In the field of artificial neural network, BP neural network is a multi-layer feed-forward neural network. Because it is difficult to figure the number of hidden layer nodes in a BP neural network, the theoretical basis and the existing methods for BP network hidden layer nodes are studied. Then based on traditional empirical formulas, we propose a new approach to rapidly figure the quantity of hidden layer nodes in two-layer network. That is, with the assistance of experience formulas, the horizon of unit number in hidden layer can be confirmed and its optimal value will be found in this horizon. Finally, a new formula for figuring the quantity of hidden layer codes is obtained through fitting input dimension, output dimension and the optimal value of hidden layer codes. Under some given input dimension and output dimension, efficiency and precision of BP algorithm may be improved by applying the proposed formula.
Optimized Parameter of Wavelet Neural Network (WNN) using INGArahulmonikasharma
Genetic algorithm has been one of the most popular methods for many challenging optimization problems. It is a critical problem in which the evacuation time is an important issues. The continuous air traffic growth and limits of resources, there is a need for reducing the congestion of the airspace system. The main objective of this work is to automatically adapt the airspace configurations, according to the evolution of traffic Niche genetic algorithm(INGA) was used in reliability optimization of software system. And also the searching performance of the genetic algorithm was improved by the stochastic tournament model. The multi-module complex software system reliability allocation effectively. Genetic algorithm (GA) and FGA are compared though seven benchmark function. It can be applied to a wider range of problem including multi-level problem. The uniform schema crossover operator and the non-uniform mutation in the genetic algorithm.
On the High Dimentional Information Processing in Quaternionic Domain and its...IJAAS Team
There are various high dimensional engineering and scientific applications in communication, control, robotics, computer vision, biometrics, etc.; where researchers are facing problem to design an intelligent and robust neural system which can process higher dimensional information efficiently. The conventional real-valued neural networks are tried to solve the problem associated with high dimensional parameters, but the required network structure possesses high complexity and are very time consuming and weak to noise. These networks are also not able to learn magnitude and phase values simultaneously in space. The quaternion is the number, which possesses the magnitude in all four directions and phase information is embedded within it. This paper presents a well generalized learning machine with a quaternionic domain neural network that can finely process magnitude and phase information of high dimension data without any hassle. The learning and generalization capability of the proposed learning machine is presented through a wide spectrum of simulations which demonstrate the significance of the work.
Machine Learning Algorithms for Image Classification of Hand Digits and Face ...IRJET Journal
This document discusses machine learning algorithms for image classification using five different classification schemes. It summarizes the mathematical models behind each classification algorithm, including Nearest Class Centroid classifier, Nearest Sub-Class Centroid classifier, k-Nearest Neighbor classifier, Perceptron trained using Backpropagation, and Perceptron trained using Mean Squared Error. It also describes two datasets used in the experiments - the MNIST dataset of handwritten digits and the ORL face recognition dataset. The performance of the five classification schemes are compared on these datasets.
This document summarizes a research paper that uses an artificial neural network approach to forecast stock market prices in India. The paper trains a feedforward neural network using a backpropagation algorithm on data from 5 Indian companies between 2004 and 2013. The network is tested in MATLAB to predict stock prices and calculate an error rate for accuracy. The neural network model is found to provide a computational method for predicting stock market movements based on historical price and volume data.
This document describes a study using artificial neural networks (ANNs) to model complex nonlinear systems. Specifically, it discusses:
1) Using an ANN to predict pressure distributions on a rotor wing during ramping motion, with results showing accurate prediction of spatial and temporal evolution.
2) Applying the same ANN model to predict performance of a bank stock based on trends in the stock and stock market index.
3) Proposing a framework combining ANNs with mathematical models to obtain better predictions and representations of financial data trends.
This document compares the k-means and grid density clustering algorithms. K-means partitions data into k clusters based on minimizing distances between points and cluster centroids. It works well with numerical data but can be affected by outliers. Grid density determines dense grids based on neighbor densities and can handle different shaped and multi-density clusters without knowing the number of clusters beforehand. It has advantages over k-means in that it can handle categorical data, noise and arbitrary shaped clusters.
International Journal of Engineering and Science Invention (IJESI)inventionjournals
This document discusses multidimensional clustering methods for data mining and their industrial applications. It begins with an introduction to clustering, including definitions and goals. Popular clustering algorithms are described, such as K-means, fuzzy C-means, hierarchical clustering, and mixture of Gaussians. Distance measures and their importance in clustering are covered. The K-means and fuzzy C-means algorithms are explained in detail. Examples are provided to illustrate fuzzy C-means clustering. Finally, applications of clustering algorithms in fields such as marketing, biology, and earth sciences are mentioned.
Application of Hybrid Genetic Algorithm Using Artificial Neural Network in Da...IOSRjournaljce
The main purpose of data mining is to extract knowledge from large amount of data. Artificial Neural network (ANN) has already been applied in a variety of domains with remarkable success. This paper presents the application of hybrid model for stroke disease that integrates Genetic algorithm and back propagation algorithm. Selecting a good subset of features, without sacrificing accuracy, is of great importance for neural networks to be successfully applied to the area. In addition the hybrid model that leads to further improvised categorization, accuracy compared to the result produced by genetic algorithm alone. In this study, a new hybrid model of Neural Networks and Genetic Algorithm (GA) to initialize and optimize the connection weights of ANN so as to improve the performance of the ANN and the same has been applied in a medical problem of predicting stroke disease for verification of the results.
Optimized Parameter of Wavelet Neural Network (WNN) using INGArahulmonikasharma
Genetic algorithm has been one of the most popular methods for many challenging optimization problems. It is a critical problem in which the evacuation time is an important issues. The continuous air traffic growth and limits of resources, there is a need for reducing the congestion of the airspace system. The main objective of this work is to automatically adapt the airspace configurations, according to the evolution of traffic Niche genetic algorithm(INGA) was used in reliability optimization of software system. And also the searching performance of the genetic algorithm was improved by the stochastic tournament model. The multi-module complex software system reliability allocation effectively. Genetic algorithm (GA) and FGA are compared though seven benchmark function. It can be applied to a wider range of problem including multi-level problem. The uniform schema crossover operator and the non-uniform mutation in the genetic algorithm.
On the High Dimentional Information Processing in Quaternionic Domain and its...IJAAS Team
There are various high dimensional engineering and scientific applications in communication, control, robotics, computer vision, biometrics, etc.; where researchers are facing problem to design an intelligent and robust neural system which can process higher dimensional information efficiently. The conventional real-valued neural networks are tried to solve the problem associated with high dimensional parameters, but the required network structure possesses high complexity and are very time consuming and weak to noise. These networks are also not able to learn magnitude and phase values simultaneously in space. The quaternion is the number, which possesses the magnitude in all four directions and phase information is embedded within it. This paper presents a well generalized learning machine with a quaternionic domain neural network that can finely process magnitude and phase information of high dimension data without any hassle. The learning and generalization capability of the proposed learning machine is presented through a wide spectrum of simulations which demonstrate the significance of the work.
Machine Learning Algorithms for Image Classification of Hand Digits and Face ...IRJET Journal
This document discusses machine learning algorithms for image classification using five different classification schemes. It summarizes the mathematical models behind each classification algorithm, including Nearest Class Centroid classifier, Nearest Sub-Class Centroid classifier, k-Nearest Neighbor classifier, Perceptron trained using Backpropagation, and Perceptron trained using Mean Squared Error. It also describes two datasets used in the experiments - the MNIST dataset of handwritten digits and the ORL face recognition dataset. The performance of the five classification schemes are compared on these datasets.
This document summarizes a research paper that uses an artificial neural network approach to forecast stock market prices in India. The paper trains a feedforward neural network using a backpropagation algorithm on data from 5 Indian companies between 2004 and 2013. The network is tested in MATLAB to predict stock prices and calculate an error rate for accuracy. The neural network model is found to provide a computational method for predicting stock market movements based on historical price and volume data.
This document describes a study using artificial neural networks (ANNs) to model complex nonlinear systems. Specifically, it discusses:
1) Using an ANN to predict pressure distributions on a rotor wing during ramping motion, with results showing accurate prediction of spatial and temporal evolution.
2) Applying the same ANN model to predict performance of a bank stock based on trends in the stock and stock market index.
3) Proposing a framework combining ANNs with mathematical models to obtain better predictions and representations of financial data trends.
This document compares the k-means and grid density clustering algorithms. K-means partitions data into k clusters based on minimizing distances between points and cluster centroids. It works well with numerical data but can be affected by outliers. Grid density determines dense grids based on neighbor densities and can handle different shaped and multi-density clusters without knowing the number of clusters beforehand. It has advantages over k-means in that it can handle categorical data, noise and arbitrary shaped clusters.
International Journal of Engineering and Science Invention (IJESI)inventionjournals
This document discusses multidimensional clustering methods for data mining and their industrial applications. It begins with an introduction to clustering, including definitions and goals. Popular clustering algorithms are described, such as K-means, fuzzy C-means, hierarchical clustering, and mixture of Gaussians. Distance measures and their importance in clustering are covered. The K-means and fuzzy C-means algorithms are explained in detail. Examples are provided to illustrate fuzzy C-means clustering. Finally, applications of clustering algorithms in fields such as marketing, biology, and earth sciences are mentioned.
Application of Hybrid Genetic Algorithm Using Artificial Neural Network in Da...IOSRjournaljce
The main purpose of data mining is to extract knowledge from large amount of data. Artificial Neural network (ANN) has already been applied in a variety of domains with remarkable success. This paper presents the application of hybrid model for stroke disease that integrates Genetic algorithm and back propagation algorithm. Selecting a good subset of features, without sacrificing accuracy, is of great importance for neural networks to be successfully applied to the area. In addition the hybrid model that leads to further improvised categorization, accuracy compared to the result produced by genetic algorithm alone. In this study, a new hybrid model of Neural Networks and Genetic Algorithm (GA) to initialize and optimize the connection weights of ANN so as to improve the performance of the ANN and the same has been applied in a medical problem of predicting stroke disease for verification of the results.
Comparison Between Clustering Algorithms for Microarray Data AnalysisIOSR Journals
Currently, there are two techniques used for large-scale gene-expression profiling; microarray and
RNA-Sequence (RNA-Seq).This paper is intended to study and compare different clustering algorithms that used
in microarray data analysis. Microarray is a DNA molecules array which allows multiple hybridization
experiments to be carried out simultaneously and trace expression levels of thousands of genes. It is a highthroughput
technology for gene expression analysis and becomes an effective tool for biomedical research.
Microarray analysis aims to interpret the data produced from experiments on DNA, RNA, and protein
microarrays, which enable researchers to investigate the expression state of a large number of genes. Data
clustering represents the first and main process in microarray data analysis. The k-means, fuzzy c-mean, selforganizing
map, and hierarchical clustering algorithms are under investigation in this paper. These algorithms
are compared based on their clustering model.
International Journal of Engineering Research and DevelopmentIJERD Editor
Electrical, Electronics and Computer Engineering,
Information Engineering and Technology,
Mechanical, Industrial and Manufacturing Engineering,
Automation and Mechatronics Engineering,
Material and Chemical Engineering,
Civil and Architecture Engineering,
Biotechnology and Bio Engineering,
Environmental Engineering,
Petroleum and Mining Engineering,
Marine and Agriculture engineering,
Aerospace Engineering.
X-TREPAN : A Multi Class Regression and Adapted Extraction of Comprehensible ...csandit
The document describes an algorithm called X-TREPAN that extracts decision trees from trained neural networks. X-TREPAN is an enhancement of the TREPAN algorithm that allows it to handle both multi-class classification and multi-class regression problems. It can also analyze generalized feed forward networks. The algorithm was tested on several real-world datasets and was found to generate decision trees with good classification accuracy while also maintaining comprehensibility.
