Sigmoid Activation Function Research Articles

Design and Performance Analysis of Low Power and High Throughput of Analog Data Compression and Decompression using ANN in 32nm FinFET Technology

The development and fabrication of integrated circuits for the applicational areas of VLSI such as processing of the signal, medicine tomography, telecommunication turn out to be a novel technology for the upcoming innovations. The fabrication of IC’s is attributable to the methodology in the technology of VLSI and when compared to artificial Neural Network, the genetic performance of these productions is approximately the same and are typically employed for diagnosing the syndrome, compression as well as the decompression of signal used in the medical domain. Techniques such as HMM, DCT, as well as PCA are employed for compression and decompression of signals but these approaches still possess some disadvantages. Therefore, to overcome these issues, a chip-level design for Artificial Neural Network is proposed that makes use of FinFET 32 nm technology and includes sigmoid activation function (SAF), Gilbert cell number, as well as bias circuits to prolong the compressed magnitude relation and accuracy. As a result, with the help of the Cadence Virtuoso analog tool, the Artificial Neural Network has been designed using FinFET 32nm technology along with all the details of sub-units such as Layout vs Schematic (LVS), Design rule check (DRC), RC extraction as well as chip level (GDS-II). Feed Forward Artificial Neural Network (FWANN) is considered as one of the most basic types of ANN and it is implemented using the concept of Back Propagation (BP). The simulation results of the suggested 16-bit 6TRAM cell were found to have 8%, 21%, and 0.9% improvement in consuming power, delay, and compressed data losses respectively.

DECISION SUPPORT SYSTEM FOR DETERMINING THE OPTIMAL COMPOSITION OF THE TEAM OF PERFORMERS ON THE EXAMPLE OF THE SPORTS VERSION OF THE GAME “WHAT? WHERE? WHEN?"

The paper describes the task to create a decision support system that allows you to determine the optimal composition of the team of performers. The sports version of the game “What? Where? When?" is chosen. The principles of holding tournaments on the intellectual game “What? Where? When?" and the rules for the formation of teams to participate in such competitions. It is concluded that to predict the impact of changes in the team composition on the result, it is advisable to use modern mathematical and intellectual methods, including the method of artificial neural networks. The available data on the results of synchronous tournaments LUK (ST) since 2011 and city tournaments (GT) since 2017 are presented, the main indicators for each competition are characterized. The introduction of the output factors is substantiated: the ratio of the team’s result to the average result and the ratio of the team’s result to the winner’s result. The forecasting problem is formulated as a prediction of the relative result of a team on a specific game based on the available list of team players for this game. It is proposed to take into account the location of a particular game, and fix the participation of players in the form of a “share” of the contribution to the team’s result, while the sum of the “shares” of all players should be equal to one. A method of artificial neural networks with a two-layer perceptron architecture, a sigmoid activation function and an error propagation algorithm for training a network is proposed. Examples of calculation in the Deductor Studio Lite environment are given. It is concluded that for the practical application of the model, the constant use of standard packages is inapplicable. In addition, it is also necessary to solve the problem of automating the selection of the team composition. Described is an application developed in a visual programming environment – a decision support system that allows you to import source data from an XLS file, configure input and output factors, change the architecture of the neural network (the number of hidden layers and the number of neurons in each layer), train the neural network using the backpropagation of errors, save the trained network on disk and load it again, calculate values for the input data, search for options for the composition of the team. The developed decision support system makes it possible to give recommendations on the formation of a team for a specific tournament by enumerating options.
 Keywords: intellectual games, questions package, team building, forecasting, artificial neural network, perceptron, sigmoid, network training.

FORECASTING THE RESULTS OF THE SINGLE ENTRANCE EXAM IN A FOREIGN LANGUAGE BY BACHELORS OF THE INSTITUTION OF HIGHER EDUCATION

