Search results for: Feed forward multilayer neural network

Authors: Ali Keshavarzi, Fereydoon Sarmadian

Abstract:

Investigation of soil properties like Cation Exchange Capacity (CEC) plays important roles in study of environmental reaserches as the spatial and temporal variability of this property have been led to development of indirect methods in estimation of this soil characteristic. Pedotransfer functions (PTFs) provide an alternative by estimating soil parameters from more readily available soil data. 70 soil samples were collected from different horizons of 15 soil profiles located in the Ziaran region, Qazvin province, Iran. Then, multivariate regression and neural network model (feedforward back propagation network) were employed to develop a pedotransfer function for predicting soil parameter using easily measurable characteristics of clay and organic carbon. The performance of the multivariate regression and neural network model was evaluated using a test data set. In order to evaluate the models, root mean square error (RMSE) was used. The value of RMSE and R2 derived by ANN model for CEC were 0.47 and 0.94 respectively, while these parameters for multivariate regression model were 0.65 and 0.88 respectively. Results showed that artificial neural network with seven neurons in hidden layer had better performance in predicting soil cation exchange capacity than multivariate regression.

Keywords: Easily measurable characteristics, Feed-forwardback propagation, Pedotransfer functions, CEC.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 2172

Authors: Koray U. Erbas

Abstract:

Neural networks are appealing for many applications since they are able to learn complex non-linear relationships between input and output data. As the number of neurons and layers in a neural network increase, it is possible to represent more complex relationships with automatically extracted features. Nowadays Deep Neural Networks (DNNs) are widely used in Computer Vision problems such as; classification, object detection, segmentation image editing etc. In this work, Facial Emotion Recognition task is performed by proposed Convolutional Neural Network (CNN)-based DNN architecture using FER2013 Dataset. Moreover, the effects of different hyperparameters (activation function, kernel size, initializer, batch size and network size) are investigated and ablation study results for Pooling Layer, Dropout and Batch Normalization are presented.

Keywords: Convolutional Neural Network, Deep Learning, Deep Learning Based FER, Facial Emotion Recognition.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 1307

Authors: Jasmin Velagic, Nedim Osmic, Bakir Lacevic

Abstract:

In this paper the neural network-based controller is designed for motion control of a mobile robot. This paper treats the problems of trajectory following and posture stabilization of the mobile robot with nonholonomic constraints. For this purpose the recurrent neural network with one hidden layer is used. It learns relationship between linear velocities and error positions of the mobile robot. This neural network is trained on-line using the backpropagation optimization algorithm with an adaptive learning rate. The optimization algorithm is performed at each sample time to compute the optimal control inputs. The performance of the proposed system is investigated using a kinematic model of the mobile robot.

Keywords: Mobile robot, kinematic model, neural network, motion control, adaptive learning rate.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 3287

Authors: Suresh Alapati, Sreehari Rao Patri, K. S. R. Krishna Prasad

Abstract:

Anultra-low power capacitor less low-dropout voltage regulator with improved transient response using gain enhanced feed forward path compensation is presented in this paper. It is based on a cascade of a voltage amplifier and a transconductor stage in the feed forward path with regular error amplifier to form a composite gainenhanced feed forward stage. It broadens the gain bandwidth and thus improves the transient response without substantial increase in power consumption. The proposed LDO, designed for a maximum output current of 100 mA in UMC 180 nm, requires a quiescent current of 69 )A. An undershot of 153.79mV for a load current changes from 0mA to 100mA and an overshoot of 196.24mV for current change of 100mA to 0mA. The settling time is approximately 1.1 )s for the output voltage undershooting case. The load regulation is of 2.77 )V/mA at load current of 100mA. Reference voltage is generated by using an accurate band gap reference circuit of 0.8V.The costly features of SOC such as total chip area and power consumption is drastically reduced by the use of only a total compensation capacitance of 6pF while consuming power consumption of 0.096 mW.

Keywords: Capacitor-less LDO, frequency compensation, Transient response, latch, self-biased differential amplifier.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 4022

Authors: P. R. Jayathilaka, K.L. Jayaratne, H.L. Premaratne

Abstract:

This research presented in this paper is an on-going project of an application of neural network and fuzzy models to evaluate the sociological factors which affect the educational performance of the students in Sri Lanka. One of its major goals is to prepare the grounds to device a counseling tool which helps these students for a better performance at their examinations, especially at their G.C.E O/L (General Certificate of Education-Ordinary Level) examination. Closely related sociological factors are collected as raw data and the noise of these data are filtered through the fuzzy interface and the supervised neural network is being utilized to recognize the performance patterns against the chosen social factors.