The document proposes a novel Spatial Fuzzy C-Means (PET-SFCM) clustering algorithm to segment PET scan images of patients with neurodegenerative disorders like Alzheimer's disease. The algorithm incorporates spatial neighborhood information into the traditional Fuzzy C-Means algorithm. It was tested on real patient data sets and showed satisfactory results compared to conventional FCM and K-Means clustering algorithms. The PET-SFCM algorithm provides an effective way to segment PET images and analyze brain changes related to neurological conditions.
System for Prediction of Non Stationary Time Series based on the Wavelet Radi...IJECEIAES
This document proposes and examines the performance of a hybrid time series forecasting model called the wavelet radial bases function neural network (WRBFNN) model. The WRBFNN model uses wavelet transforms to extract coefficients from time series data as inputs to a radial basis function neural network for forecasting. The performance of the WRBFNN model is compared to a wavelet feed forward neural network (WFFNN) model using four types of non-stationary time series data. Results show the WRBFNN model achieves more accurate forecasts than the WFFNN model for certain types of non-stationary data, and trains significantly faster with fewer computational resources required.
Survey on classification algorithms for data mining (comparison and evaluation)Alexander Decker
This document provides an overview and comparison of three classification algorithms: K-Nearest Neighbors (KNN), Decision Trees, and Bayesian Networks. It discusses each algorithm, including how KNN classifies data based on its k nearest neighbors. Decision Trees classify data based on a tree structure of decisions, and Bayesian Networks classify data based on probabilities of relationships between variables. The document conducts an analysis of these three algorithms to determine which has the best performance and lowest time complexity for classification tasks based on evaluating a mock dataset over 24 months.
This document discusses using an artificial neural network to forecast stock price indices in a stock exchange. It begins with an abstract that notes ANNs have been used successfully for non-linear business forecasting. The paper then aims to present a better prediction model for stock indices using neural network techniques in the Indian context. It reviews single and multilayer networks, and the backpropagation method for training multilayer networks.
This document discusses clustering of uncertain data objects. It first provides background on clustering uncertain data and challenges involved. It then reviews various existing approaches for clustering uncertain data, including using soft classifiers and probabilistic databases. The document proposes combining k-means clustering with Voronoi diagrams and indexing techniques to improve the performance and efficiency of clustering uncertain datasets. It outlines a plan to integrate k-means with Voronoi diagrams and indexing to reduce execution time and increase clustering performance and results for uncertain data. Finally, it concludes that combining clustering with indexing approaches can better handle uncertain data clustering challenges.
K-MEDOIDS CLUSTERING USING PARTITIONING AROUND MEDOIDS FOR PERFORMING FACE R...ijscmc
Face recognition is one of the most unobtrusive biometric techniques that can be used for access control as well as surveillance purposes. Various methods for implementing face recognition have been proposed with varying degrees of performance in different scenarios. The most common issue with effective facial biometric systems is high susceptibility of variations in the face owing to different factors like changes in pose, varying illumination, different expression, presence of outliers, noise etc. This paper explores a novel technique for face recognition by performing classification of the face images using unsupervised learning approach through K-Medoids clustering. Partitioning Around Medoids algorithm (PAM) has been used for performing K-Medoids clustering of the data. The results are suggestive of increased robustness to noise and outliers in comparison to other clustering methods. Therefore the technique can also be used to increase the overall robustness of a face recognition system and thereby increase its invariance and make it a reliably usable biometric modality
Spatial correlation based clustering algorithm for random and uniform topolog...eSAT Publishing House
IJRET : International Journal of Research in Engineering and Technology is an international peer reviewed, online journal published by eSAT Publishing House for the enhancement of research in various disciplines of Engineering and Technology. The aim and scope of the journal is to provide an academic medium and an important reference for the advancement and dissemination of research results that support high-level learning, teaching and research in the fields of Engineering and Technology. We bring together Scientists, Academician, Field Engineers, Scholars and Students of related fields of Engineering and Technology
Architecture neural network deep optimizing based on self organizing feature ...journalBEEI
Forward neural network (FNN) execution relying on the algorithm of training and architecture selection. Different parameters using for nip out the architecture of FNN such as the connections number among strata, neurons hidden number in each strata hidden and hidden strata number. Feature architectural combinations exponential could be uncontrollable manually so specific architecture can be design automatically by using special algorithm which build system with ability generalization better. Determination of architecture FNN can be done by using the algorithm of optimization numerous. In this paper methodology new proposes achievement where FNN neurons respective with hidden layers estimation work where in this work collect algorithm training self organizing feature map (SOFM) with advantages to explain how the best architectural selected automatically by SOFM from criteria error testing based on architecture populated. Different size of dataset benchmark of 4 classifications tested for approach proposed.
A COMPREHENSIVE ANALYSIS OF QUANTUM CLUSTERING : FINDING ALL THE POTENTIAL MI...IJDKP
Quantum clustering (QC), is a data clustering algorithm based on quantum mechanics which is
accomplished by substituting each point in a given dataset with a Gaussian. The width of the Gaussian is a
σ value, a hyper-parameter which can be manually defined and manipulated to suit the application.
Numerical methods are used to find all the minima of the quantum potential as they correspond to cluster
centers. Herein, we investigate the mathematical task of expressing and finding all the roots of the
exponential polynomial corresponding to the minima of a two-dimensional quantum potential. This is an
outstanding task because normally such expressions are impossible to solve analytically. However, we
prove that if the points are all included in a square region of size σ, there is only one minimum. This bound
is not only useful in the number of solutions to look for, by numerical means, it allows to to propose a new
numerical approach “per block”. This technique decreases the number of particles by approximating some
groups of particles to weighted particles. These findings are not only useful to the quantum clustering
problem but also for the exponential polynomials encountered in quantum chemistry, Solid-state Physics
and other applications.
Welcome to International Journal of Engineering Research and Development (IJERD)IJERD Editor
This document discusses techniques for handling missing data values in datasets. It compares the K-means clustering and k-nearest neighbor (kNN) classifier approaches for imputing missing values. The K-means method replaces missing values in clusters with the cluster mean, while kNN replaces missing values in groups with the group mean. When these approaches are tested on a dataset with missing values and compared for accuracy, the kNN method is shown to perform better. The document proposes a framework where the two techniques are applied separately to a dataset with missing values, the results compared, and kNN imputation is found to be more accurate than K-means clustering imputation.
This document provides an overview of self-organizing maps (SOM) as an unsupervised learning technique. It discusses the principles of self-organization including self-amplification, competition, and cooperation. The Willshaw-von der Malsburg model and Kohonen feature maps are presented as two approaches to building topographic maps through self-organization. The Kohonen SOM learning algorithm is described as involving competition between neurons to determine a winning neuron, cooperation between neighboring neurons, and adaptive changes to synaptic weights based on Hebbian learning principles.
International Journal of Engineering Research and Applications (IJERA) is an open access online peer reviewed international journal that publishes research and review articles in the fields of Computer Science, Neural Networks, Electrical Engineering, Software Engineering, Information Technology, Mechanical Engineering, Chemical Engineering, Plastic Engineering, Food Technology, Textile Engineering, Nano Technology & science, Power Electronics, Electronics & Communication Engineering, Computational mathematics, Image processing, Civil Engineering, Structural Engineering, Environmental Engineering, VLSI Testing & Low Power VLSI Design etc.
There are very few examples of the use of various architectures for recurrent neural
networks to predict student learning outcomes. In fact, the only architecture used to
solve this problem is the LSTM architecture. In the works devoted to the use of LSTM
to predict educational outcomes, the results of a detailed theoretical substantiation of
the preference of this particular architecture of the RNN are not presented. In this
regard, it seems advisable to provide such justification in the framework of this study.
The main property of input data for prediction of educational outcomes is its
temporary nature. Some sequence of user actions unfolds in time and is evaluated
(classified) by an external observer as evidence of the presence or absence of an
educational result (objective or metaobjective). In this regard, the RNN used to classify
user actions should perform a procedure for adjusting the weights of neurons for a
certain set of states in the past. At the same time, the length of the sequence of these
states is not predetermined: it can be both short (for example, for objective results),
and quite long.
Optimized Parameter of Wavelet Neural Network (WNN) using INGArahulmonikasharma
Genetic algorithm has been one of the most popular methods for many challenging optimization problems. It is a critical problem in which the evacuation time is an important issues. The continuous air traffic growth and limits of resources, there is a need for reducing the congestion of the airspace system. The main objective of this work is to automatically adapt the airspace configurations, according to the evolution of traffic Niche genetic algorithm(INGA) was used in reliability optimization of software system. And also the searching performance of the genetic algorithm was improved by the stochastic tournament model. The multi-module complex software system reliability allocation effectively. Genetic algorithm (GA) and FGA are compared though seven benchmark function. It can be applied to a wider range of problem including multi-level problem. The uniform schema crossover operator and the non-uniform mutation in the genetic algorithm.
Image classification is perhaps the most important part of digital image analysis. In this paper, we compare the most widely used model CNN Convolutional Neural Network , and MLP Multilayer Perceptron . We aim to show how both models differ and how both models approach towards the final goal, which is image classification. Souvik Banerjee | Dr. A Rengarajan "Hand-Written Digit Classification" Published in International Journal of Trend in Scientific Research and Development (ijtsrd), ISSN: 2456-6470, Volume-5 | Issue-4 , June 2021, URL: https://www.ijtsrd.compapers/ijtsrd42444.pdf Paper URL: https://www.ijtsrd.comcomputer-science/artificial-intelligence/42444/handwritten-digit-classification/souvik-banerjee
Neural Network Based Individual Classification SystemIRJET Journal
This document describes a neural network model developed for individual classification. The model was designed to measure personality traits through a questionnaire. It then uses a neural network trained on sample data through unsupervised and supervised learning with multi-layer perceptions. The backpropagation algorithm was used to train the network. The neural network architecture included multiple neuron layers trained on a 200 item data set, achieving 99.82% accuracy. The goal was to classify individuals into high, middle, and low personality categories for use in job selection or training.
RunPool: A Dynamic Pooling Layer for Convolution Neural NetworkPutra Wanda
Deep learning (DL) has achieved a significant performance in computer vision problems, mainly in automatic feature extraction and representation. However, it is not easy to determine the best pooling method in a different case study. For instance, experts can implement the best types of pooling in image processing cases, which might not be optimal for various tasks. Thus, it is
required to keep in line with the philosophy of DL. In dynamic neural network architecture, it is not practically possible to find
a proper pooling technique for the layers. It is the primary reason why various pooling cannot be applied in the dynamic and multidimensional dataset. To deal with the limitations, it needs to construct an optimal pooling method as a better option than max pooling and average pooling. Therefore, we introduce a dynamic pooling layer called RunPool to train the convolutional
neuralnetwork(CNN)architecture.RunPoolpoolingisproposedtoregularizetheneuralnetworkthatreplacesthedeterministic
pooling functions. In the final section, we test the proposed pooling layer to address classification problems with online social network (OSN) dataset
CONTRAST OF RESNET AND DENSENET BASED ON THE RECOGNITION OF SIMPLE FRUIT DATA...rinzindorjej
In this paper, a fruit image data set is used to compare the efficiency and accuracy of two widely used
Convolutional Neural Network, namely the ResNet and the DenseNet, for the recognition of 50 different
kinds of fruits. In the experiment, the structure of ResNet-34 and DenseNet_BC-121 (with bottleneck layer)
are used. The mathematic principle, experiment detail and the experiment result will be explained through
comparison.
In this paper, a fruit image data set is used to compare the efficiency and accuracy of two widely used Convolutional Neural Network, namely the ResNet and the DenseNet, for the recognition of 50 different kinds of fruits. In the experiment, the structure of ResNet-34 and DenseNet_BC-121 (with bottleneck layer) are used. The mathematic principle, experiment detail and the experiment result will be explained through comparison.
