• The Transactions of The Korean Institute of Electrical Engineers
• /
• v.66 no.12
• /
• pp.1772-1781
• /
• 2017

In this study, the design methodology for alleviating the overfitting problem of Polynomial Neural Networks(PNN) is realized with the aid of two kinds techniques such as L2 regularization and Sum of Squared Coefficients (SSC). The PNN is widely used as a kind of mathematical modeling methods such as the identification of linear system by input/output data and the regression analysis modeling method for prediction problem. PNN is an algorithm that obtains preferred network structure by generating consecutive layers as well as nodes by using a multivariate polynomial subexpression. It has much fewer nodes and more flexible adaptability than existing neural network algorithms. However, such algorithms lead to overfitting problems due to noise sensitivity as well as excessive trainning while generation of successive network layers. To alleviate such overfitting problem and also effectively design its ensuing deep network structure, two techniques are introduced. That is we use the two techniques of both SSC(Sum of Squared Coefficients) and $L_2$ regularization for consecutive generation of each layer's nodes as well as each layer in order to construct the deep PNN structure. The technique of $L_2$ regularization is used for the minimum coefficient estimation by adding penalty term to cost function. $L_2$ regularization is a kind of representative methods of reducing the influence of noise by flattening the solution space and also lessening coefficient size. The technique for the SSC is implemented for the minimization of Sum of Squared Coefficients of polynomial instead of using the square of errors. In the sequel, the overfitting problem of the deep PNN structure is stabilized by the proposed method. This study leads to the possibility of deep network structure design as well as big data processing and also the superiority of the network performance through experiments is shown.

• Journal of Institute of Control, Robotics and Systems
• /
• v.9 no.1
• /
• pp.22-29
• /
• 2003

A piezoelectric actuator yields hysteresis effect due to its composed ferroelectric. Hysteresis nonlinearty is neglected when a piezoelectric actuator moves with short stroke. However when it moves with long stroke and high frequency, the hysteresis nonlinearty can not be neglected. The hysteresis nonlinearty of piezoelectric actuator degrades the control performance in precision position control. In this paper, in order to improve the control performance of piezoelectric actuator, an inverse modeling scheme is proposed to compensate the hysteresis nonlinearty. And feedforward - feedback controller is proposed to give a good tracking performance. The Feedforward controller is an inverse hysteresis model, base on neural network and the feedback control is implemented with PID control. To show the feasibility of the proposed controller and hysteresis modeling, some experiments have been carried out. It is concluded that the proposed control scheme gives good tracking performance.

• Journal of Information Technology Services
• /
• v.16 no.3
• /
• pp.113-125
• /
• 2017

Customers have been affected by others' opinions when they make a purchase. Thanks to the development of technologies, people are sharing their experiences such as reviews or ratings through online or social network services, However, although ratings are intuitive information for others, many reviews include only texts without ratings. Also, because of huge amount of reviews, customers and companies can't read all of them so they are hard to evaluate to a product without ratings. Therefore, in this study, we propose a methodology to predict ratings based on reviews for a product. In a methodology, we first estimate the topic-review matrix using the Latent Dirichlet Allocation technic which is widely used in topic modeling. Next, we predict ratings based on the topic-review matrix using the artificial neural network model which is based on the backpropagation algorithm. Through experiments with actual reviews, we find that our methodology can predict ratings based on customers' reviews. And our methodology performs better with reviews which include certain opinions. As a result, our study can be used for customers and companies that want to know exactly a product with ratings. Moreover, we hope that our study leads to the implementation of future studies that combine machine learning and topic modeling.

• The Korean Journal of Applied Statistics
• /
• v.37 no.2
• /
• pp.139-155
• /
• 2024

This study performed a multi-classification task to classify mesial temporal lobe epilepsy with left hippocampal sclerosis patients (left mTLE), mesial temporal lobe epilepsy with right hippocampal sclerosis (right mTLE), and healthy controls (HC) using magnetoencephalography (MEG) data. We applied various artificial neural networks and compared the results. As a result of modeling with convolutional neural networks (CNN), recurrent neural networks (RNN), and graph neural networks (GNN), the average k-fold accuracy was excellent in the order of CNN-based model, GNN-based model, and RNN-based model. The wall time was excellent in the order of RNN-based model, GNN-based model, and CNN-based model. The graph neural network, which shows good figures in accuracy, performance, and time, and has excellent scalability of network data, is the most suitable model for brain research in the future.

