Weights Of Artificial Neural Network Research Articles

PENENTUAN PARAMETER LEARNING RATE SELAMA PEMBELAJARAN JARINGAN SYARAF TIRUAN BACKPROPAGATION MENGGUNAKAN ALGORITMA GENETIKA


 
 
 
 One of the weakness in backpropagation Artificial neural network(ANN) is being stuck in local minima. Learning rate parameter is an important parameter in order to determine how fast the ANN Learning. This research is conducted to determine a method of finding the value of learning rate parameter using a genetic algorithm when neural network learning stops and the error value is not reached the stopping criteria or has not reached the convergence. Genetic algorithm is used to determine the value of learning rate used is based on the calculation of the fitness function with the input of the ANN weights, gradient error, and bias. The calculation of the fitness function will produce an error value of each learning rate which represents each candidate solutions or individual genetic algorithms. Each individual is determined by sum of squared error value. One with the smallest SSE is the best individual. The value of learning rate has chosen will be used to continue learning so that it can lower the value of the error or speed up the learning towards convergence.
 
 
 
 
 The final result of this study is to provide a new solution to resolve the problem in the backpropagation learning that often have problems in determining the learning parameters. These results indicate that the method of genetic algorithms can provide a solution for backpropagation learning in order to decrease the value of SSE when learning of ANN has been static in large error conditions, or stuck in local minima

Localizing and quantifying the intra-monomer contributions to the glass transition temperature using artificial neural networks

We used fully connected artificial neural networks (ANN) to localize and quantify, based on the monomer structure of several polymers, the specific features responsible for their observed glass transition temperatures (Tg). The use of ANNs allows us not only to successfully predict the Tg of the polymers but, even more important, to understand what parts of the monomer are mainly contributing to it. For this task, we used the weights of a trained ANN as obtained after fitting the input data (monomer structure) to the corresponding Tg value. The study was performed for a set of more than 200 atactic acrylates for which typical Tg defining features were identified. Thus, the ANN is able to recognize the relevance of the backbone stiffness, the length of pending groups or the presence of methyl groups on the value of the glass transition temperature. This approach can be easily extended to many other interesting properties of polymers and it is worth noting that only the monomer chemical structure is needed as input. This method is potentially useful for identifying orthogonal ways of tuning polymer properties during the design and development of new materials and it is expected that it will contribute to a better understanding of the polymer's behavior.

Classification of Erythemato-squamous diseases using Artificial Neural Network and Genetic algorithm

This paper introduces a hybrid model using artificial neural network (ANN) and genetic algorithm (GA) to develop an efficient classification technique for classification of different categories of Erythemato-squamous diseases. Neural network has been extensively used in many applications like classification, regression, web mining, system identification and pattern recognition. Weight optimization in neural network has been a matter of concern for researchers in the field of soft computing. In this paper the weights of ANN are optimized with GA. The proposed hybrid model is applied on the Erythemato-squamous dataset taken from UCI machine learning repository. The dataset contains six different categories: psoriasis, seboreic dermatitis, lichen planus, pityriasis rosea, chronic dermatitis and pityriasis rubra pilaris of Erythemato-squamous diseases. The main aim of this paper is to determine the type of Eryhemato-Squamous disease using the hybrid model. The performance of the hybrid model is evaluated using statistical measures like accuracy, specificity and sensitivity. The accuracy of the proposed model is found to be 99.34% on test dataset. The experimental result shows the effectiveness of the hybrid model in classification of Erythematosquamous diseases.

PENERAPAN ARTIFICIAL NEURAL NETWORK DENGAN OPTIMASI MODIFIED ARTIFICIAL BEE COLONY UNTUK MERAMALKAN HARGA BITCOIN TERHADAP RUPIAH

Bitcoin is one of digital assets that can be used to make a profit. One of the ways to use Bitcoin profitly is to trade Bitcoin. At trade activities, decisions making whether to buy or not are very crucial. If we can predict the price of Bitcoin in the future period, we can make a decisions whether to buy Bitcoin or not. Artificial Neural Network can be used to predict Bitcoin price data which is time series data. There are many learning algorithm in Artificial Neural Network, Modified Artificial Bee Colony is one of optimization algorithm that used to solve the optimal weight of Artificial Neural Network. In this study, the Bitcoin exchage rate against Rupiah starting September 1, 2017 to January 4, 2019 are used. Based on the training results obtained that MAPE value is 3,12% and the testing results obtained that MAPE value is 2,02%. This represent that the prediction results from Artificial Neural Network optimized by Modified Artificial Bee Colony algorithm are quite accurate because of small MAPE value.