Comparison Between Clustering Algorithms for Microarray Data AnalysisIOSR Journals
Currently, there are two techniques used for large-scale gene-expression profiling; microarray and
RNA-Sequence (RNA-Seq).This paper is intended to study and compare different clustering algorithms that used
in microarray data analysis. Microarray is a DNA molecules array which allows multiple hybridization
experiments to be carried out simultaneously and trace expression levels of thousands of genes. It is a highthroughput
technology for gene expression analysis and becomes an effective tool for biomedical research.
Microarray analysis aims to interpret the data produced from experiments on DNA, RNA, and protein
microarrays, which enable researchers to investigate the expression state of a large number of genes. Data
clustering represents the first and main process in microarray data analysis. The k-means, fuzzy c-mean, selforganizing
map, and hierarchical clustering algorithms are under investigation in this paper. These algorithms
are compared based on their clustering model.
International Journal of Engineering Research and DevelopmentIJERD Editor
Electrical, Electronics and Computer Engineering,
Information Engineering and Technology,
Mechanical, Industrial and Manufacturing Engineering,
Automation and Mechatronics Engineering,
Material and Chemical Engineering,
Civil and Architecture Engineering,
Biotechnology and Bio Engineering,
Environmental Engineering,
Petroleum and Mining Engineering,
Marine and Agriculture engineering,
Aerospace Engineering.
X-TREPAN : A Multi Class Regression and Adapted Extraction of Comprehensible ...csandit
The document describes an algorithm called X-TREPAN that extracts decision trees from trained neural networks. X-TREPAN is an enhancement of the TREPAN algorithm that allows it to handle both multi-class classification and multi-class regression problems. It can also analyze generalized feed forward networks. The algorithm was tested on several real-world datasets and was found to generate decision trees with good classification accuracy while also maintaining comprehensibility.
The document proposes a novel Spatial Fuzzy C-Means (PET-SFCM) clustering algorithm to segment PET scan images of patients with neurodegenerative disorders like Alzheimer's disease. The algorithm incorporates spatial neighborhood information into the traditional Fuzzy C-Means algorithm. It was tested on real patient data sets and showed satisfactory results compared to conventional FCM and K-Means clustering algorithms. The PET-SFCM algorithm provides an effective way to segment PET images and analyze brain changes related to neurological conditions.
System for Prediction of Non Stationary Time Series based on the Wavelet Radi...IJECEIAES
This document proposes and examines the performance of a hybrid time series forecasting model called the wavelet radial bases function neural network (WRBFNN) model. The WRBFNN model uses wavelet transforms to extract coefficients from time series data as inputs to a radial basis function neural network for forecasting. The performance of the WRBFNN model is compared to a wavelet feed forward neural network (WFFNN) model using four types of non-stationary time series data. Results show the WRBFNN model achieves more accurate forecasts than the WFFNN model for certain types of non-stationary data, and trains significantly faster with fewer computational resources required.
Survey on classification algorithms for data mining (comparison and evaluation)Alexander Decker
This document provides an overview and comparison of three classification algorithms: K-Nearest Neighbors (KNN), Decision Trees, and Bayesian Networks. It discusses each algorithm, including how KNN classifies data based on its k nearest neighbors. Decision Trees classify data based on a tree structure of decisions, and Bayesian Networks classify data based on probabilities of relationships between variables. The document conducts an analysis of these three algorithms to determine which has the best performance and lowest time complexity for classification tasks based on evaluating a mock dataset over 24 months.
This document discusses using an artificial neural network to forecast stock price indices in a stock exchange. It begins with an abstract that notes ANNs have been used successfully for non-linear business forecasting. The paper then aims to present a better prediction model for stock indices using neural network techniques in the Indian context. It reviews single and multilayer networks, and the backpropagation method for training multilayer networks.
This document discusses clustering of uncertain data objects. It first provides background on clustering uncertain data and challenges involved. It then reviews various existing approaches for clustering uncertain data, including using soft classifiers and probabilistic databases. The document proposes combining k-means clustering with Voronoi diagrams and indexing techniques to improve the performance and efficiency of clustering uncertain datasets. It outlines a plan to integrate k-means with Voronoi diagrams and indexing to reduce execution time and increase clustering performance and results for uncertain data. Finally, it concludes that combining clustering with indexing approaches can better handle uncertain data clustering challenges.
K-MEDOIDS CLUSTERING USING PARTITIONING AROUND MEDOIDS FOR PERFORMING FACE R...ijscmc
Face recognition is one of the most unobtrusive biometric techniques that can be used for access control as well as surveillance purposes. Various methods for implementing face recognition have been proposed with varying degrees of performance in different scenarios. The most common issue with effective facial biometric systems is high susceptibility of variations in the face owing to different factors like changes in pose, varying illumination, different expression, presence of outliers, noise etc. This paper explores a novel technique for face recognition by performing classification of the face images using unsupervised learning approach through K-Medoids clustering. Partitioning Around Medoids algorithm (PAM) has been used for performing K-Medoids clustering of the data. The results are suggestive of increased robustness to noise and outliers in comparison to other clustering methods. Therefore the technique can also be used to increase the overall robustness of a face recognition system and thereby increase its invariance and make it a reliably usable biometric modality
Spatial correlation based clustering algorithm for random and uniform topolog...eSAT Publishing House
IJRET : International Journal of Research in Engineering and Technology is an international peer reviewed, online journal published by eSAT Publishing House for the enhancement of research in various disciplines of Engineering and Technology. The aim and scope of the journal is to provide an academic medium and an important reference for the advancement and dissemination of research results that support high-level learning, teaching and research in the fields of Engineering and Technology. We bring together Scientists, Academician, Field Engineers, Scholars and Students of related fields of Engineering and Technology
Architecture neural network deep optimizing based on self organizing feature ...journalBEEI
Forward neural network (FNN) execution relying on the algorithm of training and architecture selection. Different parameters using for nip out the architecture of FNN such as the connections number among strata, neurons hidden number in each strata hidden and hidden strata number. Feature architectural combinations exponential could be uncontrollable manually so specific architecture can be design automatically by using special algorithm which build system with ability generalization better. Determination of architecture FNN can be done by using the algorithm of optimization numerous. In this paper methodology new proposes achievement where FNN neurons respective with hidden layers estimation work where in this work collect algorithm training self organizing feature map (SOFM) with advantages to explain how the best architectural selected automatically by SOFM from criteria error testing based on architecture populated. Different size of dataset benchmark of 4 classifications tested for approach proposed.
A COMPREHENSIVE ANALYSIS OF QUANTUM CLUSTERING : FINDING ALL THE POTENTIAL MI...IJDKP
Quantum clustering (QC), is a data clustering algorithm based on quantum mechanics which is
accomplished by substituting each point in a given dataset with a Gaussian. The width of the Gaussian is a
σ value, a hyper-parameter which can be manually defined and manipulated to suit the application.
Numerical methods are used to find all the minima of the quantum potential as they correspond to cluster
centers. Herein, we investigate the mathematical task of expressing and finding all the roots of the
exponential polynomial corresponding to the minima of a two-dimensional quantum potential. This is an
outstanding task because normally such expressions are impossible to solve analytically. However, we
prove that if the points are all included in a square region of size σ, there is only one minimum. This bound
is not only useful in the number of solutions to look for, by numerical means, it allows to to propose a new
numerical approach “per block”. This technique decreases the number of particles by approximating some
groups of particles to weighted particles. These findings are not only useful to the quantum clustering
problem but also for the exponential polynomials encountered in quantum chemistry, Solid-state Physics
and other applications.
Welcome to International Journal of Engineering Research and Development (IJERD)IJERD Editor
This document discusses techniques for handling missing data values in datasets. It compares the K-means clustering and k-nearest neighbor (kNN) classifier approaches for imputing missing values. The K-means method replaces missing values in clusters with the cluster mean, while kNN replaces missing values in groups with the group mean. When these approaches are tested on a dataset with missing values and compared for accuracy, the kNN method is shown to perform better. The document proposes a framework where the two techniques are applied separately to a dataset with missing values, the results compared, and kNN imputation is found to be more accurate than K-means clustering imputation.
This document provides an overview of self-organizing maps (SOM) as an unsupervised learning technique. It discusses the principles of self-organization including self-amplification, competition, and cooperation. The Willshaw-von der Malsburg model and Kohonen feature maps are presented as two approaches to building topographic maps through self-organization. The Kohonen SOM learning algorithm is described as involving competition between neurons to determine a winning neuron, cooperation between neighboring neurons, and adaptive changes to synaptic weights based on Hebbian learning principles.
International Journal of Engineering Research and Applications (IJERA) is an open access online peer reviewed international journal that publishes research and review articles in the fields of Computer Science, Neural Networks, Electrical Engineering, Software Engineering, Information Technology, Mechanical Engineering, Chemical Engineering, Plastic Engineering, Food Technology, Textile Engineering, Nano Technology & science, Power Electronics, Electronics & Communication Engineering, Computational mathematics, Image processing, Civil Engineering, Structural Engineering, Environmental Engineering, VLSI Testing & Low Power VLSI Design etc.
There are very few examples of the use of various architectures for recurrent neural
networks to predict student learning outcomes. In fact, the only architecture used to
solve this problem is the LSTM architecture. In the works devoted to the use of LSTM
to predict educational outcomes, the results of a detailed theoretical substantiation of
the preference of this particular architecture of the RNN are not presented. In this
regard, it seems advisable to provide such justification in the framework of this study.
The main property of input data for prediction of educational outcomes is its
temporary nature. Some sequence of user actions unfolds in time and is evaluated
(classified) by an external observer as evidence of the presence or absence of an
educational result (objective or metaobjective). In this regard, the RNN used to classify
user actions should perform a procedure for adjusting the weights of neurons for a
certain set of states in the past. At the same time, the length of the sequence of these
states is not predetermined: it can be both short (for example, for objective results),
and quite long.
Optimized Parameter of Wavelet Neural Network (WNN) using INGArahulmonikasharma
Genetic algorithm has been one of the most popular methods for many challenging optimization problems. It is a critical problem in which the evacuation time is an important issues. The continuous air traffic growth and limits of resources, there is a need for reducing the congestion of the airspace system. The main objective of this work is to automatically adapt the airspace configurations, according to the evolution of traffic Niche genetic algorithm(INGA) was used in reliability optimization of software system. And also the searching performance of the genetic algorithm was improved by the stochastic tournament model. The multi-module complex software system reliability allocation effectively. Genetic algorithm (GA) and FGA are compared though seven benchmark function. It can be applied to a wider range of problem including multi-level problem. The uniform schema crossover operator and the non-uniform mutation in the genetic algorithm.
Image classification is perhaps the most important part of digital image analysis. In this paper, we compare the most widely used model CNN Convolutional Neural Network , and MLP Multilayer Perceptron . We aim to show how both models differ and how both models approach towards the final goal, which is image classification. Souvik Banerjee | Dr. A Rengarajan "Hand-Written Digit Classification" Published in International Journal of Trend in Scientific Research and Development (ijtsrd), ISSN: 2456-6470, Volume-5 | Issue-4 , June 2021, URL: https://www.ijtsrd.compapers/ijtsrd42444.pdf Paper URL: https://www.ijtsrd.comcomputer-science/artificial-intelligence/42444/handwritten-digit-classification/souvik-banerjee
Neural Network Based Individual Classification SystemIRJET Journal
This document describes a neural network model developed for individual classification. The model was designed to measure personality traits through a questionnaire. It then uses a neural network trained on sample data through unsupervised and supervised learning with multi-layer perceptions. The backpropagation algorithm was used to train the network. The neural network architecture included multiple neuron layers trained on a 200 item data set, achieving 99.82% accuracy. The goal was to classify individuals into high, middle, and low personality categories for use in job selection or training.
