtering out malicious clients and enhancing local training efficiency. Furthermore, the integration of the Earth Mover's Distance (EMD) mechanism serves to lessen the impact of non-IID (non-Independent and Identically Distributed) data on the global model, leading to a reduction in the frequency of model training cycles and an improvement in model accuracy. Experimental results confirm that this approach maintains high model accuracy in non-IID data settings, outperforming traditional federated learning algorithms.</p> </abstract>

Towards addressing aggregation deviation for model training in resource-scarce edge environment

Federated Learning (FL) can train models in an edge environment without sending raw data. However, the performance is still constrained by data heterogeneity. To address the problems of data heterogeneity and resource scarcity in edge devices, we propose Federal Learning via Dynamic Aggregation (FedDA), which eliminates the influence of data heterogeneity and improves model accuracy. FedDA updates the impact of individual local models on the global model in real-time at different stages. It adjusts the local epoch in each round to prevent the device from dropping out while obtaining a more accurate local model. The core module is the model impact factor (MIF) that inscribes the aggregation weights to solve the impact of fixed weights on the aggregation model with improper extraction of local information. We conducted several experiments to evaluate the convergence speed using different algorithms on the MINIST. FedDA consistently outperforms the other six SOTA algorithms on MNIST, Cifar10, and Cifar100 datasets. In significant data heterogeneity, FedDA improves accuracy by up to 6% over the different algorithms and at least about 3%, especially in resource-scarce environments. To reach the specified accuracy, FedDA is 3 times faster than SCAFFOLD and at least 50% faster than other algorithms.

Optimizing Artificial Neural Network Learning Using Improved Reinforcement Learning in Artificial Bee Colony Algorithm

Artificial neural networks (ANNs) are widely used machine learning techniques with applications in various fields. Heuristic search optimization methods are typically used to minimize the loss function in ANNs. However, these methods can lead the network to become stuck in local optima, limiting performance. To overcome this challenge, this study introduces an improved optimization approach, the improvement of reinforcement learning in the artificial bee colony (improved R‐ABC) algorithm, to enhance the optimization process for ANNs. The proposed method aims to overcome the limitations of heuristic search and improve the efficiency of weight adjustment in ANNs. This new approach enhances the discovery phase of the traditional R‐ABC by including the parameters of neighboring food sources, augmenting the search capabilities for finding the optimal solution. The performance of the improved R‐ABC was compared with ANNs utilizing backpropagation with stochastic gradient descent (SGD) and Adam optimizers, as well as other swarm intelligence (SI) methods such as particle swarm optimization (PSO) and traditional R‐ABC. The results showed that both PSO and R‐ABC continuously improved the solutions across all benchmark datasets. In the iris dataset, all SI approaches consistently achieved F1‐scores exceeding 0.94, outperforming SGD and Adam. For the other datasets, the SI approach generally outperformed the other optimization methods. The results indicate that when the improved R‐ABC is applied to ANNs, it outperforms heuristic search optimization, especially as the network size expands. Although SGD and Adam achieved faster execution times with TensorFlow, the study suggests that using PSO and improved R‐ABC can improve model accuracy and efficiency. Advanced SI methods enhance the optimization process and increase the ability of ANNs to obtain optimal solutions. Enhanced R‐ABC and PSO algorithms can significantly improve ANN training performance and efficiency, especially in complex and high‐dimensional datasets.

APPLICATION OF PINN AND THE METHOD OF DIFFERENTIAL CONSTRUCTION OF ADAPTIVE ONE-DIMENSIONAL COMPUTATIONAL GRIDS

This paper presents an approach to constructing adaptive one-dimensional computational grids using the Beltrami equation and Physics-Informed Neural Networks (PINNs). The main focus is on exploring the potential for precise control of grid node density through the control function ω(s), which allows the grid to adapt to the local features of the problem. The Beltrami equation used, being a key component of the method, regulates the distribution of nodes by modifying the function’s derivatives depending on the values of the control function. The effectiveness of this approach is demonstrated through examples involving one and two regions of node clustering. The results showed that the PINN method combined with the Beltrami equation allows for the creation of computational grids with a high degree of adaptation to given conditions, providing detailed modeling in critical regions. This approach has advantages over traditional numerical methods, as integrating physical laws in the grid construction process minimizes numerical errors and improves modeling accuracy. The use of neural networks offers flexibility in model tuning and the ability to account for complex nonlinear dependencies. The discussion of the results highlights the potential of using PINNs for adaptive grid construction in various fields requiring precise and efficient modeling. In conclusion, this study confirms that the combination of the Beltrami equation and PINNs is a powerful tool for adaptive grid construction, opening new possibilities for numerical modeling of complex physical processes.

