INTSOM: Gray Image Compression Using an Intelligent Self-Organizing Map

Technological innovation talent is the core element of economic and social development.It is a symbol of a country's soft power reflected from the perspective of knowledge.Therefore it is theoretically valuable and practically significant for a country to evaluate the comprehensive level technological innovation.Based on the construction of evaluation index system of the technological innovation talents' competitiveness, we will use BP neural network to evaluate in this paper.We determine the actual situation of the evaluation system according to the input layer, hidden layer and output layer of the model as well as the number of neurons in each layer.The number of neurons in the input layer is the number of indices of technological innovation talents evaluation.The output layer represents the comprehensive ability level and has only one neuron.However, it is difficult to figure out the number of hidden layers of BP neural networks.We use the golden section method to accurately select their layers.This can effectively avoid the problem of excessive training error due to the random selection of hidden layer.Then we illustrate the accuracy and usefulness of this algorithm with the case of the technological innovation talents of the six central provinces.Tests show that the comprehensive ability of technological innovation talents of Hubei is the most powerful in the six central provinces and is followed by Anhui.The remaining four provinces have similar levels.The actual development of technological innovation talents agree well with the experimental results.

This paper investigates the error of determining the coordinates of an acoustic signal source depending on various parameters of the developed software-mathematical models for training a neural network. The authors determined the optimal values of these parameters according to the criterion of minimum error. The study found that the optimal amount of data for training a neural network is 25,000 training pairs, and the number of hidden layers should not exceed 5. Among several forms of sensor arrangement, the best is the square, as it provides a smaller error and uniform filling of the microphone’s perimeter. Their number should not exceed 9, since the best algorithm for training the neural network trainlm requires significant computing resources of the system on which training is performed. An increase in the number of sensors leads to an increase in the training time of the neural network but does not lead to an improvement in accuracy. When studying the effect of the distance to the acoustic signal source and the algorithm for selecting the base sensor for calculating the time difference, it was found that changing these parameters does not affect the accuracy of the neural network. The best structure of the number of hidden neurons in each hidden layer is an expanding structure, in which the number of neurons in each layer increases from the input layer to the middle-hidden layer, then decreases to the output layer. Using the optimized parameters, the neural network was trained and its accuracy was evaluated on two data sets: those used for training and those unknown to the neural network. In both cases, the neural network demonstrated good results. According to the results of the study, it was found that the error of the optimised system is 6.7681 × 10-5 m in the Y coordinate and 6.6120 × 10-5 m in the X coordinate.

The article discusses the possibilities of using artificial intelligence systems to solve clustering problems. The value of the optimality criterion for various combinations of the number of clusters and the number of neurons of the output network layer is determined. Self-organizing maps (SOM, Self Organizing Maps), developed by T. Kohonen and representing a powerful tool combining two important paradigms of data analysis - clustering and projecting, visualization of multidimensional data on a plane are considered. An example of the location of cluster nuclei after training the Kohonen neural network for different values of the number of neurons in the source layer is given. Comparing the speed of modern computers with the speed of the Kohonen neural network, with other types of neural networks, allows you to conduct a large number of network exercises in a short time, so you can use one of many methods to determine the maximum value of the function. The results of experimental studies to determine the criterion of optimality are presented in the article for different combinations of the number of clusters and the number of neurons in the original layer of the network. According to the method at the initial stage, a set of input vectors is formed, each of which includes three values. A general sequence of actions is formulated to calculate the optimal number of neurons in the output layer of the Kohonen network. The methodology presented in the article is a further development of teaching methods without a teacher. The technique proposed in the article avoids the need to specify the number of outputs of the Kohonen neural network and can be widely used both in solving data mining problems and in recognizing new unknown classes and situations in different fields.

A new hierarchal structure of the self-organizing map (SOM) with dynamic growth is presented and applied to codebook design in vector quantization (VQ) and image compression. The tree-structured approach for codebook design is motivated for reducing the high computational efforts in the training and image coding phases in traditional VQ algorithms. The DHSOM has the ability to self determine the structure of the network through heuristically rules, and its final structure reflects the variability of the data (image blocks). It is shown that training that training and coding times obtained with DHSOM algorithm are faster than conventional SOM and LBG algorithms, while the qualitative results are given.