RunPool: A Dynamic Pooling Layer for Convolution Neural NetworkPutra Wanda
Deep learning (DL) has achieved a significant performance in computer vision problems, mainly in automatic feature extraction and representation. However, it is not easy to determine the best pooling method in a different case study. For instance, experts can implement the best types of pooling in image processing cases, which might not be optimal for various tasks. Thus, it is
required to keep in line with the philosophy of DL. In dynamic neural network architecture, it is not practically possible to find
a proper pooling technique for the layers. It is the primary reason why various pooling cannot be applied in the dynamic and multidimensional dataset. To deal with the limitations, it needs to construct an optimal pooling method as a better option than max pooling and average pooling. Therefore, we introduce a dynamic pooling layer called RunPool to train the convolutional
neuralnetwork(CNN)architecture.RunPoolpoolingisproposedtoregularizetheneuralnetworkthatreplacesthedeterministic
pooling functions. In the final section, we test the proposed pooling layer to address classification problems with online social network (OSN) dataset
CONTRAST OF RESNET AND DENSENET BASED ON THE RECOGNITION OF SIMPLE FRUIT DATA...rinzindorjej
In this paper, a fruit image data set is used to compare the efficiency and accuracy of two widely used
Convolutional Neural Network, namely the ResNet and the DenseNet, for the recognition of 50 different
kinds of fruits. In the experiment, the structure of ResNet-34 and DenseNet_BC-121 (with bottleneck layer)
are used. The mathematic principle, experiment detail and the experiment result will be explained through
comparison.
In this paper, a fruit image data set is used to compare the efficiency and accuracy of two widely used Convolutional Neural Network, namely the ResNet and the DenseNet, for the recognition of 50 different kinds of fruits. In the experiment, the structure of ResNet-34 and DenseNet_BC-121 (with bottleneck layer) are used. The mathematic principle, experiment detail and the experiment result will be explained through comparison.
CONTRAST OF RESNET AND DENSENET BASED ON THE RECOGNITION OF SIMPLE FRUIT DATA...rinzindorjej
In this paper, a fruit image data set is used to compare the efficiency and accuracy of two widely used Convolutional Neural Network, namely the ResNet and the DenseNet, for the recognition of 50 different kinds of fruits. In the experiment, the structure of ResNet-34 and DenseNet_BC-121 (with bottleneck layer)
are used. The mathematic principle, experiment detail and the experiment result will be explained through comparison.
IJRET : International Journal of Research in Engineering and Technology is an international peer reviewed, online journal published by eSAT Publishing House for the enhancement of research in various disciplines of Engineering and Technology. The aim and scope of the journal is to provide an academic medium and an important reference for the advancement and dissemination of research results that support high-level learning, teaching and research in the fields of Engineering and Technology. We bring together Scientists, Academician, Field Engineers, Scholars and Students of related fields of Engineering and Technology
Design of c slotted microstrip antenna using artificial neural network modeleSAT Journals
Abstract In this paper, neural network model has been used to estimation of resonance frequency of a coaxial feed C-slotted Microstrip Antenna. The Multi-Layer Perceptron Feed forward back Propagation (MLPFFBP) and Radial basis function Artificial Neural Network (RBFANN) have been used to implement the neural network model. A relative performance analysis of the proposed neural network for different training algorithms. Number of neurons and number of hidden layer is also carried out for estimating the resonance frequency. The method of moment (MOM) based IE3D software was used to generate data dictionary for training and validation set of ANN. The results obtain using ANN are compared with simulation feeding and found quite satisfactory and also it is concluded that RBFANN network is more accurate and fast compared to MLPFFBP network algorithm. Index Terms: Artificial Neural Network, C slot, Microstrip Antenna, Multilayer Feed Forward Networks, Radial basis function Artificial Neural Network, Resonance frequency.
AN IMPROVED METHOD FOR IDENTIFYING WELL-TEST INTERPRETATION MODEL BASED ON AG...IAEME Publication
This paper presents an approach based on applying an aggregated predictor formed by multiple versions of a multilayer neural network with a back-propagation optimization algorithm for helping the engineer to get a list of the most appropriate well-test interpretation models for a given set of pressure/ production data. The proposed method consists of three stages: (1) data decorrelation through principal component analysis to reduce the covariance between the variables and the dimension of the input layer in the artificial neural network, (2) bootstrap replicates of the learning set where the data is repeatedly sampled with a random split of the data into train sets and using these as new learning sets, and (3) automatic reservoir model identification through aggregated predictor formed by a plurality vote when predicting a new class. This method is described in detail to ensure successful replication of results. The required training and test dataset were generated by using analytical solution models. In our case, there were used 600 samples: 300 for training, 100 for cross-validation, and 200 for testing. Different network structures were tested during this study to arrive at optimum network design. We notice that the single net methodology always brings about confusion in selecting the correct model even though the training results for the constructed networks are close to 1. We notice also that the principal component analysis is an effective strategy in reducing the number of input features, simplifying the network structure, and lowering the training time of the ANN. The results obtained show that the proposed model provides better performance when predicting new data with a coefficient of correlation approximately equal to 95% Compared to a previous approach 80%, the combination of the PCA and ANN is more stable and determine the more accurate results with lesser computational complexity than was feasible previously. Clearly, the aggregated predictor is more stable and shows less bad classes compared to the previous approach.
Web spam classification using supervised artificial neural network algorithmsaciijournal
Due to the rapid growth in technology employed by the spammers, there is a need of classifiers that are more efficient, generic and highly adaptive. Neural Network based technologies have high ability of adaption as well as generalization. As per our knowledge, very little work has been done in this field using neural network. We present this paper to fill this gap. This paper evaluates performance of three supervised learning algorithms of artificial neural network by creating classifiers for the complex problem of latest web spam pattern classification. These algorithms are Conjugate Gradient algorithm, Resilient Backpropagation learning, and Levenberg-Marquardt algorithm.
Handwritten Digit Recognition using Convolutional Neural NetworksIRJET Journal
This document discusses using a convolutional neural network called LeNet to perform handwritten digit recognition on the MNIST dataset. It begins with an abstract that outlines using LeNet, a type of convolutional network, to accurately classify handwritten digits from 0 to 9. It then provides background on convolutional networks and how they can extract and utilize features from images to classify patterns with translation and scaling invariance. The document implements LeNet using the Keras deep learning library in Python to classify images from the MNIST dataset, which contains labeled images of handwritten digits. It analyzes the architecture of LeNet and how convolutional and pooling layers are used to extract features that are passed to fully connected layers for classification.
IRJET-Breast Cancer Detection using Convolution Neural NetworkIRJET Journal
This document discusses using a convolutional neural network (CNN) to detect breast cancer from medical images. CNNs are a type of deep learning model that can learn image features without manual feature engineering. The proposed system would take a sample medical image as input, preprocess it, and compare it to images in a database labeled as cancerous or non-cancerous. If cancer is detected, the system would determine the cancer stage and recommend appropriate treatment. The CNN model would be built and trained using libraries like Keras, TensorFlow, and Numpy to classify images and detect breast cancer at early stages for better treatment outcomes.
Optimal neural network models for wind speed predictionIAEME Publication
The document describes using artificial neural networks for wind speed prediction. Specifically, it analyzes the performance of multilayer perceptron networks and radial basis function networks for wind speed forecasting using real-time data collected from wind farms in Coimbatore, India over one year. The models are trained on 3000 samples and tested on 1000 samples. Performance is evaluated using statistical metrics like coefficient of determination, mean absolute error, root mean square error, and mean bias error. Results show that the neural network models improve prediction accuracy compared to other approaches and the optimal model depends on factors like the number of hidden neurons and spread value.
Optimal neural network models for wind speed predictionIAEME Publication
The document presents two neural network models - multilayer perceptron (MLP) and radial basis function (RBF) networks - for wind speed prediction. It evaluates these models on real wind speed data collected over one year from wind farms in Coimbatore, India. The experimental results show that the RBF and MLP models can improve wind speed prediction accuracy compared to other approaches, according to statistical metrics like coefficient of determination, mean absolute error, root mean square error, and mean bias error.
Optimal neural network models for wind speed predictionIAEME Publication
The document presents two neural network models - multilayer perceptron (MLP) and radial basis function (RBF) networks - for wind speed prediction. It evaluates these models on real-time wind speed data collected from wind farms in Coimbatore, India over one year. The experimental results show that RBF and MLP networks can improve wind speed prediction accuracy compared to other approaches, according to statistical metrics like coefficient of determination, mean absolute error, root mean square error and mean bias error. The RBF and MLP models are able to handle the non-linear patterns in wind speed data, which conventional models struggle with, increasing prediction precision.
Survey on Artificial Neural Network Learning Technique AlgorithmsIRJET Journal
This document discusses different types of learning algorithms used in artificial neural networks. It begins with an introduction to neural networks and their ability to learn from their environment through adjustments to synaptic weights. Four main learning algorithms are then described: error correction learning, which uses algorithms like backpropagation to minimize error; memory based learning, which stores all training examples and analyzes nearby examples to classify new inputs; Hebbian learning, where connection weights are adjusted based on the activity of neurons; and competitive learning, where neurons compete to respond to inputs to become specialized feature detectors through a winner-take-all mechanism. The document provides details on how each type of learning algorithm works.
This document discusses characterizing polymeric membranes under large deformations using an artificial neural network model. It presents an experimental study of blowing circular thermoplastic ABS membranes using free blowing technique. A multilayer neural network is used to model the non-linear behavior of the membrane under biaxial deformation. The neural network results are compared to experimental data and a finite difference model using a hyperelastic Mooney-Rivlin model. The neural network accurately reproduces the membrane behavior with minimal error margins compared to experimental measurements.
PADDY CROP DISEASE DETECTION USING SVM AND CNN ALGORITHMIRJET Journal
- The document discusses a study on detecting diseases in paddy/rice crops using deep learning algorithms like convolutional neural networks (CNN) and support vector machines (SVM).
- A dataset of rice leaf images was created and a CNN model using transfer learning with MobileNet was developed and trained on the dataset to classify rice diseases.
- The proposed method aims to automatically classify rice disease images to help farmers more accurately identify diseases, as manual identification can be difficult and inaccurate. This could help improve treatment and support farmers.
Web Spam Classification Using Supervised Artificial Neural Network Algorithmsaciijournal
Due to the rapid growth in technology employed by the spammers, there is a need of classifiers that are
more efficient, generic and highly adaptive. Neural Network based technologies have high ability of
adaption as well as generalization. As per our knowledge, very little work has been done in this field using
neural network. We present this paper to fill this gap. This paper evaluates performance of three supervised
learning algorithms of artificial neural network by creating classifiers for the complex problem of latest
web spam pattern classification. These algorithms are Conjugate Gradient algorithm, Resilient Backpropagation learning, and Levenberg-Marquardt algorithm.
Similar to A New Method for Figuring the Number of Hidden Layer Nodes in BP Algorithm (20)
Data Mining is a significant field in today’s data-driven world. Understanding and implementing its concepts can lead to discovery of useful insights. This paper discusses the main concepts of data mining, focusing on two main concepts namely Association Rule Mining and Time Series Analysis
A Review on Real Time Integrated CCTV System Using Face Detection for Vehicle...rahulmonikasharma
We are describes the technique for real time human face detection and counting the number of passengers in vehicle and also gender of the passengers.The Image processing technology is very popular,now at present all are going to use it for various purpose. It can be applied to various applications for detecting and processing the digital images. Face detection is a part of image processing. It is used for finding the face of human in a given area. Face detection is used in many applications such as face recognition, people tracking, or photography. In this paper,The webcam is installed in public vehicle and connected with Raspberry Pi model. We use face detection technique for detecting and counting the number of passengers in public vehicle via webcam with the help of image processing and Raspberry Pi.