The paper provides information on the need to pass the “Unified entrance exam” in a foreign (English) language by students who have received a bachelor’s degree and wish to continue their studies to obtain a master’s degree. It is determined that when working with undergraduate students, it is advisable, firstly, to determine the percentage of graduates whose passing EVE is unlikely, and secondly, to intensify work with such graduates to increase this probability. The task was set to create a model for predicting the results of the unified entrance exam in a foreign language by bachelor’s graduates of higher education institutions upon entering the master’s program. A number of factors that affect the EVE score are proposed, namely: competitive score at enrollment (indicator of the student’s basic level), rating (assessment) based on the results of the first year of study (exam in the compulsory subject “Foreign Language”), choice “Foreign language” in the 2-3rd year (maximum of all or “0”, if the student did not choose), the rating of additional classes “Foreign language” in the 4th year, the average rating for the penultimate session (indicator “current” student’s attitude to the educational process), the fact of having additional points (an indicator of the student’s interest in other activities than learning), the average rating of a bachelor’s degree (an indicator of the general student’s attitude to the educational process). The available data concerning students of two years of the department of intelligent decision-making systems of the Donbas State Engineering Academy are given. A method of artificial neural networks with a two-layer perceptron architecture with ten neurons in each hidden layer, a sigmoid activation function, and an error backpropagation algorithm for network training is proposed. Calculations were performed in the Deductor Studio Lite environment, their results were analyzed. It is noted that the proposed approach to forecasting can be applied when working with undergraduate students, to determine the percentage of graduates whose EVI is unlikely to pass, and to intensify work with such graduates to increase this probability.
 Keywords: educational and qualification level, the only entrance exam, forecasting, artificial neural network, perceptron, sigmoid, network training.

An Artificial Intelligence Solution for Predicting Short-Term Degradation Behaviors of Proton Exchange Membrane Fuel Cell

The dead-ended anode (DEA) and anode recirculation operations are commonly used to improve the hydrogen utilization of automotive proton exchange membrane (PEM) fuel cells. The cell performance will decline over time due to the nitrogen crossover and liquid water accumulation in the anode. Highly efficient prediction of the short-term degradation behaviors of the PEM fuel cell has great significance. In this paper, we propose a data-driven degradation prediction method based on multivariate polynomial regression (MPR) and artificial neural network (ANN). This method first predicts the initial value of cell performance, and then the cell performance variations over time are predicted to describe the degradation behaviors of the PEM fuel cell. Two cases of degradation data, the PEM fuel cell in the DEA and anode recirculation modes, are employed to train the model and demonstrate the validation of the proposed method. The results show that the mean relative errors predicted by the proposed method are much smaller than those by only using the ANN or MPR. The predictive performance of the two-hidden-layer ANN is significantly better than that of the one-hidden-layer ANN. The performance curves predicted by using the sigmoid activation function are smoother and more realistic than that by using rectified linear unit (ReLU) activation function.

Photovoltaic System MPPT Evaluation Using Classical, Meta-Heuristics, and Reinforcement Learning-Based Controllers: A Comparative Study

Maximum power point tracking (MPPT) entails constraining photovoltaic (PV) modules to operate under a specified power condition. It has previously been shown that some meta-heuristic techniques often suffer from steady-state oscillations around maximum points and experience difficulty in adapting to environmental variations, such as irradiation and/or temperature. To address the aforementioned limitation, this work proposed an adaptable reinforcement learning (RL) technique based on a novel deep deterministic policy gradient (DDPG) agent and a reward function. The actor–network top layer uses a sigmoid activation function and the critic–network contains bottleneck layers with non-uniform nodal distributions as well as exponential linear unit (ELU) activation functions in some of the layers. The RL based on DDPG method was compared with Particle Swarm Optimization (PSO) and Perturb-and-Observe (P&O) in order to determine the optimal duty-cycle command needed for controlling the PV modules MPPT. All the investigated systems were implemented in MATLAB/Simulink. The results show that the proposed RL technique based on DDPG agent yielded superior tracking efficiency than all the other approaches. However, as the step change in irradiation at a constant temperature increases, the RL technique based on DDPG agent shows a decrease in tracking efficiency.

Logish: A new nonlinear nonmonotonic activation function for convolutional neural network

Activation function is an important component of the convolutional neural network. Recently, nonlinear nonmonotonic activation functions such as Swish and Mish have illustrated good performance in deep learning structures. In this paper, we propose a new nonlinear nonmonotonic activation function called Logish, which can be represented by f(x)=x·ln[1+sigmoid(x)]. Firstly, we take the logarithmic operation to reduce the numerical range of sigmoid(x)+1, then we employ variable x to make the negative output have a strong regularization effect. Furthermore, we evaluate the image classification performance of Logish and its variant f(x)=αx·ln[1+sigmoid(βx)] in simple and complex networks with top–1 accuracy. Experimental results demonstrate that Logish’s variant (α=1,β=10) can achieve 94.8% top-1 accuracy with ResNet–50 network on CIFAR 10 dataset, and can reach 99.24% top-1 accuracy with DenseNet on MNIST dataset and 88.52% top-1 accuracy with SE-Inception-v4 network on SVHN dataset respectively. It is higher than the Sigmoid, Tanh, ReLU, Swish and Mish activation functions in the corresponding dataset. It also verifies the performance and effectiveness of Logish.