The effectiveness of the binary artificial neural network at low volume of training samples was investigated. The studies were conducted with the MATLAB software’s Neural Network Toolbox module used for the classification task. Feed-forward artificial neural network was thoroughly studied. It was discovered, that the number of neurons in the inner layer in the range of changes had little effect on efficiency. The volume of training samples reduces efficiency when the value is very small. The influence the of the input training samples dimension on the efficiency was studied; the dimension was reduced from 64 to 16 by eliminating a different set of input parameters. It is established that such changes have a different impact on efficiency, and under certain conditions, the efficiency increases. For a cascade binary network of feed forward propagation and a cascade binary network with reverse propagation of error, it was established that the efficiency changes slightly depending on the number of neurons in the hidden layer in the studied range. A two-layer cascade network with reverse error propagation with a certain combination of neurons has a higher efficiency compared to a single-layer one. The results allow us to outline ways to improve efficiency.

Artificial neural networks (ANNs) especially Backpropagation Neural Network (BPNN) was used largely in image processing. The backpropagation neural network algorithm (BP) was used for training the BPNN for image compression/decompression. The BP requires long time to train the BPNN with small error. Therefore, in this research, a three layered BPNN was designed for building image compression system. The Fast backpropagation neural network algorithm (FBP) was used for training the designed BPNN to reduce the training time (convergence time) of BPNN as possible as. Many techniques were used to improve the use of FBP for BPNN training. This is done by using different architecture of BPNN by changing the number of input layer neurons and number of hidden layer neurons. Also we trained the BPNN with different FBP parameters. Finally, FBP results such as compression ratio (CR) and peak signal to noise ratio (PSNR) are computed and compared with BP results. From the results, we noticed that the use of FBP improve the BPNN training by reducing the convergence time of image compression learning process.

An important benchmark for closed-loop flow control is the suppression of the von Karman vortex street in the wake of a circular cylinder at a Reynolds number of 100. A low-dimensional Proper Orthogonal Decomposition (POD) is applied to the flow field and a four sensor configuration is placed based on the intensity of the resulting spatial Eigenfunctions. This effort focuses on the performance of a non-linear Artificial Neural Network Estimator (ANNE) used for real-time mapping of flow field sensors to POD states. This investigation examines the sensitivity of ANNE by varying the number of neurons in the hidden layer. The aim is to determine if there is an optimum number of neurons in the hidden layer. Three input data sets were studied, the first had no noise on the training and testing data, the second had 10% noise applied to the training and testing data, and the third had 25% noise applied to the training and testing data sets. In all cases 50 randomly selected training data sets were applied to the back-propagating network. A separate data set was used for testing the feed-forward network. During the validation process, network outputs were compared to desired outputs and the Root Mean Squared (RMS) error was calculated for each of the four output nodes. Results show that in general, the more nodes in the hidden layer the better the performance of the network. However, there appeared to be a consistently reduced error when eight (8) nodes were used in the hidden layer. Results from ANNE when compared to the state-of-the-art Linear Stochastic Estimator (LSE) shows significant benefits.