Considering Two Sides of One Review Using Stanford NLP Frameworkrahulmonikasharma
Sentiment analysis is a type of natural language processing for tracking the mood of the public about a particular product or a topic and is useful in several ways. Polarity shift is the most classical task which aims at classifying the reviews either positive or negative. But in many cases, in addition to the positive and negative reviews, there still many neutral reviews exist. However, the performance sometimes limited due to the fundamental deficiencies in handling the polarity shift problem. We propose an Improvised Dual Sentiment Analysis (IDSA) model to address this problem for sentiment classification. We first propose a novel data expansion technique by creating sentiment-reversed review for each training and test review. We develop a corpus based method to construct a pseudo-antonym dictionary. It removes DSA’s dependency on an external antonym dictionary for review reversion. We conduct a range of experiments and the results demonstrates the effectiveness of DSA in addressing the polarity shift in sentiment classification. .
A New Detection and Decoding Technique for (2×N_r ) MIMO Communication Systemsrahulmonikasharma
The requirements of fifth generation new radio (5G- NR) access networks are very high capacity and ultra-reliability. In this paper, we proposed a V-BLAST2 × N_r MIMO system that is analyzed, improved, and expected to achieve both very high throughput and ultra- reliability simultaneously.A new detection technique called parallel detection algorithm is proposed. The performance of the proposed algorithm compared with existing linear detection algorithms. It was seen that the proposed technique increases the speed of signal transmission and prevents error propagation which may be present in serial decoding techniques. The new algorithm reduces the bit error probability and increases the capacity simultaneouslywithout using a standard STC technique. However, it was seen that the BER of systems using the proposed algorithm is slightly higher than a similar system using only STC technique. Simulation results show the advantages of using the proposed technique.
Broadcasting Scenario under Different Protocols in MANET: A Surveyrahulmonikasharma
A wireless network enables people to communicate and access applications and information without wires. This provides freedom of movement and the ability to extend applications to different parts of a building, city, or nearly anywhere in the world. Wireless networks allow people to interact with e-mail or browse the Internet from a location that they prefer. Adhoc Networks are self-organizing wireless networks, absent any fixed infrastructure. broadcasting of data through proper channel is essential. Various protocols are designed to avoid the loss of data. In this paper an overview of different broadcast protocols are discussed.
Sybil Attack Analysis and Detection Techniques in MANETrahulmonikasharma
Security is important for many sensor network applications. A particularly harmful attack against sensor and ad hoc networks is known as the Sybil attack [6], where a node Illegitimately claims multiple identities.Mobility cause a main problem when we talk about security in Mobile Ad-hoc networks. It doesn’t depend on fixed architecture, the nodes are continuously moving in a random fashion. In this article we will focus on identifying the Sybil attack in MANET. It uses air medium for communication so it is more prone to the attack. Sybil attack is one in which single node present multiple fake identities to other nodes, which cause destruction.
A Landmark Based Shortest Path Detection by Using A* and Haversine Formularahulmonikasharma
In 1900, less than 20 percent of the world populace lived in cities, in 2007, fair more than 50 percent of the world populace lived in cities. In 2050, it has been anticipated that more than 70 percent of the worldwide population (about 6.4 billion individuals) will be city tenants. There's more weight being set on cities through this increment in population [1]. With approach of keen cities, data and communication technology is progressively transforming the way city regions and city inhabitants organize and work in reaction to urban development. In this paper, we create a nonspecific plot for navigating a route throughout city A asked route is given by utilizing combination of A* Algorithm and Haversine equation. Haversine Equation gives least distance between any two focuses on spherical body by utilizing latitude and longitude. This least distance is at that point given to A* calculation to calculate minimum distance. The method for identifying the shortest path is specify in this paper.
Processing Over Encrypted Query Data In Internet of Things (IoTs) : CryptDBs,...rahulmonikasharma
This document discusses techniques for processing queries over encrypted data in Internet of Things (IoT) systems. It describes CryptDB and MONOMI, which are database systems that can execute SQL queries over encrypted data. CryptDB uses a database proxy to encrypt/decrypt data and rewrite queries to execute on encrypted data. MONOMI builds on CryptDB and introduces a split client/server approach to query execution to improve efficiency of analytical queries over encrypted data. The document also outlines various encryption schemes that can be used for encrypted query processing, including deterministic encryption, order-preserving encryption, homomorphic encryption, and others.
Quality Determination and Grading of Tomatoes using Raspberry Pirahulmonikasharma
This document describes a system for determining the quality and grading tomatoes using image processing techniques on a Raspberry Pi. The system uses a USB camera to capture images of tomatoes and then performs preprocessing, masking, contour detection, image enhancement and color detection algorithms to analyze features like shape, size, color and texture. It can grade tomatoes into four categories: red, orange, green, and turning green. The system was able to accurately determine tomato quality and estimate expiry dates with 90% accuracy and had low computational time of 0.52 seconds compared to other machine learning methods.
Comparative of Delay Tolerant Network Routings and Scheduling using Max-Weigh...rahulmonikasharma
Network management and Routing is supportively done by performing with the nodes, due to infrastructure-less nature of the network in Ad hoc networks or MANET. The nodes are maintained itself from the functioning of the network, for that reason the MANET security challenges several defects. Routing process and Scheduling is a significant idea to enhance the security in MANET. Other than, scheduling has been recognized to be a key issue for implementing throughput/capacity optimization in Ad hoc networks. Designed underneath conventional (LT) light tailed assumptions, traffic fundamentally faces Heavy-tailed (HT) assumption of the validity of scheduling algorithms. Scheduling policies are utilized for communication networks such as Max-Weight, backpressure and ACO, which are provably throughput optimality and the Pareto frontier of the feasible throughput region under maximal throughput vector. In wireless ad-hoc network, the issue of routing and optimal scheduling performs with time varying channel reliability and multiple traffic streams. Depending upon the security issues within MANETs in this paper presents a comparative analysis of existing scheduling policies based on their performance to progress the delay performance in most scenarios. The security issues of MANETs considered from this paper presents a relative analysis of existing scheduling policies depend on their performance to progress the delay performance in most developments.
DC Conductivity Study of Cadmium Sulfide Nanoparticlesrahulmonikasharma
The dc conductivity of consolidated nanoparticle of CdS has been studied over the temperature range from 303 K to 523 K and the conductivity has been found to be much larger than that of single crystals.
A Survey on Peak to Average Power Ratio Reduction Methods for LTE-OFDMrahulmonikasharma
OFDM (Orthogonal Frequency Division Multiplexing) is generally preferred for high data rate transmission in digital communication. The Long-Term Evolution (LTE) standards for the fourth generation (4G) wireless communication systems. Orthogonal Frequency Division Multiple Access (OFDMA) and Single Carrier Frequency Division Multiple Access (SC-FDMA) are the two multiple access techniques which are generally used in LTE.OFDM system has a major shortcoming of high peak to average power ratio (PAPR) value. This paper explains different PAPR reduction techniques and presents a comparison of the various techniques based on theoretical results. It also presents a survey of the various PAPR reduction techniques and the state of the art in this area.
IOT Based Home Appliance Control System, Location Tracking and Energy Monitoringrahulmonikasharma
Home automation has been a dream of sciences for so many years. It could wind up conceivable in twentieth century simply after power all family units and web administrations were begun being utilized on across the board level. The point of home robotization is to give enhanced accommodation, comfort, vitality effectiveness and security. Vitality checking and protection holds prime significance in this day and age in view of the irregularity between control age and request observing frameworks accessible in the market. Ordinarily, customers are disappointed with the power charge as it doesn't demonstrate the power devoured at the gadget level. This paper shows the outline and execution of a vitality meter utilizing Arduino microcontroller which can be utilized to gauge the power devoured by any individual electrical apparatus. The primary expectation of the proposed vitality meter is to screen the power utilization at the gadget level, transfer it to the server and build up remote control of any apparatus. So we can screen the power utilization remotely and close down gadgets if vital. The car segment is additionally one of the application spaces where vehicle can be made keen by utilizing "IOT". So a vehicle following framework is additionally executed to screen development of vehicles remotely.
Thermal Radiation and Viscous Dissipation Effects on an Oscillatory Heat and ...rahulmonikasharma
An anticipated outcome that is intended chapter is to investigate effects of magnetic field on an oscillatory flow of a viscoelastic fluid with thermal radiation, viscous dissipation with Ohmic heating which bounded by a vertical plane surface, have been studied. Analytical solutions for the quasi – linear hyperbolic partial differential equations are obtained by perturbation technique. Solutions for velocity and temperature distributions are discussed for various values of physical parameters involving in the problem. The effects of cooling and heating of a viscoelastic fluid compared to the Newtonian fluid have been discussed.
Advance Approach towards Key Feature Extraction Using Designed Filters on Dif...rahulmonikasharma
In fast growing database repository system, image as data is one of the important concern despite text or numeric. Still we can’t replace test on any cost but for advancement, information may be managed with images. Therefore image processing is a wide area for the researcher. Many stages of processing of image provide researchers with new ideas to keep information safe with better way. Feature extraction, segmentation, recognition are the key areas of the image processing which helps to enhance the quality of working with images. Paper presents the comparison between image formats like .jpg, .png, .bmp, .gif. This paper is focused on the feature extraction and segmentation stages with background removal process. There are two filters, one is integer filter and second one is floating point Filter, which is used for the key feature extraction from image. These filters applied on the different images of different formats and visually compare the results.
Alamouti-STBC based Channel Estimation Technique over MIMO OFDM Systemrahulmonikasharma
This document summarizes research on using Alamouti space-time block coding (STBC) for channel estimation in MIMO-OFDM wireless communication systems. The proposed system uses 16-PSK modulation with up to 4 transmit and 32 receive antennas. Simulation results show that the proposed approach reduces bit error rate and mean square error at higher signal-to-noise ratios, compared to existing MISO systems. Alamouti-STBC channel estimation improves performance for MIMO-OFDM by achieving full diversity gain from multiple transmit antennas.
Empirical Mode Decomposition Based Signal Analysis of Gear Fault Diagnosisrahulmonikasharma
A vibration investigation is about the specialty of searching for changes in the vibration example, and after that relating those progressions back to the machines mechanical outline. The level of vibration and the example of the vibration reveal to us something about the interior state of the turning segment. The vibration example can let us know whether the machine is out of adjust or twisted. Al-so blames with the moving components and coupling issues can be distinguished. This paper shows an approach for equip blame investigation utilizing signal handling plans. The information has been taken from college of ohio, joined states. The investigation has done utilizing MATLAB software.
1) The document discusses using the ARIMA technique for short term load forecasting of electricity demand in West Bengal, India.
2) It analyzed historical hourly load data from 2017 to build an ARIMA model and forecast demand for July 31, 2017, achieving a Mean Absolute Percentage Error of 2.1778%.
3) ARIMA is identified as an appropriate univariate time series method for short term load forecasting that provides more accurate results than other techniques.
Impact of Coupling Coefficient on Coupled Line Couplerrahulmonikasharma
The coupled line coupler is a type of directional coupler which finds practical utility. It is mainly used for sampling the microwave power. In this paper, 3 couplers A,B & C are designed with different values of coupling coefficient 6dB,10dB & 18dB respectively at a frequency of 2.5GHz using ADS tool. The return loss, isolation loss & transmission loss are determined. The design & simulation is done using microstrip line technology.
Design Evaluation and Temperature Rise Test of Flameproof Induction Motorrahulmonikasharma
The ignition of flammable gases, vapours or dust in presence of oxygen contained in the surrounding atmosphere may lead to explosion. Flameproof three phase induction motors are the most common and frequently used in the process industries such as oil refineries, oil rigs, petrochemicals, fertilizers, etc. The design of flameproof motor is such that it allows and sustain explosion within the enclosure caused by ignition of hazardous gases without transmitting it to the external flammable atmosphere. The enclosure is mechanically strong enough to withstand the explosion pressure developed inside it. To prevent an explosion due to hot spot on the surface of the motor, flameproof induction motors are subjected to heat run test to determine the maximum surface temperature and temperature class with respect to the ignition temperature of the surrounding flammable gas atmosphere. This paper highlights the design features of flameproof motors and their surface temperature classification for different sizes.