Keywords: Education, Fuzzy, neural network, prediction, Sociology

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 1586

Authors: Gizelle K. Vianna, Gabriel V. Cunha, Gustavo S. Oliveira

Abstract:

Foliage diseases in plants can cause a reduction in both quality and quantity of agricultural production. Intelligent detection of plant diseases is an essential research topic as it may help monitoring large fields of crops by automatically detecting the symptoms of foliage diseases. This work investigates ways to recognize the late blight disease from the analysis of tomato digital images, collected directly from the field. A pair of multilayer perceptron neural network analyzes the digital images, using data from both RGB and HSL color models, and classifies each image pixel. One neural network is responsible for the identification of healthy regions of the tomato leaf, while the other identifies the injured regions. The outputs of both networks are combined to generate the final classification of each pixel from the image and the pixel classes are used to repaint the original tomato images by using a color representation that highlights the injuries on the plant. The new images will have only green, red or black pixels, if they came from healthy or injured portions of the leaf, or from the background of the image, respectively. The system presented an accuracy of 97% in detection and estimation of the level of damage on the tomato leaves caused by late blight.

Keywords: Artificial neural networks, digital image processing, pattern recognition.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 2510

Authors: Yong-Quan Zhou, Deng-Xu He, Zheng Nong

Abstract:

In this paper two models using a functional network were employed to solving classification problem. Functional networks are generalized neural networks, which permit the specification of their initial topology using knowledge about the problem at hand. In this case, and after analyzing the available data and their relations, we systematically discuss a numerical analysis method used for functional network, and apply two functional network models to solving XOR problem. The XOR problem that cannot be solved with two-layered neural network can be solved by two-layered functional network, which reveals a potent computational power of functional networks, and the performance of the proposed model was validated using classification problems.

Keywords: Functional network, neural network, XOR problem, classification, numerical analysis method.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 1265

Authors: Li Kewen, Su Zhaoxin, Wang Xingmou, Zhu Jian Bing

Abstract:

Conventional reservoir prediction methods ar not sufficient to explore the implicit relation between seismic attributes, and thus data utilization is low. In order to improve the predictive classification accuracy of reservoir lithology, this paper proposes a deep learning lithology prediction method based on ResNet (Residual Neural Network) and SENet (Squeeze-and-Excitation Neural Network). The neural network model is built and trained by using seismic attribute data and lithology data of Shengli oilfield, and the nonlinear mapping relationship between seismic attribute and lithology marker is established. The experimental results show that this method can significantly improve the classification effect of reservoir lithology, and the classification accuracy is close to 70%. This study can effectively predict the lithology of undrilled area and provide support for exploration and development.

Keywords: Convolutional neural network, lithology, prediction of reservoir lithology, seismic attributes.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 598

Authors: Anita S. Gangal, P. K. Kalra, D. S. Chauhan

Abstract:

The backpropagation algorithm in general employs quadratic error function. In fact, most of the problems that involve minimization employ the Quadratic error function. With alternative error functions the performance of the optimization scheme can be improved. The new error functions help in suppressing the ill-effects of the outliers and have shown good performance to noise. In this paper we have tried to evaluate and compare the relative performance of complex valued neural network using different error functions. During first simulation for complex XOR gate it is observed that some error functions like Absolute error, Cauchy error function can replace Quadratic error function. In the second simulation it is observed that for some error functions the performance of the complex valued neural network depends on the architecture of the network whereas with few other error functions convergence speed of the network is independent of architecture of the neural network.

Keywords: Complex backpropagation algorithm, complex errorfunctions, complex valued neural network, split activation function.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 2367

Authors: E. Nemati Lay, M. Peymani, E. Sanjari

Abstract:

Prediction of viscosity of natural gas is an important parameter in the energy industries such as natural gas storage and transportation. In this study viscosity of different compositions of natural gas is modeled by using an artificial neural network (ANN) based on back-propagation method. A reliable database including more than 3841 experimental data of viscosity for testing and training of ANN is used. The designed neural network can predict the natural gas viscosity using pseudo-reduced pressure and pseudo-reduced temperature with AARD% of 0.221. The accuracy of designed ANN has been compared to other published empirical models. The comparison indicates that the proposed method can provide accurate results.