In the adult barrel cortex of the rat the calcium-binding proteins calbindin D28k (CALB) and parvalbumin (PARV) are found in separate populations of GABAergic nonpyramidal neurons. In layers II to IV of the barrel cortex most PARV-immunoreactive neurons are likely to derive from a subpopulation of CALB-immunoreactive neurons whose CALB immunoreactivity ceases when they begin to express PARV between the second and third postnatal weeks. The aim of this study was to investigate the influence of subcortical afferents on the neurochemical differentiation of cortical PARV- and CALB-immunoreactive nonpyramidal neurons during development of the barrel cortex. We produced unilateral excitotoxic lesions with a single injection of ibotenic acid (0.5 microl, 0.05 M) in different subcortical nuclei in 7- to 8-day-old rats. Lesions involving the ventroposterior thalamic nuclei resulted in delayed development of PARV and CALB immunoreactivity in the barrel cortex. One week after ibotenic acid injections a transient decrease in the number of PARV-immunoreactive neurons in layer IV was observed, together with increased numbers of CALB-immunoreactive neurons in all cortical layers. The number of nonpyramidal neurons displaying coexistence of PARV and CALB in the lesioned hemisphere also increased compared with the numbers in the control hemisphere or control littermates. In contrast, lesions affecting the globus pallidus, zona incerta and reticular thalamic nucleus transiently increased the number of PARV-immunoreactive neurons in layers II and III, but had no effect on the number of CALB-positive cells. From 3 weeks onwards no differences were found between control and lesioned hemispheres after injections into either the ventroposterior thalamic nuclei or the magnocellular basal forebrain. These results suggest that CALB and PARV expression in nonpyramidal cortical neurons can be reversibly modulated in opposite directions by different cortical afferents during postnatal development.

Bayesian regularization of artificial neural networks (BRANNs) were used to predict body mass index (BMI) in mice using single nucleotide polymorphism (SNP) markers. Data from 1896 animals with both phenotypic and genotypic (12 320 loci) information were used for the analysis. Missing genotypes were imputed based on estimated allelic frequencies, with no attempt to reconstruct haplotypes based on family information or linkage disequilibrium between markers. A feed-forward multilayer perceptron network consisting of a single output layer and one hidden layer was used. Training of the neural network was done using the Bayesian regularized backpropagation algorithm. When the number of neurons in the hidden layer was increased, the number of effective parameters, γ, increased up to a point and stabilized thereafter. A model with five neurons in the hidden layer produced a value of γ that saturated the data. In terms of predictive ability, a network with five neurons in the hidden layer attained the smallest error and highest correlation in the test data although differences among networks were negligible. Using inherent weight information of BRANN with different number of neurons in the hidden layer, it was observed that 17 SNPs had a larger impact on the network, indicating their possible relevance in prediction of BMI. It is concluded that BRANN may be at least as useful as other methods for high-dimensional genome-enabled prediction, with the advantage of its potential ability of capturing non-linear relationships, which may be useful in the study of quantitative traits under complex gene action.

In order to ensure the safe and stable operation of power system, enrich the means of power grid analysis and control, and expand the application of deep intelligent learning methods in power grid systems, the application of deep learning intelligent machine learning method in frequency prediction of large power grid is explored. First, on the basis of deep learning, the frequency response mode of large power grid is analyzed and the key characteristic quantities that affect the frequency response mode are extracted. Second, the deep belief neural network (DBN. DNN) frequency prediction model is constructed. Also, the training and testing of the model are introduced. Finally, the input and output based on the DBN.CNN prediction model and the network structure design of the model are analyzed. The prediction performance of the model is evaluated. The results show that when the number of neurons in the hidden layer is 50, the model achieves the optimal prediction effect. Increasing the number of training samples helps to improve the modeling ability and prediction accuracy of the model. For frequency prediction problems, the number of training samples should be set to ≥400, and the number of hidden layers corresponding to the model should be 5. When the number of hidden layer neurons is 10, the prediction accuracy of the DBN/DNN network is poor. When the number of hidden layer neurons is 50, the model can achieve the best prediction effect. Overall, the DBN.DNN prediction model has good prediction performance. The RMSE of the forecast data is O. 0073Hz can basically meet the actual application requirements. Therefore, the frequency prediction method based on deep belief neural network has certain advantages in accuracy and efficiency.

In the present paper, we have employed the application of artificial neural networks (ANN) to predict effective heat storage coefficient (HSC) of building materials. First we prepared a database to train and test the models developed here. Two types of architectures from different networks are developed, one with three inputs and the other with four inputs mixed architecture combining an ANN with a theoretical model developed by us previously. These ANN models are built, trained and tested by the feed forward back propagation algorithm, to obtain the effective properties of building materials from the properties of their constituents. Feed forward back propagation neural network structure has been developed, which includes an input layer, a hidden layer and an output layer. The number of neurons in the input layer is equal to the number of input parameters and the number of neurons in the output layer is equal to the output parameters. A good agreement has been found between the predicted values using ANN and the experimental results reported in the literature.