Optimizing Gradle Builds - Gradle DPE Tour Berlin 2024Sinan KOZAK
Sinan from the Delivery Hero mobile infrastructure engineering team shares a deep dive into performance acceleration with Gradle build cache optimizations. Sinan shares their journey into solving complex build-cache problems that affect Gradle builds. By understanding the challenges and solutions found in our journey, we aim to demonstrate the possibilities for faster builds. The case study reveals how overlapping outputs and cache misconfigurations led to significant increases in build times, especially as the project scaled up with numerous modules using Paparazzi tests. The journey from diagnosing to defeating cache issues offers invaluable lessons on maintaining cache integrity without sacrificing functionality.
Null Bangalore | Pentesters Approach to AWS IAMDivyanshu
#Abstract:
- Learn more about the real-world methods for auditing AWS IAM (Identity and Access Management) as a pentester. So let us proceed with a brief discussion of IAM as well as some typical misconfigurations and their potential exploits in order to reinforce the understanding of IAM security best practices.
- Gain actionable insights into AWS IAM policies and roles, using hands on approach.
#Prerequisites:
- Basic understanding of AWS services and architecture
- Familiarity with cloud security concepts
- Experience using the AWS Management Console or AWS CLI.
- For hands on lab create account on [killercoda.com](https://killercoda.com/cloudsecurity-scenario/)
# Scenario Covered:
- Basics of IAM in AWS
- Implementing IAM Policies with Least Privilege to Manage S3 Bucket
- Objective: Create an S3 bucket with least privilege IAM policy and validate access.
- Steps:
- Create S3 bucket.
- Attach least privilege policy to IAM user.
- Validate access.
- Exploiting IAM PassRole Misconfiguration
-Allows a user to pass a specific IAM role to an AWS service (ec2), typically used for service access delegation. Then exploit PassRole Misconfiguration granting unauthorized access to sensitive resources.
- Objective: Demonstrate how a PassRole misconfiguration can grant unauthorized access.
- Steps:
- Allow user to pass IAM role to EC2.
- Exploit misconfiguration for unauthorized access.
- Access sensitive resources.
- Exploiting IAM AssumeRole Misconfiguration with Overly Permissive Role
- An overly permissive IAM role configuration can lead to privilege escalation by creating a role with administrative privileges and allow a user to assume this role.
- Objective: Show how overly permissive IAM roles can lead to privilege escalation.
- Steps:
- Create role with administrative privileges.
- Allow user to assume the role.
- Perform administrative actions.
- Differentiation between PassRole vs AssumeRole
Try at [killercoda.com](https://killercoda.com/cloudsecurity-scenario/)
Redefining brain tumor segmentation: a cutting-edge convolutional neural netw...IJECEIAES
Medical image analysis has witnessed significant advancements with deep learning techniques. In the domain of brain tumor segmentation, the ability to
precisely delineate tumor boundaries from magnetic resonance imaging (MRI)
scans holds profound implications for diagnosis. This study presents an ensemble convolutional neural network (CNN) with transfer learning, integrating
the state-of-the-art Deeplabv3+ architecture with the ResNet18 backbone. The
model is rigorously trained and evaluated, exhibiting remarkable performance
metrics, including an impressive global accuracy of 99.286%, a high-class accuracy of 82.191%, a mean intersection over union (IoU) of 79.900%, a weighted
IoU of 98.620%, and a Boundary F1 (BF) score of 83.303%. Notably, a detailed comparative analysis with existing methods showcases the superiority of
our proposed model. These findings underscore the model’s competence in precise brain tumor localization, underscoring its potential to revolutionize medical
image analysis and enhance healthcare outcomes. This research paves the way
for future exploration and optimization of advanced CNN models in medical
imaging, emphasizing addressing false positives and resource efficiency.
Advanced control scheme of doubly fed induction generator for wind turbine us...IJECEIAES
This paper describes a speed control device for generating electrical energy on an electricity network based on the doubly fed induction generator (DFIG) used for wind power conversion systems. At first, a double-fed induction generator model was constructed. A control law is formulated to govern the flow of energy between the stator of a DFIG and the energy network using three types of controllers: proportional integral (PI), sliding mode controller (SMC) and second order sliding mode controller (SOSMC). Their different results in terms of power reference tracking, reaction to unexpected speed fluctuations, sensitivity to perturbations, and resilience against machine parameter alterations are compared. MATLAB/Simulink was used to conduct the simulations for the preceding study. Multiple simulations have shown very satisfying results, and the investigations demonstrate the efficacy and power-enhancing capabilities of the suggested control system.
An improved modulation technique suitable for a three level flying capacitor ...IJECEIAES
This research paper introduces an innovative modulation technique for controlling a 3-level flying capacitor multilevel inverter (FCMLI), aiming to streamline the modulation process in contrast to conventional methods. The proposed
simplified modulation technique paves the way for more straightforward and
efficient control of multilevel inverters, enabling their widespread adoption and
integration into modern power electronic systems. Through the amalgamation of
sinusoidal pulse width modulation (SPWM) with a high-frequency square wave
pulse, this controlling technique attains energy equilibrium across the coupling
capacitor. The modulation scheme incorporates a simplified switching pattern
and a decreased count of voltage references, thereby simplifying the control
algorithm.
An improved modulation technique suitable for a three level flying capacitor ...
A New Method for Figuring the Number of Hidden Layer Nodes in BP Algorithm
1. International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 5 Issue: 9 101 – 114
_______________________________________________________________________________________________
101
IJRITCC | September 2017, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
A New Method for Figuring the Number of Hidden Layer Nodes in BP
Algorithm
Chang Mu*, Bo-Zhang Qiu, Xiu-Hai Liu
School of Science, Nanjing University of Science and Technology,
Nanjing, Jiangsu 210094, China
*Corresponding author: jt_mc17@sina.com
Abstract: In the field of artificial neural network, BP neural network is a multi-layer feed-forward neural network. Because it is difficult to
figure the number of hidden layer nodes in a BP neural network, the theoretical basis and the existing methods for BP network hidden layer
nodes are studied. Then based on traditional empirical formulas, we propose a new approach to rapidly figure the quantity of hidden layer nodes
in two-layer network. That is, with the assistance of experience formulas, the horizon of unit number in hidden layer can be confirmed and its
optimal value will be found in this horizon. Finally, a new formula for figuring the quantity of hidden layer codes is obtained through fitting
input dimension, output dimension and the optimal value of hidden layer codes. Under some given input dimension and output dimension,
efficiency and precision of BP algorithm may be improved by applying the proposed formula.
Key words: Artificial neural network; BP algorithm; Hidden layer nodes.
__________________________________________________*****_________________________________________________
I. Introduction
With the coming of information age, the data analysis
and its forecast and decision-making have become more and
more important. Artificial neural network is initiated through
imitation of the structure and characteristics of human brain.
Interconnecting a large number of simple processing units
(neurons or nodes), we get massively parallel distributed
information processing and nonlinear dynamic systems which
have several advantages, such as excellent fault-tolerance,
powerful learning ability, fast running speed and so on. It
contains a lot of networks, among which BP algorithm
proposed in 1986 is pretty mature. A group of scientists led by
Rumelhart and McCelland have detailedly discussed an error
back propagation algorithm error for multi-layer perceptron
with nonlinear continuous transformation function in their
book "Parallel distributed processing”. BP algorithm model
has become the most widely used neural network model, and
its application field is still expanding.
BP algorithm owns many merits including good
self-study character and the realization of nonlinear mapping
from the input space to the output space. However, there are
some disadvantages concerning BP algorithm, for example,
network training calling for a large scale of input and output
data, slow convergence speed, easy to fall into local minimum
point, strict choice of initial weights and difficulty of network
structure design, namely, there is no theoretical guidance
about the selection concerning quantity of hidden layer codes,
etc. In BP network, the choice of hidden layer nodes is very
crucial, not only having a great influence on the performance
of establishing neural network model, but also being the direct
reason of “overfitting” appearing in network training, but in
theory there exists no scientific and universal method for
determining hidden layer codes
]31[
.
In fact, applying artificial neural network to solve
nonlinear problems is a matching process between the optimal
solution of practical problem and the balance of network
through employing artificial neurons in the network. The
performance of network is an important index for judging the
2. International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 5 Issue: 9 101 – 114
_______________________________________________________________________________________________
102
IJRITCC | September 2017, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
quality of network structure, and the selection problem of
artificial neural network structure is actually a determining
problem about the number of hidden layer neurons. But at
present the formulas of the hidden layer nodes in lots of
articles are proposed provided that training samples can be
obtained without any limitation, and furthermore they are
given for the worst situation, which both seldom appear in
practice. In this paper, we mainly discuss the selection
problem for the quantity of hidden layer neurons in two-layer
BP neural network, and through the investigation about the
simulation for 6 common functions when input dimension and
output dimension are both limited from 1 to 10, a new method
for determining the number of neurons in hidden layer of BP
algorithm is presented.
1 BP neural network model
1.1 Basic Theory of BP Neural Network
BP neural network (back-propagation neural network) is
a multi-layer forward network, and its structure shows as the
following:
The first layer of BP neural network is called input layer,
and the last layer is output layer, and every middle layer is
called hidden layer. The nonlinear input-output relationship
between hidden layer k and the neurons in output layer is
denoted as kf ( mk ,,2 ), and the connection weight
from the j -th neuron of the ( 1k )th layer to the i -th
neuron of the k -th layer is
k
ijw . The total inputs of the i -th
neuron in the k -th layer are
k
iu , and the output of this
neuron is .yk
i The relationships between the variables are:
k
iy = kf (
k
iu ) ,
k
iu = ,w 11
l
j
j
k
ij y
.,,2 mk
If the input data of BP neural network is X [ 1x p ,
2x p ,…, npx ]
T
(assuming there are np neurons in input
layer), going through every hidden layer node in turn from
input layer, the output data Y [
m
a1
y ,
m
a2
y ,…,
m
am
y ]
T
(assuming there are ma neurons in output layer) will be
obtained. Hence BP neural network can be treated as a
nonlinear mapping from input space to output space.
BP learning algorithm minimizes the error according to
the reverse learning process, so the objective function is
selected as the following:
.)(
2
1
1j
2
ma
sj
m
j yyJMin
That is, the sum of the squares of the difference between the
expected output sjy and the actual output
m
jy in the neural
network is minimized through choosing appropriate neural
network weights. Actually, this learning algorithm is to look
for the minimum value of the objective function J subject
to constraint conditions:
k
iy = kf (
k
iu ),
k
iu =
11
w
l
j
j
k
ij y ,
.,,2 mk
Employing "steepest descent method" in nonlinear
programming, changing the weight along the negative
gradient direction of the objective function, we get the neural
network weight correction △ 1k
ijw 1
k
ijw
J
( 0 ),
where is the learning step size.
3. International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 5 Issue: 9 101 – 114
_______________________________________________________________________________________________
103
IJRITCC | September 2017, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
Therefore BP learning algorithm can be summarized as
follows:
△ 1k
ijw 1
j
k
i yd
))(1( si
m
i
m
i
m
i
m
i yyyyd
1
)1(
k
l
l
k
li
k
i
k
i
k
i dwyyd
( 2,,1 mk )
]5,4[
.
Thus, the selection of the error function is a recursive
process that begins with the back propagation of output layer,
so it is named as back propagation learning algorithm.
Through learning plenty of samples, modifying the weight,
then constantly reducing the deviation, finally well-pleasing
results will be achieved.
The basic idea of the BP algorithm is that the learning
process consists of two processes: the forward propagation of
the signal and the reverse propagation of the error. Forward
propagation is that the input samples come from input layer
and are processed by every hidden layer to output layer in the
end. If the actual output of output layer does not match the
expected output, it turns into the reverse propagation phase of
the error. The reverse propagation of the error makes output
error going to input layer through every hidden layer by a
given form, and the error is distributed to all units of every
layer, so the error signal of every layer unit is obtained to be
the foundation for the correction of every unit weight. This
forward propagation of the signal and every layer weight’s
adjustment process of error backwards propagation are carried
out round and round. In reality, weight’s adjustment process is
learning and training process of the network. This process
continues running until the error of the network output is
reduced to an acceptable level, or until the pre-set learning
times is reached.