Ultimate strength prediction of corroded plates with center-longitudinal crack using FEM and ANN

Corrosion and cracking are two main reasons for failure of plates in ships and offshore structures. In the current paper, the ultimate strength of plates with pitting corrosion and a longitudinal crack is investigated under uniform in-plane compression using non-linear finite element method (FEM). Ultimate strength of cracked-pitted plates are predicted using proper artificial neural network (ANN) under different aspect ratio, thickness of plate, number of pits, crack length, and pit depth over thickness of the plate. Based on the numerical results, in rectngular plates, ultimate strength for pitted plate and cracked-pitted plate are almost the same, while in square plates when degree of pitting is low, negligible effect of longitudinal crack on the ultimate strength is detected. For thin plates, less influence of pit corrosion on ultimate strength reduction is observed in comparison to thick plates. Also, it is found that the arrangements of pits in the cracked-pitted plates on ultimate strength reduction is more influential than number of the pits. Finally, an equation is suggested to predict the ultimate strength of corroded plated with longitudinal crack using ANN's weights and bias.

Optimal trained artificial neural network for Telugu speaker diarization

Speaker indexing or diarization is the process of automatically partitioning the conversation involving multiple speakers into homogeneous segments and grouping together all the segments that correspond to the same speaker. So far, certain works have been done under this aspect; still, the need of accurate partitioning process gets lagged under certain criteria. With this in mind, this paper aims to introduce a new speaker indexing or diarization model (Telugu language) that initially involves Mel Frequency Cepstral coefficient based feature extraction. Subsequently, a new Optimized Artificial Neural Network (ANN) is introduced for clustering process. The novelty behind the clustering process is: the training of ANN takes place through optimization logic that updates the weight of ANN by a hybrid concept of Artificial Bee Colony (ABC) and Lion Algorithm (LA). Thereby, the proposed model is named as ANN-ABC-LA model. Finally, the performance of the proposed ANN-ABC-LA model is compared over the state-of-the-art models with respect to different performance measures.

Prediction of Base Pressure in a Suddenly Expanded Flow Process at Supersonic Mach Number Regimes using ANN and CFD

Sudden expansion of flow in supersonic flow regime has gained relevance in the recent pasts for a wide run of applications. A number of kinematic as well as geometric parameters have been significantly found to impact the base pressure created within the suddenly expanded stream. The current research intends to create a predictive model for base pressure that is established in the abruptly extended stream. The artificial neural network (ANN) approach is being utilized for this purpose. The database utilized for training the network was assembled utilizing computational fluid dynamics (CFD). This was done by the design of experiments based L27 Orthogonal array. The three input parameters were Mach number (M), nozzle pressure ratio (NPR) and area ratio (AR) and base pressure was the output parameter. The CFD numerical demonstrate was approved by an experimental test rig that developed results for base pressure, and used a nozzle and sudden extended axis-symmetric duct to do so. The ANN architecture comprised of three layers with eight neurons in the hidden layer. The algorithm for optimization was Levenberg-Marquardt. The ANN was able to successfully predict the base pressure with a regression coefficient R2 of less than 0.99 and RMSE=0.0032. The importance of input parameters influencing base pressure was estimated by using the ANN weight coefficients. Mach number obtained a relative importance of 47.16% claiming to be the most dominating factor.

Complementary Modeling of Gravel Road Traffic-Generated Dust Levels Using Bayesian Regularization Feedforward Neural Networks and Binary Probit Regression

Gravel roads require extensive maintenance and rehabilitation. That is because of the dynamic behavior of gravel road conditions. This study is aimed at investigating road condition factors affecting the traffic-generated dust on gravel roads. The study concerns Laramie County, Wyoming. The parametric binary probit regression structure and the non-parametric Bayesian regularization artificial neural network (BRANN) methods were implemented to model traffic-generated dust levels as a function of the factors that contribute to dust generation. The BRANNs method simplifies the model building process by precluding irrelevant and redundant weights of artificial neural networks (ANNs). In this study, the BRANN method is utilized with one single hidden layer using the MATLAB® function for neural networks. In the hidden layer, multiple neuron counts ranging from three to thirty were attempted. The parametric and non-parametric techniques mentioned were adopted to provide comprehensive insights into important factors that contribute to dust generation. Therefore, both techniques complement each other. A total of 206 gravel road segments were used for model building for both analyses. As per the results of the BRANN model, it was found that twenty neurons produced the most accurate results. Furthermore, it was found that the BRANN model had more variables than the binary probit regression model, whereas the probit model provided general insights into the factors affecting the dust on gravel roads such as average travel speed and soil type. Also, it would be an easy-to-use method to assist local agencies and DOT practitioners in addressing the dust problems on gravel roads.