Mangrove plants are one of the plants that really support the improvement of the world's ecology nowadays. There are several studies and researches which use mangrove’s data which spanning from image data to morphological data. In several previous studies, we hardly find researches about mangrove morphological data which resulted in classification or identification. Therefore, it is necessary to try out a methodology that is lighter and easy to use than the previous mangrove morphological classification. This study aims to obtain a classification of three classes of mangrove sprouts, such as Avicennia Marina, Sonneratia Caseolaris and Ceriops Tagal, using the Multi-Layer Perceptron (MLP). We have collected 3000 mangrove sprouts data, which consist of 2100 training data and 900 testing data. The results of this study show an accuracy value of 98.1% for the training data, and 91.9% testing data accuracy value using a double hidden layer and the number of neurons in each hidden layer is more than the number of neurons in the input layer resulting in an optimal accuracy value at the 1000th epoch.

In this work the artificial neural networks robust design methodology was used to find the optimal parameters in backpropagation artificial neural network architecture applied to the solution of the inverse kinematics of a six degrees of freedom robotic manipulator. A systematic and experimental strategy is proposed according to the requirements of the problem. Among the various parameters that affect the performance of a backpropagation neural network, four design variables were selected since they are the variables that can be controlled by the user: the number of neurons in the first hidden layer, the number of neurons in the second hidden layer, the momentum, and the learning rate. The results show that the robust design methodology can be used to find a better configuration with high performance and generalization capability; significantly reducing the time spent determining the optimal architecture of the neural network compared to the trial and error approach, which requires long periods of time.

Mobile Ad Hoc network is a self-organizing infrastructure less network where the nodes are responsible for routing the packets. MANET has dynamic topology which allows nodes to join and leave the network at any time. This paper proposes a channel adaptive routing protocol which extends AOMDV routing protocol to accommodate channel fading. The proposed Channel Aware AOMDV protocol uses channel average nonfading duration to select stable links for route discovery. The main objective of this paper is to demonstrate approximation capability of artificial neural network. Feed forward network architecture with back propagation algorithm is used. Depending upon the size of database number of hidden layers and number of neurons in hidden layer are decided. Number of neurons in hidden layer is varied and effect on percentage deviation in the output parameter is observed. Depending upon nature of input and output parameters synthesis and analysis artificial neural network (ANN) models are trained.

Multilayer artificial neural networks (ANN) are often used for the solution of classification problems or for the time series forecasting. An appropriate number of learning and testing patterns must be available for the ANN training. Each training pattern is composed of n input parameters and m output parameters. Number m is usually given by the problem formulation, but the number n may be often selected from a greater set of input parameters. An optimal selection of input parameters is a very important task especially in a situation when the number of usable input parameters is great and the analytical relations between the input and output parameters are not known. The number of neurons in all ANN layers must be generally kept as small as possible because of the optimal generalisation ability. In this paper we present a possible way for the selection of significant input parameters (the so called markers), which are the most important ones from the point of view of influence on the output parameters. These parameters are later used for the training of ANN. A statistical approach is usually used for this reason [5]. After some experience in the ANN application we recognised that the approach based on mathematical logic, i. e. the GUHA method (General Unary Hypotheses Automaton) is also suitable for the determination of markers. Besides the minimisation of the number of elements in the input layer of ANN, also the number of neurons in hidden layers must be optimised. For this reason standard methods of pruning can be used, described e. g. in [1]. We have used this method in the following applications: - Optimisation of the intervals between the major overhaul of plane engines by the analysis of tribodiagnostic data. Only selected types of chemical pollution in oil can be taken into account. - Prediction of bleeding of patients with chronic lymphoblastic leukemia. Only a part of parameters about the patient is important from this point of view (see [2]). - Optimisation of the quality and reliability prediction of artificial resin production in chemical factory. Only a part of the production parameters (times of production phases, temperatures, percentage of components etc.) have straight influence on the product. - Optimisation of the prosody control in the text-to-speech synthesis. This application is described in the paper.