]6[
1.2 Improvement of standard BP algorithm
BP algorithm is applied to a three-layer perceptron with a
nonlinear transformation function, which can be employed to
approximate any nonlinear function with arbitrary precision.
This extraordinary advantage makes the multi-layer sensor
more and more widely used. However, the standard BP
algorithm exposes a lot of inherent defects in practice, such as
easy to fall into local minimum point and thus hard to
approach the global optimal value. When training new
samples it has the tendency to forget old samples.
Furthermore more training times make the learning efficiency
lower and the convergence speed is slow. For these
shortcomings of BP algorithm, many researchers have put
forward lots of effective improvement algorithms. A brief
introduction about three commonly used methods is presented
as follows.
1.2.1 Increase momentum
Some mathematicians in 1986 stated the standard BP
algorithm adjusts the weight only according to negative
gradient direction of the error in time t , but neglects the
gradient direction before time t , hence leading to training
process concussion and slow convergence speed. In order to
improve training speed of the network, a momentum item may
be added in the weight adjustment formula. If W represents
the weight matrix of a layer, and X represents the input
vector of a layer, then the weight adjustment vector including
the momentum item can be expressed as:
).1()( tWXtW
1.2.2 Adaptive adjustment learning rate
The learning rate is also called the step size and
keeps constant in standard BP algorithm. However, in practice,
it is difficult to determine the best learning rate that is
appropriate from beginning to end. It can be seen from the
error surface that training times may increase in the flat area if
is too small, so is expected to become bigger. In the
region where the error variation changes rapidly, the narrow "
Pound " may be missed if is too large which makes
training effect shocking and iteration times increasing. In
order to speed up the convergence process, a better idea is to
change the learning rate adaptively, so that the rate
increases and decreases in proper situations.
4. International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 5 Issue: 9 101 – 114
_______________________________________________________________________________________________
104
IJRITCC | September 2017, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
1.2.3 Introduce steepness factor
There are some flat areas on the error surface, and the
weight adjustment enters into a flat area because the neuron
output enters into a saturation region of the transferring
function. If we reduce the net input of the neuron after the
adjustment entering into the flat area, letting the output depart
from the unsaturated region of the transferring function, hence
the shape of the error function can be converted so that the
adjustment will be separated from the flat area.
]8,7,5[
1.3 Research results on the number of hidden nodes
1.3.1 Empirical formula method
At present, there does not exist a complete theoretical
guide concerning the method for determining the number of
hidden neurons in artificial neural networks which in
operation is usually based on neural network, designer’s
experience and estimation.
(1) Kolmogorov gave an equality about inN and hidN
for the artificial neural network containing only three layers:
12 inhid NN ;
(2) Jadid and Fairbarin presented the upper bound
formula of the hidden neurons:
outin
train
hid NNR
N
N
,
where trainN is the number of training samples, and
105 R ;
(3) Hecht-Nieslsen theoretically proved the inequality
relation between the number of hidden nodes and the number
of input nodes according to Kolmogorov theorem:
1 inhid NN ;
(4) Gao obtained an empirical formula by simulations:
212
35.077.054.212.043.0 outinoutoutinhid NNNNNN
,
and the right part of the results retained integer as the value of
hidN ;
]9[
(5) According to some given learning samples, Zhou and
other researchers selected different numbers of neurons in
hidden layer, training the network model and then comparing
different results of the model, and finally proposed an
empirical formula:
21
11 outinhid NNN .
In fact, once the number of neurons in hidden layer is
determined, the number of neurons will never be changed
throughout the network adjustment. In this case, the number
of neurons may be either too large or too small in the
established neural network model, so that it is hard to ensure
whether the network structure is optimal or not.
1.3.2 Repeated test method
Most researchers do not advocate to employ empirical
formula, and they consider that the most effective way to find
the optimal number of hidden neurons is based on repeated
tests, so that the number of hidden layer neurons which
enables the sample error to achieve the preset default accuracy
can be regarded as the optimal number of hidden layer
neurons in the network model. Cross-validation is a typically
iterative test method, finding effective information from the
existing learning samples, and the connection strength
between different network neurons will be adjusted according
to the information. Applying the cross-validation method, we
are able to find the optimal number of hidden layer neurons,
so that the predetermined accuracy can be achieved for the
network learning samples. The cross-validation method is
simple and easy to understand. Firstly, the learning samples
are divided into two parts: learning samples and test samples;
secondly, the learning samples are networked to adjust the
neuron connection strength until the preset accuracy of the
network is approached; finally learning for test samples is
made. In short, in order to get a better learning performance,
both learning samples and test samples should be as much as
possible to participate in learning.
]10[
1.3.3 Growth Law
5. International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 5 Issue: 9 101 – 114
_______________________________________________________________________________________________
105
IJRITCC | September 2017, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
The so-called growth method refers to choose a few
neurons of hidden layer before the operation of the artificial
neural network, and then gradually increases the number of
hidden neurons (increasing one or more nerves each time) in
learning process for a given practical problem until the
performance requirements of the network model are satisfied.
1.3.4 Global minimum method
In 1989, Mezard from the perspective of the complexity
of space, increasing the number of hidden nodes will reduce
the practical problem of optimization of the dimension of
space, if the appropriate increase in neurons and connectivity,
reasonable to determine the network learning parameters, the
network error may be out Local minima and converge to
global minima. Based on this idea, he proposed the Tiling
algorithm to construct the main neuron module and the
auxiliary neuron module on each layer to achieve the desired
mapping:
Rn
→Bm
={0,1}m
or{-1,1}m
Frean in 1990 gave the upstar algorithm to add neurons
to reduce the network learning error in the way the network at
the global minimum point of convergence, to balance the data.
In 1992, Chang et al. established a three-layer general network
model. When the network into a local minimum, add a hidden
layer of neurons to ensure that after adding neurons to the
network learning error than adding neurons before the school
learning error is small. The neurons are added repeatedly until
the network error converges to zero.
1.3.5 Cascade-Conelation algorithm
Fahhnan's CC algorithm is an optimization algorithm that
can construct the optimal network structure in the network
learning process. Before the network begins to learn, the
number of hidden neurons is set to zero; then the network is
trained to add a hidden layer neuron until the network learning
error does not fall again. The intensity of the interaction
associated with the newly added neurons can be determined
by the covariance of the actual output value of the neuron and
the network learning error is determined. Follow this step to
learn more until you meet the requirements of the network
model default accuracy. It is not difficult to find that the CC
algorithm is a greedy algorithm that seeks to make each
neuron contribute to the convergence of the network to the
preset accuracy, but there is a benefit that at any one time,
only one neuron's intensity is adjusted , And do not need
feedback.
1.3.6 Modular growth method
In the process of online learning, each time not add a
neuron, but to contain multiple neurons of the module.
Littmann et al. constructed a modular approach to a neural
network that only applies to single output neurons, which
adjusts the connection strength between neurons by
introducing an error evaluation function. Lu and other
proposed PMSN structure, each increase is a filter module SM
(SievingModel), do not have to re-adjust the parameters have
been learned, so the network learning speed quickly.
II. BP algorithm test network structure
2.1 Determination of the number of hidden layers
In the BP algorithm, although the hidden layer structure
design involves network weight, incentive function, learning
rate and other factors, but the direct impact of network
performance is the hidden layer structure of the layers and the
number of nodes. From the existing research results, we can
see that the more complex the number of hidden layers, the
more complex the mapping expressed by the network, not
only increase the computational complexity, but also lead to
data distortion, and according to Kolmogorov theorem, when
the hidden layer is greater than or equal to 1 , The BP network
has the ability to approach all nonlinear maps, that is, when
the hidden layer nodes can be set freely according to the need,
then with two layers of forward neural network can achieve
arbitrary precision approximation of any continuous function,
so Experiment from a simple and practical point of view,
choose a hidden layer of the network structure.
]11[
Kolmogorov theorem: Given any ε> 0, there is a
three-layer BP neural network for any continuous function f,
6. International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 5 Issue: 9 101 – 114
_______________________________________________________________________________________________
106
IJRITCC | September 2017, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
the input layer has 1p neurons, the middle layer has 2 1p +1
neurons, and the output layer has mp neurons. It can be
approximated by f in any ε square error accuracy. The
theorem proves not only the existence of the mapping network,
but also the structure of the mapping network. That is, there is
always a three-layer forward neural network with a structure
of 1p ×(2 1p +1)× mp that can accurately approximate any
continuous function f.
]10[
2.2 Determination of the number of hidden nodes
It is known that a two-layer BP network with infinite
hidden layer nodes can realize any non-linear mapping from
input to output, but for a limited number of input-to-output
mappings, there is no need for infinite hidden nodes. How to
choose the number of hidden nodes. If the number of hidden
nodes is too large, the generalization ability of the network
will be worse after the study. If the number of hidden nodes is
too small, the learning time will be too long, the network fault
tolerance will be poor, the local minimum will be, the error
will not be the smallest, and do not produce enough number of
connection weights to meet a number of sample learning. Due
to the complexity of the problem, so far has not yet found a
good analytical formula, the number of hidden nodes is often
based on the empirical formula derived from the empirical test
and their own to determine, if the requirements of the
approximation of the variable function changes, The number
of nodes in the hidden layer should be more; if the specified
approximation accuracy is high, the number of hidden layer
units should be more; if the beginning is to put less hidden
layer unit , According to the subsequent study of the situation
gradually increased, or the beginning to join enough hidden
nodes, through learning to not work on the connection and
hidden layer nodes deleted.
]12[
III. A fitting experiment about number formula of hidden
layer nodes
3.1 Experimental route
The number of neurons in establishing hidden layer
neural network should follow the principle --- selecting
compact structure under the condition that the precision
requirement is satisfied, that is, as few hidden layer nodes as
possible. In this experiment, the empirical formula is
employed to determine the range of the number of hidden
layer nodes, and the optimal value will be searched in this
range. According to the empirical formula
amnN 2/1
)(
(where N is the number of corresponding hidden layer nodes
for minimum error, n is the dimension of input data, m
is the dimension of output data , a is a constant number
between 1 and 10), the optimal number of nodes is determined
in the interval 15,1 provided that input dimension and
output dimension both change from 1 to 10. With the help of
Matlab program, the number of hidden layer nodes is
traversed from 1 to 15, then recording the error about different
numbers of nodes and making comparison between these
errors, determining and recording the corresponding numbers
of hidden layer nodes for minimum error under different
dimensions of input and output data, finally the corresponding
formula is obtained according to the fitting between input and
output dimensions and the number of hidden layer nodes with
minimum error. The experimental processes are listed as
follows:
(1) Obtaining data sets (sample sets and experimental
sets);
(2) Inputting sample sets, and training the network;
(3) Inputting experimental sets to the network trained,
and traversing different numbers of hidden layer nodes and
then getting corresponding errors;
(4) Selecting the corresponding number of nodes for the
minimum error under different input and output data with
different dimensions, and then recording m (dimension of
input layer), n (dimension of output layer), N (the
7. International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 5 Issue: 9 101 – 114
_______________________________________________________________________________________________
107
IJRITCC | September 2017, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
number of corresponding hidden layer nodes for minimum
error);
(5) Fitting function relations according to the optimal
number of codes and input and output data;
]13[
3.2 Experimental data
3.2.1 Selection of experimental functions
In the experiment, the functions of generating
experimental data are constructed by using the compound
operations of elementary functions (including trigonometric
function, exponential function, logarithmic function and so
on), for example,
153 3
3
2
4
1 xxx ,
)(log3)(log2)(log 310210110 xxx
,
)cos()(log3)tan(*)tan( 431021 xxxx
,
4
4321 )cot(*)tan(2)cos(3)sin( xxxx
,
etc.