Keywords: Artificial neural network, Empirical correlation, Natural gas, Viscosity

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 3188

Authors: A. Chaouachi, R.M. Kamel, R. Ichikawa, H. Hayashi, K. Nagasaka

Abstract:

This paper presents the applicability of artificial neural networks for 24 hour ahead solar power generation forecasting of a 20 kW photovoltaic system, the developed forecasting is suitable for a reliable Microgrid energy management. In total four neural networks were proposed, namely: multi-layred perceptron, radial basis function, recurrent and a neural network ensemble consisting in ensemble of bagged networks. Forecasting reliability of the proposed neural networks was carried out in terms forecasting error performance basing on statistical and graphical methods. The experimental results showed that all the proposed networks achieved an acceptable forecasting accuracy. In term of comparison the neural network ensemble gives the highest precision forecasting comparing to the conventional networks. In fact, each network of the ensemble over-fits to some extent and leads to a diversity which enhances the noise tolerance and the forecasting generalization performance comparing to the conventional networks.

Keywords: Neural network ensemble, Solar power generation, 24 hour forecasting, Comparative study

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 3227

Authors: John Kabuba, Antoine Mulaba-Bafubiandi, Kim Battle

Abstract:

In this study a neural network (NN) was proposed to predict the sorption of binary mixture of copper-cobalt ions into clinoptilolite as ion-exchanger. The configuration of the backpropagation neural network giving the smallest mean square error was three-layer NN with tangent sigmoid transfer function at hidden layer with 10 neurons, linear transfer function at output layer and Levenberg-Marquardt backpropagation training algorithm. Experiments have been carried out in the batch reactor to obtain equilibrium data of the individual sorption and the mixture of coppercobalt ions. The obtained modeling results have shown that the used of neural network has better adjusted the equilibrium data of the binary system when compared with the conventional sorption isotherm models.

Keywords: Adsorption isotherm, binary system, neural network; sorption

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 2000

Authors: Jaswinder Kaur, Satwinder Singh, Karanjeet Singh Kahlon

Abstract:

Accurate software cost estimates are critical to both developers and customers. They can be used for generating request for proposals, contract negotiations, scheduling, monitoring and control. The exact relationship between the attributes of the effort estimation is difficult to establish. A neural network is good at discovering relationships and pattern in the data. So, in this paper a comparative analysis among existing Halstead Model, Walston-Felix Model, Bailey-Basili Model, Doty Model and Neural Network Based Model is performed. Neural Network has outperformed the other considered models. Hence, we proposed Neural Network system as a soft computing approach to model the effort estimation of the software systems.

Keywords: Effort Estimation, Neural Network, Halstead Model, Walston-Felix Model, Bailey-Basili Model, Doty Model.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 2181

Authors: M. Govindarajan, R. M. Chandrasekaran

Abstract:

Text Mining is around applying knowledge discovery techniques to unstructured text is termed knowledge discovery in text (KDT), or Text data mining or Text Mining. In Neural Network that address classification problems, training set, testing set, learning rate are considered as key tasks. That is collection of input/output patterns that are used to train the network and used to assess the network performance, set the rate of adjustments. This paper describes a proposed back propagation neural net classifier that performs cross validation for original Neural Network. In order to reduce the optimization of classification accuracy, training time. The feasibility the benefits of the proposed approach are demonstrated by means of five data sets like contact-lenses, cpu, weather symbolic, Weather, labor-nega-data. It is shown that , compared to exiting neural network, the training time is reduced by more than 10 times faster when the dataset is larger than CPU or the network has many hidden units while accuracy ('percent correct') was the same for all datasets but contact-lences, which is the only one with missing attributes. For contact-lences the accuracy with Proposed Neural Network was in average around 0.3 % less than with the original Neural Network. This algorithm is independent of specify data sets so that many ideas and solutions can be transferred to other classifier paradigms.

Keywords: Back propagation, classification accuracy, textmining, time complexity.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 4164

Authors: Liu Zhiyuan, Sun Zongdi

Abstract:

In this paper, the BP neural network model is established to predict the carbon trading price and carbon trading volume in Shanghai City. First of all, we find the data of carbon trading price and carbon trading volume in Shanghai City from September 30, 2015 to December 23, 2016. The carbon trading price and trading volume data were processed to get the average value of each 5, 10, 20, 30, and 60 carbon trading price and trading volume. Then, these data are used as input of BP neural network model. Finally, after the training of BP neural network, the prediction values of Shanghai carbon trading price and trading volume are obtained, and the model is tested.