Detection of the location of pneumothorax in chest X-rays using small artificial neural networks and a simple training process

The purpose of this study was to evaluate the diagnostic performance achieved by using fully-connected small artificial neural networks (ANNs) and a simple training process, the Kim-Monte Carlo algorithm, to detect the location of pneumothorax in chest X-rays. A total of 1,000 chest X-ray images with pneumothorax were taken randomly from NIH (the National Institutes of Health) public image database and used as the training and test sets. Each X-ray image with pneumothorax was divided into 49 boxes for pneumothorax localization. For each of the boxes in the chest X-ray images contained in the test set, the area under the receiver operating characteristic (ROC) curve (AUC) was 0.882, and the sensitivity and specificity were 80.6% and 83.0%, respectively. In addition, a common currently used deep-learning method for image recognition, the convolution neural network (CNN), was also applied to the same dataset for comparison purposes. The performance of the fully-connected small ANN was better than that of the CNN. Regarding the diagnostic performances of the CNN with different activation functions, the CNN with a sigmoid activation function for fully-connected hidden nodes was better than the CNN with the rectified linear unit (RELU) activation function. This study showed that our approach can accurately detect the location of pneumothorax in chest X-rays, significantly reduce the time delay incurred when diagnosing urgent diseases such as pneumothorax, and increase the effectiveness of clinical practice and patient care.

Evaluating the Performance of Extreme Learning Machine Technique for Ore Grade Estimation

Due to the complex geology of vein deposits and their erratic grade distributions, there is the tendency of overestimating or underestimating the ore grade. These estimated grade results determine the profitability of mining the ore deposit or otherwise. In this study, five Extreme Learning Machine (ELM) variants based on hard limit, sigmoid, triangular basis, sine and radial basis activation functions were applied to predict ore grade. The motive is that the activation function has been identified to play a key role in achieving optimum ELM performance. Therefore, assessing the extent of influence the activation functions will have on the final outputs from the ELM has some scientific value worth investigating. This study therefore applied ELM as ore grade estimator which is yet to be explored in the literature. The obtained results from the five ELM variants were analysed and compared with the state-of-the-art benchmark methods of Backpropagation Neural Network (BPNN) and Ordinary Kriging (OK). The statistical test results revealed that the ELM with sigmoid activation function (ELM-Sigmoid) was the best among all the other investigated methods (ELM-Hard limit, ELM-Triangular basis, ELM-Sine, ELM-Radial Basis, BPNN and OK). This is because the ELM-sigmoid produced the lowest MAE (0.0175), MSE (0.0005) and RMSE (0.0229) with highest R2 (91.93%) and R (95.88%) respectively. It was concluded that ELM-Sigmoid can be used by field practitioners as a reliable alternative ore grade estimation technique.

Some special families of holomorphic and Sălăgean type bi-univalent functions associated with Horadam polynomials involving a modified sigmoid activation function

The aim of this paper is to introduce some special families of holomorphic and S\u{a}l\u{a}gean type bi-univalent functions by making use of Horadam polynomials involving the modified sigmoid activation function $\phi(s)=\frac{2}{1+e^{-s} },\,s\geq0$ in the open unit disc $\mathfrak{D}$. We investigate the upper bounds on initial coefficients for functions of the form $g_{\phi}(z)=z+\sum\limits_{j=2}^{\infty}\phi(s)d_jz^j$, in these newly introduced special families and also discuss the Fekete-Szegö problem. Some interesting consequences of the results established here are also indicated.