3.2.2 Normalization (pretreatment)
Normalization processing means that the input and
output data of a network are limited to 1,0 or 11- ,
through transformation processing, and main reasons for
normalization are:
(1) The input data of the network usually have different
physical meanings or different dimensions, and the
normalization makes all components change in 1,0 or
11- , , so that the network training gives every input
component equal position in the beginning.
(2) All neurons of BP neural network adopt transferring
function of Sigmoid class, which can prevent the neuron
output from being saturating because the absolute value of net
input is too large, and then adjust the weight into the flat area
of error surface.
(3) We know the output of Sigmoid transferring function
changes in [0, 1] or [-1, 1], if the output data are not
transformed, the absolute error of the output component
owning big numerical value will be large, and the absolute
error of the output component with small value will be small.
The weight adjustment only for total error of output in
network training leads to a problem that an output component
with small proportion possessed in total error has large
absolute error, while this problem can be settled after the
normalization of output data.
Normalized formula is:
𝑥𝑖 =
𝑥𝑖 − 𝑥 𝑚𝑖𝑛
𝑥 𝑚𝑎𝑥 − 𝑥 𝑚𝑖𝑛
where ix is input or output data, minx represents the
minimum value of the data, maxx stands for the maximum
value of the data.
]14[
3.2.3 Normalized instance
Input 1x
Original value/ After
normalization
Input 2x
Original value/ After
normalization
Input 3x
Original value/ After
normalization
Output 1y
Original value/ After
normalization
Output 2y
Original value/After
normalization
2771/-0.4465 191/-0.96219 5254/0.05042 9.44416/0.13394 19.166/0.23053
7132/0.426484 5717/0.143229 1322/-0.73629 10.7316/0.57939 20.731/0.485137
8. International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 5 Issue: 9 101 – 114
_______________________________________________________________________________________________
108
IJRITCC | September 2017, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
1649/-0.6711 6688/0.337467 6105/0.220688 10.828/0.61282 22.225/0.728122
3753/-0.24992 9646/0.929186 9090/0.817927 11.517/0.851245 23.419/0.922346
2611/-0.47853 375/-0.92539 9982/0.996399 9.990/0.322821 20.563/0.457689
8446/0.689521 5903/0.180436 1875/-0.62565 10.971/0.662133 21.288/0.575678
2224/-0.556 6960/0.391878 2508/-0.499 10.589/0.530081 21.230/0.56633
5984/0.196677 9894/0.978796 669-0.86695 10.598/0.533097 20.244/0.405874
113/-0.97858 8039/0.607722 9324/0.864746 9.928/0.301309 21.772/0.654496
8185/0.637274 8624/0.724745 7239/0.447579 11.708/0.917371 23.363/0.913351
3.3 Experimental parameters
In the process of training the network, the learning rate is
set to be 0.001, and the training times are 300, and the training
precision is 0.00005. The total error performance function is
mean square error (MSE), and the weights and thresholds of
the network are corrected successively. Then the total error
decreases until the required error performance is reached.
When there are lots of samples, the correction of weights and
thresholds follows the input of all samples and the calculation
of total error, so that the convergence speed can be accelerated.
In general, the input-output relationships of the neurons in
hidden layer are nonlinear functions, and the transferring
function of hidden layer is defined as Tansig function, namely,
the tangent Sigmoid transferring function, which limits the
output value in the interval [0, 1] and reduces the amount of
computation. The output layer acts as buffer memory, adding
data source to the network. The input-output relationship of
input layer is usually linear, and the Purelin function is
selected, which is a pure linear transferring function, so that
the result of output layer can be taken to any value without
prejudice to the expression of the output result.
Training function is Trainrp, that is to say, the elastic
reflection propagation algorithm is employed to train the
network. The weight and threshold change according to the
variation of the gradient for many training functions. However,
if the function is very flat near the extreme value, the variation
of the gradient is very small, and therefore the change of
weight and threshold will be very small. Too small variation
of the gradient not only drive the network training
convergence speed being very slow, but also can the network
not achieve the training performance requirements.
Fortunately, Trainrp function has the ability to overcome this
shortcoming, because Trainrp function uses the direction of
error gradient to determine the variation of weight and
threshold, and the size of the error gradient has no influence
on the variation of weight and threshold.
]16,15,14[
11. International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 5 Issue: 9 101 – 114
_______________________________________________________________________________________________
111
IJRITCC | September 2017, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
3 8 5
4 8 9
5 8 14
6 8 7
7 8 12
8 8 13
9 8 7
10 8 9
1 9 6
2 9 11
3 9 8
4 9 8
5 9 14
6 9 7
7 9 14
8 9 10
9 9 7
10 9 9
1 10 6
2 10 8
3 10 8
4 10 6
5 10 12
6 10 12
7 10 11
8 10 11
9 10 5
10 10 6
Selecting some common continuous functions including sine
function, cosine function, tangent function, logarithmic
function, power function and quadratic function to combine as
the source of experimental data, 1000 groups of function data
are obtained by calculation provided that input dimension and
output dimension are both restricted from 1 to 10, in which
100 groups are the sample sets for training network and the
rest are experimental sets. The experimental parameters are
set up according to section 3.3, and the data obtained are fitted
to get the formula for figuring the number of hidden layer
nodes (four decimal digits reserved):
.0352.01049.04814.11201.06937.10366.1 22
nmmmnnN
12. International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 5 Issue: 9 101 – 114
_______________________________________________________________________________________________
112
IJRITCC | September 2017, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
IV. Simulation verification
4.1 Data
Select the function 4231 **)(f xxxxx as a test function for simulation experiments, and choose 10 sets of data as a test
sample.
Function input ( ix ) and expected output (Y )
1x 2x 3x 4x
Y
0.565809 0.276021 -0.84075 -0.96778 -0.24107
-0.74237 0.253203 -0.52786 -0.26188 -0.32922
-0.18367 0.101081 -0.50645 -0.84369 -0.37273
-0.3286 0.428943 -0.36021 -0.37796 -0.14799
-0.73236 0.058247 -0.35541 0.009507 -0.25600
0.061355 -0.48179 0.696509 0.735415 0.318415
0.671905 -0.95036 -0.69831 -0.13800 -0.28288
-0.98619 -0.45757 0.670101 -0.14720 -0.22783
-0.05375 -0.17014 -0.67230 -0.19824 -0.27711
0.472125 -0.55705 -0.47464 -0.84389 -0.36447
4.2 Experimental output
Input the function data to the network as experiment, obtaining the total error corresponding to the number of hidden nodes
in [2,15].
Number of
hidden layer
nodes
Total
error
2 0.014000
3 0.007991
4 0.009578
5 0.010067
6 0.008907
13. International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 5 Issue: 9 101 – 114
_______________________________________________________________________________________________
113
IJRITCC | September 2017, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
7 0.007160
8 0.006991
9 0.009317
10 0.008777
11 0.012671
12 0.016822
13 0.014904
14 0.022306
15 0.011066
4.3 Result analysis
It can be seen from the simulation results that in the case
of selecting the function f (x) with four-dimension input and
one-dimension output, the number of hidden neurons of the
BP algorithm with the smallest total error is 8N .
Substituting 1,4 nm into the formula:
,0352.01049.04814.11201.06937.10366.1 22
nmmmnnN
Easily, we know theoretically optimal number of hidden layer
nodes: 7N . Based on the total error of test output, the
total errors are almost identical when hidden layer nodes are 7
and 8. Furthermore, the total errors are all very small in the
interval 152, . Considering the randomness of the training
network, it is proved that the formula are able to improve the
efficiency for determining the optimal number of hidden layer
nodes within the allowable range of the error under some
given input and output dimensions, and also improve the
precision of the continuous function approximation by
employing the BP algorithm network.
V. Conclusions
BP algorithm employing reverse propagation is the most
widely used structure of artificial neural networks. In this
paper, we took the basic multi-input and multi-output single
hidden layer BP network as an example, and some typical
continuous functions were trained under different input and
output dimensions, and also the global error of each network
output has been analyzed. The result is that when input
dimension and output dimension are both less than or equal to
10, and the learning rate is 0.001, and the training times are
300, and the training precision is 0.00005, the formula for
figuring the number of hidden layer nodes with the best
learning convergence character and function approximation
accuracy is obtained through approaching six kinds of
functions. This investigation may provide inspiration for how
to figure the number of hidden layer nodes in some BP
networks with other input and output dimensions or the multi
hidden layer structure.
There are still many deficiencies in this experiment.
Firstly, interval range is derived from the most commonly
used empirical formula which perhaps is not the interval
where the optimal node locates. Secondly, most parameters
14. International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 5 Issue: 9 101 – 114
_______________________________________________________________________________________________
114
IJRITCC | September 2017, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
are certain, such as learning rate and training parameters,
while the experiment does not ensure that this formula is still
the best when the learning rate and the number of training
times change. Thirdly, the training sample set is not large
enough to approximate the six kinds of functions, for each
experimental data are only 1000 groups. In the future, we are
about to deepen the study of BP algorithm, exploring the
problem about the optimal number of hidden layer nodes in
more complex cases. We plan to determine the interval range
at first, and then make selection for the number of hidden
layer nodes, so the amount of computation may be reduced
remarkably. Moreover, the formula for determining the
number of hidden layer nodes in this experiment can also be
applied as reference and judging basis.
References:
[1] Nielson, R H., Theory of the Backpropagation Neural
Network, Neural Networks, 1(Suppl.1): 593-594,
1989.
[2] Dayhoff, J. E. and DeLeo, J. M., Artificial neural
networks: Opening the black box, Cancer, 91(s8):
1615–1635, 2001.
[3] Hamid, B. Mohamad R.M., A learning
automata-based algorithm for determination of the
number of hidden units for three-layer neural
networks, International Journal of Systems Science,
40(1): 101-118, 2009.
[4] Han, L., Introduction to Artificial Neural Network,
Beijing: Beijing University of Posts and
Telecommunications Press, 2006.
[5] Chu, L., Chen, S., and Zhou, M., Mathematical Basis
of Computing Intelligence, Beijing: Science Press,
2002.
[6] Xu, L., Neural Network Control, Beijing: Electronic
Industry Press, 2009.
[7] Li, Y., BP Neural Network Research and
Improvement and Applications,
http://cdmd.cnki.com.cn/Article/CDMD-10361-1012
421305.htm , 2012.
[8] Fan, J., Wang, Z. and Qian, F., Research progress of
hidden layer structure design of BP artificial neural
network, Control Engineering of China, 12(Suppl.1):
105-108, 2005.
[9] Gao, D., On Structures of supervised linear basics
function feedforward three-layered neural networks,
Journal of Circuits and Systems, 2 (03): 31-36. 1997.
[10] Wang, W. Artificial Intelligence and Its Application.
Beijing: Higher Education Press, 2016.
[11] Yan, H. and Guan, Y., Method to determine the
quantity of internal nodes of back propagation neural
networks and its demonstration, Control Engineering
of China, 16(S2): 100-102, 2009.
[12] Shi, Y., Han, L. and Lian, X., Neural Network
Design Method and Case Analysis. Beijing: Beijing
University of Posts and Telecommunications Press,
2009.
[13] Zhang, Y., and Cai, B., Research Progress of
Artificial Neural Network and Argument of
Publication Process, Beijing: Electronic Industry
Press, 2010.
[14] Chen, M., Matlab Neural Network Principle and
Examples of Fine Solution. Beijing: Tsinghua
University Press, 2012.
[15] He, Z., Matlab R2015b Neural Network
Technology,Beijing: Tsinghua University Press,
2016.
[16] Wang, X., Shi, F. and Yu, L., 43 Cases of Matlab
Neural Network Analysis, Beijing: Beihang
University Press, 2013.