Keywords: Carbon trading price, carbon trading volume, BP neural network model, Shanghai City.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 1336

Authors: Kexin Jia, Youxin Lu

Abstract:

To make the modulation classification system more suitable for signals in a wide range of signal to noise rate (SNR), a novel method of designing combined classifier based on fuzzy neural network (FNN) is presented in this paper. The method employs fuzzy neural network classifiers and interclass distance (ICD) to improve recognition reliability. Experimental results show that the proposed combined classifier has high recognition rate with large variation range of SNR (success rates are over 99.9% when SNR is not lower than 5dB).

Keywords: Modulation classification, combined classifier, fuzzy neural network, interclass distance.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 1180

Authors: B. Q. Huang, Tarik Rashid, M-T. Kechadi

Abstract:

this paper presents a multi-context recurrent network for time series analysis. While simple recurrent network (SRN) are very popular among recurrent neural networks, they still have some shortcomings in terms of learning speed and accuracy that need to be addressed. To solve these problems, we proposed a multi-context recurrent network (MCRN) with three different learning algorithms. The performance of this network is evaluated on some real-world application such as handwriting recognition and energy load forecasting. We study the performance of this network and we compared it to a very well established SRN. The experimental results showed that MCRN is very efficient and very well suited to time series analysis and its applications.

Keywords: Gradient descent method, recurrent neural network, learning algorithms, time series, BP

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 2991

Authors: Hossein Esbati, Jalil Shirazi

Abstract:

In this paper, in order to categorize ORL database face pictures, principle Component Analysis (PCA) and Kernel Principal Component Analysis (KPCA) methods by using Elman neural network and Support Vector Machine (SVM) categorization methods are used. Elman network as a recurrent neural network is proposed for modeling storage systems and also it is used for reviewing the effect of using PCA numbers on system categorization precision rate and database pictures categorization time. Categorization stages are conducted with various components numbers and the obtained results of both Elman neural network categorization and support vector machine are compared. In optimum manner 97.41% recognition accuracy is obtained.

Keywords: Face recognition, Principal Component Analysis, Kernel Principal Component Analysis, Neural network, Support Vector Machine.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 1887

Authors: Hajir Karimi, Fakheri Yousefi, Mahmood Reza Rahimi

Abstract:

An accurate and proficient artificial neural network (ANN) based genetic algorithm (GA) is developed for predicting of nanofluids viscosity. A genetic algorithm (GA) is used to optimize the neural network parameters for minimizing the error between the predictive viscosity and the experimental one. The experimental viscosity in two nanofluids Al2O3-H2O and CuO-H2O from 278.15 to 343.15 K and volume fraction up to 15% were used from literature. The result of this study reveals that GA-NN model is outperform to the conventional neural nets in predicting the viscosity of nanofluids with mean absolute relative error of 1.22% and 1.77% for Al2O3-H2O and CuO-H2O, respectively. Furthermore, the results of this work have also been compared with others models. The findings of this work demonstrate that the GA-NN model is an effective method for prediction viscosity of nanofluids and have better accuracy and simplicity compared with the others models.

Keywords: genetic algorithm, nanofluids, neural network, viscosity

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 2034

Authors: M. Khatami Rad, N. Jamali, M. Torabizadeh, A. Noshadi

Abstract:

In this paper, an artificial neural network simulator is employed to carry out diagnosis and prognosis on electric motor as rotating machinery based on predictive maintenance. Vibration data of the primary failed motor including unbalance, misalignment and bearing fault were collected for training the neural network. Neural network training was performed for a variety of inputs and the motor condition was used as the expert training information. The main purpose of applying the neural network as an expert system was to detect the type of failure and applying preventive maintenance. The advantage of this study is for machinery Industries by providing appropriate maintenance that has an essential activity to keep the production process going at all processes in the machinery industry. Proper maintenance is pivotal in order to prevent the possible failures in operating system and increase the availability and effectiveness of a system by analyzing vibration monitoring and developing expert system.

Keywords: Condition based monitoring, expert system, neural network, fault detection, vibration monitoring.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 1943

Authors: Ali Ghiaseddin , Akram Nemati

Abstract:

In a pilot plant scale of a fluidized bed reactor, a reduction reaction of sodium sulfate by natural gas has been investigated. Natural gas is applied in this study as a reductant. Feed density, feed mass flow rate, natural gas and air flow rate (independent parameters)and temperature of bed and CO concentration in inlet and outlet of reactor (dependent parameters) were monitored and recorded at steady state. The residence time was adjusted close to value of traditional reaction [1]. An artificial neural network (ANN) was established to study dependency of yield and carbon gradient on operating parameters. Resultant 97% accuracy of applied ANN is a good prove that natural gas can be used as a reducing agent. Predicted ANN model for relation between other sources carbon gradient (accuracy 74%) indicates there is not a meaningful relation between other sources carbon variation and reduction process which means carbon in granule does not have significant effect on the reaction yield.