Deep learning of total electron content

One of the most notable errors in the global navigation satellite system (GNSS) is the ionospheric delay due to the total electron content (TEC). TEC is the number of electrons in the ionosphere in the signal path from the satellite to the receiver, which fluctuates with time and location. This error is one of the major problems in single-frequency (SF) GPS receivers. One way to eliminate this error is to use dual-frequency. Users of SF receivers should either use estimation models or local models to reduce this error. In this study, deep learning of artificial neural networks (ANN) was used to estimate TEC for SF users. For this purpose, the ionosphere as a single-layer model (assuming that all free electrons in the ionosphere are in this thin layer) is locally modeled by the code observation method. Linear combination has been used by selecting 24 permanent GNSS stations in the northwest of Iran. TEC was modeled independently of the geometry between the satellite and the receiver, called L4. This modeling was used to train the error ANN with two 5-day periods of high and low solar and geomagnetic activity range with a hyperbolic tangential sigmoid activation function. The results show that the proposed method is capable of eliminating ionosphere error with an average accuracy of 90%. The international reference ionosphere 2016 (IRI2016) is used for the verification, which has a 96% significance correlation with estimated TEC.

Mineral bioflotation optimization: Comparison between artificial neural networks and response surface methodology

The present work studied the fundamental of polynomial modeling and artificial neural network (ANN) techniques applied to the bioflotation of apatite, calcite and dolomite using the Rhodococcus opacus biosurfactant. Both techniques were used to model the bioflotation process and optimize some bioflotation parameters (pH and biosurfactant concentration). A full quadratic model optimized with genetic algorithm (GA) and a three-layer feedforward neural network were used to describe the mineral recovery. Different training algorithms and activation functions were tested to gather to an ANN with the best generalization capacity. Although the variation of the sigmoid activation function did not lead to a considerable change in the robustness of the ANN, the use of standard numerical training algorithms produced ANN models with better accuracy. ANN models demonstrated higher adequacy to describe and predict the mineral recovery than polynomial models. For all mineral studied, mineral recovery increased for a more acidic medium and for a higher biosurfactant concentration (BC). BC exhibited a higher effect on mineral recovery than pH, showing pH an increasing effect on mineral recovery as the optimal BC values were approached. Rhodococcus opacus biosurfactant demonstrated to be a promising solution for the separation of carbonate gangue minerals from phosphate ores.

Design of Improved Version of Sigmoidal Function with Biases for Classification Task in ELM Domain

Extreme Learning Machine (ELM) is one of the latest trends in learning algorithm, which can provide a good recognition rate within less computation time. Therefore, the algorithm can sustain for a faster response application by utilizing a feed-forward neural network. In this research article, the ELM method has been designed with the presence of sigmoidal function of biases in the hidden nodes to perform the classification task. The classification task is very challenging with the existing learning algorithm and increased computation time due to the huge amount of dataset. While handling of the stochastic matrix for hidden layer, output provides the lower performance for learning rate and robustness in the determination. To address these issues, the modified version of ELM has been developed to obtain better accuracy and minimize the classification error. This research article includes the mathematical proof of sigmoidal activation function with biases of the hidden nodes present in the networks. The output matrix maintains the column rank in order to improve the speed of the training output weights (β). The proposed improved version of ELM leverages better accuracy and efficacy in classification and regression problems as well. Due to the inclusion of matrix column ranking in mathematical proof, the proposed improved version of ELM solves the slow training speed and over-fitting problems present in the existing learning approach.

Tracking System Using Artificial Neural Network for FPGA Cart Follower

Wheelchair users may have difficulty carrying their luggage while travelling. A proposed solution is introduced based on the problem stated. A tracking system for cart followers introduced to improve wheelchair mobility while transporting their luggage. The cart will track and follow the wheelchair at an appropriate distance. A Camera acts as input to enable the cart to track and follow the wheelchair. Pixy CMUcam5 is used to detect the predefined colour code (CC) in this project. The visual-based sensor gathered the information of the CC, which situated behind the wheelchair and translated the collected data into relative position information, such as distance and skew angle, which helps the cart in following the wheelchair. This translation is done in the Artificial Neural Network (ANN). The type of ANN used is a Multilayer perceptron (MLP) with the Levenberg-Marquardt training algorithm. Log sigmoid (Logsig) and Pure linear (Purelin) activation functions used in the hidden and output layer, respectively. The errors for distances and angles after computed with ANN are presented accordingly in Simulink. The MSE value obtained is 0.14007. The ANN implemented in the Field Programmable Gate Array (FPGA). The implementation of the ANN on the FPGA done through software and hardware configuration. The time taken for hardware implementation is faster than software due to parallel computations. The error value for output distance is less than 0.8000, and 0.3000 for the output angle during the simulation.