EVALUATION OF ESTIMATION PERFORMANCE FOR SOIL MOISTURE USING PARTICLE SWARM OPTIMIZATION AND ARTIFICIAL NEURAL NETWORK

Soil plays a vital role in the climate system. This paper performs a hybrid methodology that consists of particle swarm optimization (PSO) and artificial neural network (ANN) to estimate soil moisture (SM) by considering different parameters that include air temperature, time, relative humidity and soil temperature. Besides, this paper investigates the effects of the parameters of PSO-ANN by utilizing from the response surface. PSO algorithm is involved in the process of changing the weights of ANN. The coefficient of determination and mean absolute error are chosen to measure the performance of the performed hybrid PSO-ANN. The numerical results show that hybrid PSO-ANN is applied to estimate SM successfully.

Prediction of pilling of polyester–cotton blended woven fabric using artificial neural network models

In this article, an intelligent pilling prediction model using back-propagation neural network model and an optimized model with genetic algorithm is introduced. Genetic algorithm is proposed in consideration of the initial weight and threshold of back-propagation artificial neural network, and further improves training speed and the accuracy for prediction pilling of polyester–cotton blended woven fabrics. The results show that the maximum numbers of training steps of the optimized model by genetic algorithm are reduced from 164 steps to 137 steps compared with that of back-propagation model. The training fitness of optimized model by genetic algorithm is improved from 0.914 to 0.945. The simulation fitness is increased from 0.912 to 0.987. And the root mean square error decreased from 1.0431 to 0.6842. The optimized model by genetic algorithm shows a better agreement between the experimental and predicted values.

Brain tumor diagnosis based on artificial neural network and a chaos whale optimization algorithm

AbstractAccurate and early detection of the brain tumor region has a great impact on the choice of treatment, its success rate, and the follow‐up of the disease process over time. This study presents a new bioinspired technique for the early detection of the brain tumor area to improve the chance of completely healing. The study presents a multistep technique to detect the brain tumor area. Herein, after image preprocessing and image feature extraction, an artificial neural network is used to determine the tumor area in the image. The method is based on using an improved version of the whale optimization algorithm for optimal selection of the features and optimizing the artificial neural network weights for classification. Simulation results of the proposed method are applied to FLAIR, T1, and T2 datasets and are compared with different algorithms. Three performance indexes including correct detection rate, false acceptance rate, and false rejection rate are selected for the system performance analysis. Final results showed the superiority of the proposed method toward the other similar methods.

Design of artificial neural network using particle swarm optimisation for automotive spring durability

This paper presents the optimisation of spring fatigue life based on an artificial neural network (ANN) architecture and particle swarm optimisation algorithm (PSO) using ISO 2631 vertical vibration as input. The road-induced vibration of a ground vehicle caused the spring to fail due to fatigue and human discomfort. Hence, there is a need to model the relationship between these two parameters for spring design assistance. Vibration and force signals were extracted from a quarter car model simulation for fatigue life and ISO 2631 vertical vibration estimations. PSO was applied to the datasets for ANN weights and biases adjustments while the mean squared error (MSE) was set as the objective function. For validation purposes, a set of independent datasets was applied to the ANN. The residuals were analysed using Lilliefors normality and error histogram. For prediction accuracy, the predicted fatigue lives were analysed using scatter band approach and compared with traditional trained ANN. The results have shown that most of the PSO-based ANN predicted fatigue lives were in the acceptable region and the root mean square error (RMSE) value of 0.6391 life cycles in natural logarithm was obtained. The PSO-based ANN has shown improved performance compared to the conventional ANN approach in predicting fatigue life.