• ETRI Journal
• /
• v.26 no.4
• /
• pp.297-306
• /
• 2004

A plasma is a collection of charged particles and on average is electrically neutral. In fabricating integrated circuits, plasma etching is a key means to transfer a photoresist pattern into an underlayer material. To construct a predictive model of plasma-etching processes, a polynomial neural network (PNN) is applied. This process was characterized by a full factorial experiment, and two attributes modeled are its etch rate and DC bias. According to the number of input variables and type of polynomials to each node, the prediction performance of the PNN was optimized. The various performances of the PNN in diverse environments were compared to three types of statistical regression models and the adaptive network fuzzy inference system (ANFIS). As the demonstrated high-prediction ability in the simulation results shows, the PNN is efficient and much more accurate from the point of view of approximation and prediction abilities.

• 제어로봇시스템학회:학술대회논문집
• /
• 1990.10a
• /
• pp.7-12
• /
• 1990

In this paper, we propose a new manipulator control scheme based on the CMAG neural network. The proposed control consists of two components. The feedforward component is an output of trained CMAC neural network and the feedback component is a modified sliding mode control. The CMAC accepts the position, velocity and acceleration of manipulator as input and outputs two values for the controller : One is the nominal torque used for feedforward compensation(M1 network) and the other is the inertia matrix related information used for the feedback component(M2 network). Since the used control algorithm guarantees the robust trajectory tracking in spite of modeling errors, the CMAC mapping errors due to the memory limitation are little worth consideration.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.30B no.6
• /
• pp.59-66
• /
• 1993

Stereo vision is useful in obtaining three dimensional depth information from two images taken from different view points. Neural network modeling for stereo matching, the key step in stereo vision, is defined by an energy function satisfying with three constraints proposed by Marr and Poggio. Stereo matching is then carried out through the network to find minimum energy corresponding to the optimized solution of the problem. An algorithm for stereo matching using an analog neural network is presented here. The network can reduce errors in initial state an early iteration steps by adoption of continuous sigmoid function in stead of binary state. The experimental results show good matching performance for sparse random dot stereogram and real image.

• Proceedings of the KIEE Conference
• /
• 2001.07d
• /
• pp.2341-2344
• /
• 2001

In recent years the neural network known as a sort of the intelligent control strategy is used as a powerful tool for designing control system since it has learning ability. But it is difficult for neural network controllers to guarantee the stability of control systems. In this paper we try connecting a radial basis function network to an adaptive control strategy. Radial basis function networks are simpler and easier to handle than multilayer perceptrons. We use the radial basis function network to generate control input signals that are similar to the control inputs of adaptive control using linear reparameterization of the robot manipulator. We adopt the saturation function as an auxiliary controller. This paper also proves mathematically the stability of the control system under the existence of disturbances and modeling errors.

• Journal of the Ergonomics Society of Korea
• /
• v.25 no.2
• /
• pp.161-172
• /
• 2006

This study aims to compare and contrast the Kansei modeling methods for building a luxuriousness model that people feel about appearance of mobile phones. For the evaluation based on Kansei engineering approaches, 15 participants were employed to evaluate 18 mobile phones using a questionnaire. The results of evaluation were analyzed to build luxuriousness models through quantification I method, neural network, and decision tree method, respectively. The performance of Kansei modeling methods was compared and contrasted in terms of accuracy and predictability. The result of comparison of modeling methods indicated that model accuracy and predictability was closely related to the number of variables and data size. It was also revealed that quantification I method was the best in terms of model accuracy while decision tree method was the best modeling method with small variance in terms of predictability. However, it was empirically found that quantification I method showed extremely unstable predictability with small number of data. Consequently, it is expected that the research findings of this study might be utilized as a guideline for selecting proper Kansei modeling method.

• Journal of the Korean Society for Precision Engineering
• /
• v.21 no.10
• /
• pp.65-77
• /
• 2004

In this paper, we propose an adaptive mechanism based on immune algorithm and neural network identifier technique. It is also applied fur an autonomous guided vehicle (AGV) system. When the immune algorithm is applied to the PID controller, there exists the case that the plant is damaged due to the abrupt change of PID parameters since the parameters are almost adjusted randomly. To solve this problem, we use the neural network identifier (NNI) technique fur modeling the plant and humoral immune algorithm (HIA) which performs the parameter tuning of the considered model, respectively. After the PID parameters are determined in this off-line manner, these gains are then applied to the plant for the on-line control using an immune adaptive algorithm. Moreover, even though the neural network model may not be accurate enough initially, the weighting parameters are adjusted to be accurate through the on-line fine tuning. Finally, the simulation and experimental result fur the control of steering and speed of AGV system illustrate the validity of the proposed control scheme. These results for the proposed method also show that it has better performance than other conventional controller design methods.