Seizure Detection Based on Adaptive Feature Extraction by Applying Extreme Learning Machines

Epilepsy is one of the most common chronic disorder which negatively affects the patients' life. The functionality of the brain can be obtained from brain signals and it is vital to analyze and examine the brain signals in seizure detection processes. In this study, we performed machine learning-based and signal processing methods to detect epileptic signals. To do that, we examined three different EEG signals (healthy, ictal, and interictal) with two different classes (healthy ones and epileptic ones). Our proposed method consists of three stages which are preprocessing, feature extraction, and classification. In the preprocessing phase, EEG signals normalized to scale all samples into [0,1] range. After Stockwell Transform was applied and chaotic features and Parseval's Energy collected from each EEG signal. In the last part, EEG signals were classified with ELM (Extreme Learning Machines) with different parameters. Our study shows the best classification accuracy obtained from the Sigmoid activation function with the number of 100 hidden neurons. The highlights of this study are: Stockwell Transform is used; Entropy values are selected based on the adaptive process. Threshold values are determined according to the error rates; ELM classifier algorithm is applied.

Supervised learning in the presence of concept drift: a modelling framework

We present a modelling framework for the investigation of supervised learning in non-stationary environments. Specifically, we model two example types of learning systems: prototype-based learning vector quantization (LVQ) for classification and shallow, layered neural networks for regression tasks. We investigate so-called student–teacher scenarios in which the systems are trained from a stream of high-dimensional, labeled data. Properties of the target task are considered to be non-stationary due to drift processes while the training is performed. Different types of concept drift are studied, which affect the density of example inputs only, the target rule itself, or both. By applying methods from statistical physics, we develop a modelling framework for the mathematical analysis of the training dynamics in non-stationary environments. Our results show that standard LVQ algorithms are already suitable for the training in non-stationary environments to a certain extent. However, the application of weight decay as an explicit mechanism of forgetting does not improve the performance under the considered drift processes. Furthermore, we investigate gradient-based training of layered neural networks with sigmoidal activation functions and compare with the use of rectified linear units. Our findings show that the sensitivity to concept drift and the effectiveness of weight decay differs significantly between the two types of activation function.

Development of Integrated Neural Network Model for Identification of Fake Reviews in E-Commerce Using Multidomain Datasets.

Online product reviews play a major role in the success or failure of an E-commerce business. Before procuring products or services, the shoppers usually go through the online reviews posted by previous customers to get recommendations of the details of products and make purchasing decisions. Nevertheless, it is possible to enhance or hamper specific E-business products by posting fake reviews, which can be written by persons called fraudsters. These reviews can cause financial loss to E-commerce businesses and misguide consumers to take the wrong decision to search for alternative products. Thus, developing a fake review detection system is ultimately required for E-commerce business. The proposed methodology has used four standard fake review datasets of multidomains include hotels, restaurants, Yelp, and Amazon. Further, preprocessing methods such as stopword removal, punctuation removal, and tokenization have performed as well as padding sequence method for making the input sequence has fixed length during training, validation, and testing the model. As this methodology uses different sizes of datasets, various input word-embedding matrices of n-gram features of the review's text are developed and created with help of word-embedding layer that is one component of the proposed model. Convolutional and max-pooling layers of the CNN technique are implemented for dimensionality reduction and feature extraction, respectively. Based on gate mechanisms, the LSTM layer is combined with the CNN technique for learning and handling the contextual information of n-gram features of the review's text. Finally, a sigmoid activation function as the last layer of the proposed model receives the input sequences from the previous layer and performs binary classification task of review text into fake or truthful. In this paper, the proposed CNN-LSTM model was evaluated in two types of experiments, in-domain and cross-domain experiments. For an in-domain experiment, the model is applied on each dataset individually, while in the case of a cross-domain experiment, all datasets are gathered and put into a single data frame and evaluated entirely. The testing results of the model in-domain experiment datasets were 77%, 85%, 86%, and 87% in the terms of accuracy for restaurant, hotel, Yelp, and Amazon datasets, respectively. Concerning the cross-domain experiment, the proposed model has attained 89% accuracy. Furthermore, comparative analysis of the results of in-domain experiments with existing approaches has been done based on accuracy metric and, it is observed that the proposed model outperformed the compared methods.