Aging Assessment of Power Transformer Insulation Oil using Hybrid Meta-Heuristic Trained Artificial Neural Network

In today's economic state, power transformer remains as the most expensive equipment in electrical system, in which insulation oil has been taken a significant role for performing a prominent operation. Since the insulation oil happens to degrade soon due to aging, high temperature and chemical reactions such as the oxidation, the periodic checking of oil followed by its replacement is necessary to stop the unexpected failure of the transformer. Moreover, it will be very advantageous if it happens to implement an automated model for predicting the age of transformer oil from time to time. The main intent of this paper is to develop an age assessment framework of transformer insulation oil using intelligent approaches. Here, diverse parameters associated with the transformer such as Breakdown Voltage (BDV), moisture, resistivity, tan delta, interfacial tension, and flash point is given as input for predicting the age of the insulation oil. These data have been already collected using 20 working power transformers operated at various substations in Punjab, India. In the proposed model, the collected parameters are subjected to a well-performing machine learning algorithm termed as Artificial Neural Network (ANN) in order to predict the age of the insulation oil. As a main contribution, the existing training algorithm in ANN so called as Levenberg–Marquardt (LM) is replaced by a hybrid meta-heuritics algorithm. The newly developed hybrid algorithm merges the idea of Crow Search Algorithm (CSA), and Particle Swarm Optimization (PSO), and the new algorithm is termed as Particle Swarm-based Crow Search Algorithm (PS-CSA). The new training algorithm optimizes the weight of ANN using the hybrid CS-PSO updating procedure, in such a way that the difference between the predicted and actual outcome is minimum. Hence, this age prediction of transformer insulation oil will be beneficial for the environs to avoid the drastic condition.

The Concept of Synapse against Popular Opinion

Emulation of the operation process in the human brain was performed by Artificial Neural Network (ANN). The new comments of this study with the concept of progressive tense like an action to new Comparison ANN with Biological Neuron Network were stated against popular opinions. However, their opinions just pointed out the role of Synapse as the weights in the framework of ANN. In this paper, another concept was better suggested. The role of Synapse should be treated as the weights of ANN, which connect two neurons of two hidden different layers. There was a new proposed opinion in this study when an accurate ANN model was built with optimal weights. The role of Synapse should be both the converting the action potential into electrical energy and chemical energy and synaptic strengthening corresponding to long-term potentiation (LTP) in Biological Neuron Network. From the concept of pharmacology, the action of updating weights with optimal values after training more data, was similar as keeping a normal converting for LTP just using medicaments for resisting some ageing brain diseases e.g. Dementia. The new proposed opinion by comparing both Neural Networks should be reasonable in this study.

Fully-CMOS Multi-Level Embedded Non-Volatile Memory Devices With Reliable Long-Term Retention for Efficient Storage of Neural Network Weights

We present a fully CMOS-compatible multi-level non-volatile memory technology, without any special process cost. It is especially suitable for storing the weights of artificial neural networks on chip with low cost, high density, and high power-efficiency. We use hot carrier injection to program the single-transistor cells, and we conduct charge pumping experiments which identify interfacial traps, rather than bulk oxide traps, as the dominant factor in producing stable $I-V$ shifts. We also derive a new physics-based experimentally verified logarithmic model to explain the rate of interfacial trap generation in large $I-V$ shift regimes where the conventional power-law no longer applies. We fabricate two chips, one using TSMC’s 16 nm FinFET and the other in 28 nm planar, and show the FinFET cells are more favorable for non-volatile memory due to their better channel control. We store multiple levels in each FinFET cell using a “program and check” strategy which sets memory cells’ currents with standard deviations less than $2~{\mu }A$ across a shifting range of over $100~{\mu }A$ . We demonstrate 8 level FinFET cells with extrapolated 10-year charge loss within 10% at 125 °C.

Noise prediction of axial piston pump based on different valve materials using a modified artificial neural network model

In this paper, an alternative method to predict the noise of a submersible Axial Piston Pump (APP) for different valve seat materials is presented. The proposed method is composed of an Artificial Neural Network (ANN) model trained using experimental data and integrated with a hybrid algorithm consists of Cat Swarm Optimization (CSO) and Firefly Algorithm (FA) algorithms. The hybrid CSFA algorithm is used as a subroutine in the ANN model to estimate the ANN weights. The FA is used as local operator to improve the exploitation ability of CSO. The obtained results prove the excellence of the proposed method in predicting the noise of APP considering four different valve seat materials (Polytetrafluoroethylene (PTFE), Polyetheretherketone (PEEK), Aliphatic polyamides (NYLON), and stainless steel (316 L)), five speed levels, and six system pressures. Moreover, the effects of different mechanical properties of the valve seat materials as well as operating conditions (speed and system pressure) have been investigated.