Keywords: reduction by natural gas, fluidized bed, sulfate, sulfide, artificial neural network

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 1476

Authors: A. Pajaziti, H. Cana

Abstract:

In this paper, the structural genetic algorithm is used to optimize the neural network to control the joint movements of robotic arm. The robotic arm has also been modeled in 3D and simulated in real-time in MATLAB. It is found that Neural Networks provide a simple and effective way to control the robot tasks. Computer simulation examples are given to illustrate the significance of this method. By combining Genetic Algorithm optimization method and Neural Networks for the given robotic arm with 5 D.O.F. the obtained the results shown that the base joint movements overshooting time without controller was about 0.5 seconds, while with Neural Network controller (optimized with Genetic Algorithm) was about 0.2 seconds, and the population size of 150 gave best results.

Keywords: Robotic Arm, Neural Network, Genetic Algorithm, Optimization.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 3551

Authors: M. Tarafdar Haque, A. Taheri

Abstract:

Application of neural networks in execution of programmed pulse width modulation (PPWM) of a voltage source inverter (VSI) is studied in this paper. Using the proposed method it is possible to cancel out the desired harmonics in output of VSI in addition to control the magnitude of fundamental harmonic, contineously. By checking the non-trained values and a performance index, the most appropriate neural network is proposed. It is shown that neural networks may solve the custom difficulties of practical utilization of PPWM such as large size of memory, complex digital circuits and controlling the magnitude of output voltage in a discrete manner.

Keywords: Neural Network, Inverter, PPWM.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 1646

Authors: H. Soleimanjahi, M. J. Nategh, S. Falahi

Abstract:

Deciding the numerous parameters involved in designing a competent artificial neural network is a complicated task. The existence of several options for selecting an appropriate architecture for neural network adds to this complexity, especially when different applications of heterogeneous natures are concerned. Two completely different applications in engineering and medical science were selected in the present study including prediction of workpiece's surface roughness in ultrasonic-vibration assisted turning and papilloma viruses oncogenicity. Several neural network architectures with different parameters were developed for each application and the results were compared. It was illustrated in this paper that some applications such as the first one mentioned above are apt to be modeled by a single network with sufficient accuracy, whereas others such as the second application can be best modeled by different expert networks for different ranges of output. Development of knowledge about the essentials of neural networks for different applications is regarded as the cornerstone of multidisciplinary network design programs to be developed as a means of reducing inconsistencies and the burden of the user intervention.

Keywords: Artificial Neural Network, Malignancy Diagnosis, Papilloma Viruses Oncogenicity, Surface Roughness, UltrasonicVibration-Assisted Turning.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 1461

Authors: Han-wool Lee, Chin-suk Hong, Weui-bong Jung

Abstract:

In this study, noise characteristics of structure were analyzed in an effort to reduce noise passing through an opening of an enclosure surrounding the structure that generates noise. Enclosures are essential measure to protect noise propagation from operating machinery. Access openings of the enclosures are important path of noise leakage. First, noise characteristics of structure were analyzed and feed-forward noise control was performed using simulation in order to reduce noise passing through the opening of enclosure, which surrounds a structure generating noise. We then implemented a feed-forward controller to actively control the acoustic power through the opening. Finally, we conducted optimization of placement of the reference sensors for several cases of the number of sensors. Good control performances were achieved using the minimum number of microphones arranged an optimal placement.

Keywords: Active Noise Control, Feed-forward Control, Noise Attenuation, Position Optimization.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 1598

Authors: P. Latha, L. Ganesan, N. Ramaraj, P. V. Hari Venkatesh

Abstract:

Motion detection is a basic operation in the selection of significant segments of the video signals. For an effective Human Computer Intelligent Interaction, the computer needs to recognize the motion and track the moving object. Here an efficient neural network system is proposed for motion detection from the static background. This method mainly consists of four parts like Frame Separation, Rough Motion Detection, Network Formation and Training, Object Tracking. This paper can be used to verify real time detections in such a way that it can be used in defense applications, bio-medical applications and robotics. This can also be used for obtaining detection information related to the size, location and direction of motion of moving objects for assessment purposes. The time taken for video tracking by this Neural Network is only few seconds.