Development of a neural network structure for recognition of attacks in computer systems

The integration of industrial control communication networks and the Internet into the Industrial Control System (ICS) increases their vulnerability to cyber-attacks, leading to devastating consequences. Traditional intrusion detection systems (IDS) mostly rely on predefined models and are trained mostly on specific cyber attacks, which means that traditional IDS cannot deal with unknown attacks In addition, most IDSs do not take into account the imbalance of ICS datasets, and therefore suffer from low accuracy and high false positives when used. In the article, we propose an NCO-double-layer DIFF_RF-OPFYTHON intrusion detection method for ICS, which consists of NCO modules, two-layer DIFF_RF modules, and OPFYTHON modules. Detected traffic will be divided into three categories by the two-layer DIFF_RF module: known attacks, unknown attacks, and normal traffic. Next, the known attacks will be classified by the OPFYTHON module into specific attacks according to the characteristics of the attack traffic. We use the NCO module to improve model inputs and improve model accuracy. The results show that the proposed method outperforms traditional intrusion detection methods such as XGboost and SVM. Detecting unknown attacks is also significant. The accuracy of the data set used in this article reaches 98.13%. The detection rate of unknown and known attacks reaches 98.21% and 95.1%, respectively. In this article, the basic modules of the classifier are shallow machine learning algorithms. This can be improved by using a more powerful neural network architecture. In the article, all detected unknown attacks are marked as one category. In further studies, the unknown attacks can be clustered and analyzed to further classify the detected unknown attacks. It is worth noting that the number of unknown attacks is only a small part, so it is difficult to classify unknown attacks using cluster analysis. Since training a more powerful neural network requires a lot of data, it is important to investigate how to train a new model when the number of samples belonging to that class is limited.

Sonic Log Prediction Based on Extreme Gradient Boosting (XGBoost) Machine Learning Algorithm by Using Well Log Data

Sonic log is an important aspect that provides a detailed description of the subsurface properties associated with oil and gas reservoirs. The problem that frequently occurs is the unavailability of sonic log data for various reasons needs to be given an effective solution. The alternative approach proposed in this research is sonic log prediction based on Extreme Gradient Boosting (XGBoost) machine learning algorithm, using available log data to build a reliable sonic log prediction model. In this research, the predicted DT log type is the Differential Time Shear Slowness (DTSM) log, which is the velocity of shear waves propagating in a formation. Log features used for training include gamma ray (GR), density (RHOB), porosity (NPHI), resistivity (RS and RD) logs with DTSM log as the prediction target. To optimise the performance and generalisation of the XGBoost algorithm in predicting log DTSM, hyperparameter tuning was applied using grid search technique to obtain optimal parameters for the prediction model. Based on the experimental results, this research found that hyperparameter tuning using grid search technique improved the accuracy of sonic log (DTSM) model prediction based on XGBoost algorithm, as proven by the decrease of RMSE and MAPE values to 19.699 and 7.713%. The results also pointed out the need for methods other than listwise deletion to handle missing values as an alternative to improving model accuracy. This research highlighted the need for continuous improvement in data processing methods and algorithm optimization to advance the application of machine learning in geophysical exploration.

Impacts of g-g-v Constraints Formulations on Online Minimum-Time Vehicle Trajectory Planning

The g-g-v diagram is a popular representation of a vehicle’s maximum longitudinal and lateral accelerations, as a function of its speed. Recently, g-g-v diagrams have been used as performance constraints for online minimum-time vehicle trajectory planning with economic nonlinear model predictive control (E-NMPC). However, in the related literature, the modeling accuracy of the g-g-v formulations was often compromised in favor of computational efficiency for online E-NMPC. This paper compares various formulations of the g-g-v constraints, evaluating their accuracy, computational efficiency and impact on the lap times, when applied by an artificial race driver (ARD) to control a sports car model on a circuit. Also, we propose a new g-g-v diagram formulation, based on convex polytopic and non-convex polynomial constraints. Our formulation yields improved modeling accuracy while preserving a low computational burden for online E-NMPC. Finally, we analyze the vehicle trajectories, to understand how ARD achieves lower lap times through an improved knowledge of the maximum performance.