Parallel Programming of an Ionic Floating-Gate Memory Array for Scalable Neuromorphic Computing

Neuromorphic computers could overcome efficiency bottlenecks inherent to conventional computing through parallel programming and read out of artificial neural network weights in a crossbar memory array. However, selective and linear weight updates and <10nA read currents are required for learning that surpasses conventional computing efficiency. We introduce an ionic floating-gate memory array [1] based upon a polymer redox transistor connected to a volatile conductive-bridge memory (CBM). Selective and linear programming of a transistor array is executed in parallel by overcoming the bridging threshold of the CBMs. Synaptic weight readout with currents <10nA is achieved by diluting the conductive polymer in an insulating channel to decrease the conductance. The redox transistors endure > 109 ‘read-write’ operations and support > 1MHz ‘read-write’ frequencies. [1] Fuller et. al. Science, 2019 (in press)

Improving feature location accuracy via paragraph vector tuning

Abstract Context Feature location techniques are still not highly accurate despite advances in the field. Objective This paper aims at investigating the impact of applying different tunings to paragraph vector to the feature location problem. It evaluates the influence of different artificial neural network (ANN) configurations for learning rate and negative sampling loss function in paragraph vectors training. Method The suggested weight configuration relies on the search for an adequate ANN learning rate and an adequate calibration of negative sampling skip-gram mode of the Doc2vec (DV) algorithm. A dataset with 633 feature descriptions, extracted from six open-source Java projects, organized within method granularity, is used for the empirical assessment. Results Our results suggest that feature location techniques benefit from the use of paragraph vector with systematic tuning. We show that an adequate update policy for ANN weights can increase feature location accuracy. An adequate calibration for negative sampling also improved accuracy. We got it with no default values of negative sampling pointed by literature. Moreover, an ensemble of learning rate policies and the use of a tuned DV negative sampling option had overcome state-of-the-art approaches. Conclusions We show evidence of a relationship between hyper-parameter settings and accuracy gain. Modern paragraph vector approaches require adequate calibration to produce better results, and we have improved the accuracy of feature location process with proper tuning.

Machine learning models to predict bottom hole pressure in multi‐phase flow in vertical oil production wells

The precise estimation of a pressure drop in vertical multiphase flowing oil wells plays a crucial role in designing robust production facilities and evaluating optimum production plans. A significant amount of research has been conducted on determining a pressure drop via calculating the bottom hole pressure (BHP); these different methods as well as numerical, analytical, semi‐analytical, and empirical correlations can be used for doing this task. Unfortunately, most of those correlations are unable to provide reasonable precision when calculating BHP and, consequently, improvements are still required. To predict BHP in vertical wells, several hybrids of meta‐heuristic optimization methods and a bio‐inspired connectionist approach, i.e., artificial neural network (ANN), are employed. The main goal of these optimization algorithms is to optimize the parameters of the ANN models, i.e., weights and biases, to improve their performance. Based on the obtained outputs and various robustness indexes, a hybrid genetic algorithm and particle swarm optimization (HGAPSO) is highly precise and has a maximum of 10 % error compared to measured pressure data. The results of this work reveal the capability of those hybrid connectionist models for predicting the BHP of multi‐phase flow in vertical wells; these results demonstrate the promising use of connectionist methods in commercial production software in the near future.

SIMULATION OF A TRAINED TRAINED NEURAL NETWORK OF A FUZZY L FUZZY LOGIC REGUL OGIC REGULATION SY TION SYSTEM BASED ON THE CO STEM BASED ON THE COTTON DRYING PROCESS

The article discusses the modeling of a fuzzy-logical system of regulation of the process of drying of raw cotton. The tasks of overcoming uncertainties arising in the process of operation of technological units at the enterprises of the cotton-cleaning industry are presented. An example of solving such a problem by using an artificial neural network is given. Mathematical models based on the neural network have been developed that are used to formalize the process of drying raw cotton and determine the optimal tuned parameters of the fuzzy-logical PID controller, allowing the fate of changing the operating modes of the technological units of the drying drum. A method for determining the number of synoptic weights of artificial neural networks is proposed, which minimizes the number of trainings and increases the speed of management decisions. To train the neural network weights use the reverse spreading error method. The range of variation of the regulator parameter is justified, taking into account the features of the cotton drying process. As a result, the proposed model was used in the control system of the drying process in terms of quality indicators, which led to an increase in the accuracy of the technological process.