Network security situation prediction method based on strengthened LSTM neural network

As an important part of network security situation awareness, network security situation prediction describes the dynamic changes of security situation over time, and predicts future situation values based on historical situation values. In order to improve the accuracy of network security situation prediction, a long- and short-term memory network security situation prediction model based on the Sigmoid weighted reinforcement mechanism is proposed. Firstly, LSTM neural network is used to mine the temporal correlation of network security situation data. Sigmoid weighted linear element is introduced to deal with the gradient problem in the back propagation, and the input value is multiplied by Sigmoid activation function, so as to strengthen the structure of LSTM neural network and improve the accuracy of prediction.Then, the cuckoo search algorithm was used to optimize the super parameters to improve the training time. Finally, the public data set CICIDS2017 was used to verify the model. The simulation experiment results show that the model has a faster convergence rate and smaller errors.

Extreme Learning Machine based Differentiation of Pulmonary Tuberculosis in Chest Radiographs using Integrated Local Feature Descriptors

Background and ObjectiveComputer aided diagnostics of Pulmonary Tuberculosis in chest radiographs relies on the differentiation of subtle and non-specific alterations in the images. In this study, an attempt has been made to identify and classify Tuberculosis conditions from healthy subjects in chest radiographs using integrated local feature descriptors and variants of extreme learning machine. MethodsLung fields in the chest images are segmented using Reaction Diffusion Level Set method. Local feature descriptors such as Median Robust Extended Local Binary Patterns and Gradient Local Ternary Patterns are extracted. Extreme Learning Machine (ELM) and Online Sequential ELM (OSELM) classifiers are employed to identify Tuberculosis conditions and, their performances are analysed using standard metrics. ResultsResults show that the adopted segmentation method is able to delineate lung fields in both healthy and Tuberculosis images. Extracted features are statistically significant even in images with inter and intra subject variability. Sigmoid activation function yields accuracy and sensitivity values greater than 98% for both the classifiers. Highest sensitivity is observed with OSELM for minimal significant features in detecting Tuberculosis images. ConclusionAs ELM based method is able to differentiate the subtle changes in inter and intra subject variations of chest X-ray images, the proposed methodology seems to be useful for computer-based detection of Pulmonary Tuberculosis.

A Robust Automated Machine Learning System with Pseudoinverse Learning

Developing a robust deep neural network (DNN) for a specific task is not only time-consuming but also requires lots of experienced human experts. In order to make deep neural networks easier to apply or even take the human experts out of the design of network architecture completely, a growing number of researches focus on robust automated machine learning (AutoML). In this paper, we investigated the robustness problem of AutoML systems based on contractive pseudoinverse learners. In our proposed method, deep neural networks were built with stacked contractive pseudoinverse learners (CPILer). Each CPILer has a Jacobian regularized reconstruction loss function and is trained with pseudoinverse learning algorithm. When sigmoid activation function is adopted in the hidden layer, the graph Laplace regularizer is derived from square Frobenius norm of the Jacobian matrix. This learning scheme not only speeds up the training process dramatically but also reduces the effort of hyperparameter tuning. In addition, the graph Laplace regularization can improve the robustness of the learning systems by reducing the sensibility to noise. An ensemble network architecture consisting of several sub-networks was designed to build the AutoML systems. The architecture hyperparameters of the system were determined in an automated way which could be considered as a data-driven way. The proposed method shown good performance in the experiments in terms of efficiency and accuracy, and outperformed the baseline methods on a series of benchmark data sets. The robustness improvement of our proposed method was also demonstrated in the experiments.

Hardware implementation of radial-basis neural networks with Gaussian activation functions on FPGA

This article introduces a method for realizing the Gaussian activation function of radial-basis (RBF) neural networks with their hardware implementation on field-programmable gaits area (FPGAs). The results of modeling of the Gaussian function on FPGA chips of different families have been presented. RBF neural networks of various topologies have been synthesized and investigated. The hardware component implemented by this algorithm is an RBF neural network with four neurons of the latent layer and one neuron with a sigmoid activation function on an FPGA using 16-bit numbers with a fixed point, which took 1193 logic matrix gate (LUTs—LookUpTable). Each hidden layer neuron of the RBF network is designed on an FPGA as a separate computing unit. The speed as a total delay of the combination scheme of the block RBF network was 101.579 ns. The implementation of the Gaussian activation functions of the hidden layer of the RBF network occupies 106 LUTs, and the speed of the Gaussian activation functions is 29.33 ns. The absolute error is ± 0.005. The Spartan 3 family of chips for modeling has been used to get these results. Modeling on chips of other series has been also introduced in the article. RBF neural networks of various topologies have been synthesized and investigated. Hardware implementation of RBF neural networks with such speed allows them to be used in real-time control systems for high-speed objects.