Federated Data Quality Assessment Approach: Robust Learning With Mixed Label Noise.

Federated learning (FL) has been an effective way to train a machine learning model distributedly, holding local data without exchanging them. However, due to the inaccessibility of local data, FL with label noise would be more challenging. Most existing methods assume only open-set or closed-set noise and correspondingly propose filtering or correction solutions, ignoring that label noise can be mixed in real-world scenarios. In this article, we propose a novel FL method to discriminate the type of noise and make the FL mixed noise-robust, named FedMIN. FedMIN employs a composite framework that captures local-global differences in multiparticipant distributions to model generalized noise patterns. By determining adaptive thresholds for identifying mixed label noise in each client and assigning appropriate weights during model aggregation, FedMIN enhances the performance of the global model. Furthermore, FedMIN incorporates a loss alignment mechanism using local and global Gaussian mixture models (GMMs) to mitigate the risk of revealing samplewise loss. Extensive experiments are conducted on several public datasets, which include the simulated FL testbeds, i.e., CIFAR-10, CIFAR-100, and SVHN, and the real-world ones, i.e., Camelyon17 and multiorgan nuclei challenge (MoNuSAC). Compared to FL benchmarks, FedMIN improves model accuracy by up to 9.9% due to its superior noise estimation capabilities.

Self-Constructing Chebyshev Fuzzy Neural Complementary Sliding Mode Control and Its Application.

In this article, a complementary sliding mode (CSM) controller using a self-constructing Chebyshev fuzzy recurrent neural network (SCCFRNN) is proposed for harmonic suppression control of an active power filter (APF). The SCCFRNN whose structure can be automatically learned through the designed structure self-learning algorithm is introduced to approximate the unknown nonlinear term in the APF dynamic model, so as to improve modeling accuracy and reduce the burden of CSM control (CSMC). The SCCFRNN combines the advantages of a fuzzy neural network (FNN), recurrent neural network (RNN), and Chebyshev neural network (CNN), and all parameters can be adjusted according to the designed adaptive laws. Eventually, through detailed simulation, hardware experiments, and fair comparison, the feasibility and superiority of the proposed control algorithm were verified.

Nonlinear diffusion‐limited two‐dimensional colliding‐plume simulations with very high‐order numerical approximations

AbstractAtmospheric numerical models play a crucial role in operational weather forecasting, as well as in improving our understanding of atmospheric dynamics via research studies. Maximising their accuracy is of paramount importance. Use of advective flux schemes greater than O7 in atmospheric models is largely undocumented, with no studies considering O3–O17 fluxes with formal accuracy‐preserving high‐order interpolation, pressure gradient/divergence, and subgrid‐scale (SGS) turbulent fluxes. Higher‐order numerical approximations can reduce truncation, amplitude and phase errors, and potentially improve model accuracy and effective resolution. Here, simulations are presented using very‐high‐order O3–O17 fluxes, with/without high‐order O2–O18 Lagrangian interpolations, pressure gradient/divergence approximations, and SGS turbulent fluxes for a two‐dimensional, highly‐viscous (Re ∼ 100) diffusion‐limited, nonlinear colliding‐plumes problem using 25–200 m spatial resolutions. The highest‐order flux schemes coupled with higher‐order interpolations, pressure gradient/divergence and SGS flux approximations produced the best solutions, with higher‐order fluxes and interpolations being most important. Overall solution convergence of order about O1–O2 with mode‐split (fast sound/slow advective waves) O3 Runge–Kutta temporal schemes was negatively impacted by order at most O1 temporal convergence with SGS fluxes, divergence damping, and especially spatial filters, compared to about order O3 convergence with these inactivated. While very‐high‐order schemes were shown to improve solution accuracy, few cost‐effective higher‐order highly‐viscous test problem solutions (higher order vs finer resolution) were found using theoretical floating‐point operations (FPO) with Courant‐Friedrichs‐Lewy (CFL)‐limited or constant stable Courant number‐based time steps. However, employing central processing unit (CPU) time, rather than FPOs, demonstrated there was reduced computational burden using higher‐order approximations. We conclude that O9–O17 flux schemes with or without high‐order (≥O4) interpolations, pressure gradient/divergence approximations, and SGS fluxes can improve atmospheric model solution accuracy, without prohibitive computation costs, compared to O3–O7 flux with O2 interpolations, pressure gradient/divergence approximations, and SGS fluxes.