Keywords: Frame separation, Correlation Network, Neural network training, Radial Basis Function, object tracking, Motion Detection.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 3104

Authors: Svitov David, Alyamkin Sergey

Abstract:

The usage of convolutional neural networks (CNNs) in conjunction with the margin-based softmax approach demonstrates the state-of-the-art performance for the face recognition problem. Recently, lightweight neural network models trained with the margin-based softmax have been introduced for the face identification task for edge devices. In this paper, we propose a distillation method for lightweight neural network architectures that outperforms other known methods for the face recognition task on LFW, AgeDB-30 and Megaface datasets. The idea of the proposed method is to use class centers from the teacher network for the student network. Then the student network is trained to get the same angles between the class centers and face embeddings predicted by the teacher network.

Keywords: ArcFace, distillation, face recognition, margin-based softmax.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 557

Authors: Zahra Ghadampour, Gholamreza Rakhshandehroo

Abstract:

A concern that researchers usually face in different applications of Artificial Neural Network (ANN) is determination of the size of effective domain in time series. In this paper, trial and error method was used on groundwater depth time series to determine the size of effective domain in the series in an observation well in Union County, New Jersey, U.S. different domains of 20, 40, 60, 80, 100, and 120 preceding day were examined and the 80 days was considered as effective length of the domain. Data sets in different domains were fed to a Feed Forward Back Propagation ANN with one hidden layer and the groundwater depths were forecasted. Root Mean Square Error (RMSE) and the correlation factor (R2) of estimated and observed groundwater depths for all domains were determined. In general, groundwater depth forecast improved, as evidenced by lower RMSEs and higher R2s, when the domain length increased from 20 to 120. However, 80 days was selected as the effective domain because the improvement was less than 1% beyond that. Forecasted ground water depths utilizing measured daily data (set #1) and data averaged over the effective domain (set #2) were compared. It was postulated that more accurate nature of measured daily data was the reason for a better forecast with lower RMSE (0.1027 m compared to 0.255 m) in set #1. However, the size of input data in this set was 80 times the size of input data in set #2; a factor that may increase the computational effort unpredictably. It was concluded that 80 daily data may be successfully utilized to lower the size of input data sets considerably, while maintaining the effective information in the data set.

Keywords: Neural networks, groundwater depth, forecast.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 2471

Authors: Isao Taguchi, Yasuo Sugai

Abstract:

This paper proposes an efficient learning method for the layered neural networks based on the selection of training data and input characteristics of an output layer unit. Comparing to recent neural networks; pulse neural networks, quantum neuro computation, etc, the multilayer network is widely used due to its simple structure. When learning objects are complicated, the problems, such as unsuccessful learning or a significant time required in learning, remain unsolved. Focusing on the input data during the learning stage, we undertook an experiment to identify the data that makes large errors and interferes with the learning process. Our method devides the learning process into several stages. In general, input characteristics to an output layer unit show oscillation during learning process for complicated problems. The multi-stage learning method proposes by the authors for the function approximation problems of classifying learning data in a phased manner, focusing on their learnabilities prior to learning in the multi layered neural network, and demonstrates validity of the multi-stage learning method. Specifically, this paper verifies by computer experiments that both of learning accuracy and learning time are improved of the BP method as a learning rule of the multi-stage learning method. In learning, oscillatory phenomena of a learning curve serve an important role in learning performance. The authors also discuss the occurrence mechanisms of oscillatory phenomena in learning. Furthermore, the authors discuss the reasons that errors of some data remain large value even after learning, observing behaviors during learning.

Keywords: data selection, function approximation problem, multistage leaning, neural network, voluntary oscillation.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 1389

Authors: D. Singh, R. Yousefi, M. Boroushaki

Abstract:

Intelligent deep-drawing is an instrumental research field in sheet metal forming. A set of 28 different experimental data have been employed in this paper, investigating the roles of die radius, punch radius, friction coefficients and drawing ratios for axisymmetric workpieces deep drawing. This paper focuses an evolutionary neural network, specifically, error back propagation in collaboration with genetic algorithm. The neural network encompasses a number of different functional nodes defined through the established principles. The input parameters, i.e., punch radii, die radii, friction coefficients and drawing ratios are set to the network; thereafter, the material outputs at two critical points are accurately calculated. The output of the network is used to establish the best parameters leading to the most uniform thickness in the product via the genetic algorithm. This research achieved satisfactory results based on demonstration of neural networks.

Keywords: Deep-drawing, Neural network, Genetic algorithm, Sheet metal forming.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 2149