• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.33B no.4
• /
• pp.209-216
• /
• 1996

In this paper, we propose a learning method of multi-layer perceptrons (MLPs) with 8-bit data precision. The suggested method uses the cross-entropy cost function to remove the slope term of error signal in output layer. To decrease the possibility of overflows, we use 16-bit weighted sum results into the 8-bit data with appropriate range. In the forwared propagation, the range for bit-conversion is determined using the saturation property of sigmoid function. In the backwared propagation, the range for bit-conversion is derived using the probability density function of back-propagated signal. In a simulation study to classify hadwritten digits in the CEDAR database, our method shows similar generalization performance to the error back-propagation learning with 16-bit precision.

• Journal of Korea Multimedia Society
• /
• v.21 no.10
• /
• pp.1142-1149
• /
• 2018

Recently, computer vision application is increasing by using CNN which is one of the deep learning algorithms. However, CNN does not provide perfect classification performance due to gradient vanishing problem. Most of CNN algorithms use an activation function called ReLU to mitigate the gradient vanishing problem. In this study, four activation functions that can replace ReLU were applied to four different structural networks. Experimental results show that ReLU has the lowest performance in accuracy, loss rate, and speed of initial learning convergence from 20 experiments. It is concluded that the optimal activation function varied from network to network but the four activation functions were higher than ReLU.

• Journal of Power System Engineering
• /
• v.2 no.3
• /
• pp.74-82
• /
• 1998

This paper presents a new approach to the design of speed and azimuth control of a mobile robot with two drive wheels. The proposed control scheme uses a Gaussian function as a unit function in the neural-fuzzy network and back propagation algorithm to train the neural-fuzzy network controller in the framework of the specialized learning architecture. It is proposed to a learned controller with two neural-fuzzy networks based on an independent reasoning and a connection net with fixed weights to simplify the neural-fuzzy network. The performance of the proposed controller can be seen by the computer simulation for trajectory tracking of the speed and azimuth of a mobile robot driven by two independent wheels.

• Journal of the Korean Institute of Illuminating and Electrical Installation Engineers
• /
• v.27 no.6
• /
• pp.59-67
• /
• 2013

This paper presents a neural net type process model for enhancing learning compensation function in hot strip finishing rolling mill. Adequate input and output variables of process model are chosen, the proposed model was designed as single layer neural net. Equivalent carbon content, strip thickness and rolling speed are suggested as input variables, and looper's manipulation variable is proposed as output variable. According to simulation result using process data to show the validity of the proposed process model, neural net type process model's outputs give almost similar data to process output under same input conditions.

• 제어로봇시스템학회:학술대회논문집
• /
• 1994.10a
• /
• pp.703-707
• /
• 1994

This paper proposes a method of learning control in DC servomotor using a neural network. First we estimate the pulse transfer function of the servo system with an unknown load, then we determine the best gains of I-PD control system using a neural network. Each time the load changes, its best gains of the I-PD control system is computed by the neural network. And the best gains and its pulse transfer function for the case are stored in the memory. According the increase of the set of gains and its pulse transfer function, the learning control system can afford the most suitable I-PD gains instantly.

• International journal of advanced smart convergence
• /
• v.9 no.3
• /
• pp.183-191
• /
• 2020

Today, taking photos using smartphones has become an essential element of modern people. According to these social changes, modern people need a larger storage capacity, and the number of unnecessary photos has increased. To support the storage, cloud-based photo storage services from various platforms have appeared, and many people are using the services. As the number of photos increases, it is difficult for users to find the photos they want, and it takes a lot of time to organize. In this paper, we propose a cloud-based photo management service that facilitates photo management by classifying photos and recommending unnecessary photos using deep learning. The service provides the function of tagging photos by identifying what the subject is, the function of checking for wrongly taken photos, and the function of recommending similar photos. By using the proposed service, users can easily manage photos and use storage capacity efficiently.

• Journal of the Korean Data and Information Science Society
• /
• v.27 no.3
• /
• pp.827-833
• /
• 2016

In this paper, we propose a deep least squares support vector machine (LS-SVM) for regression problems, which consists of the input layer and the hidden layer. In the hidden layer, LS-SVMs are trained with the original input variables and the perturbed responses. For the final output, the main LS-SVM is trained with the outputs from LS-SVMs of the hidden layer as input variables and the original responses. In contrast to the multilayer neural network (MNN), LS-SVMs in the deep LS-SVM are trained to minimize the penalized objective function. Thus, the learning dynamics of the deep LS-SVM are entirely different from MNN in which all weights and biases are trained to minimize one final error function. When compared to MNN approaches, the deep LS-SVM does not make use of any combination weights, but trains all LS-SVMs in the architecture. Experimental results from real datasets illustrate that the deep LS-SVM significantly outperforms state of the art machine learning methods on regression problems.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 1993.06a
• /
• pp.1054-1057
• /
• 1993

A direct method of fuzzy inference and a fuzzy algorithm with learning function are applied to the steam generator level control of nuclear power plant. The fuzzy controller by use of direct inference can control the steam generator in the entire range of power level. There is a little long response time of fuzzy direct inference controller at low power level. The rule base of fuzzy controller with learning function is divided into two parts. One part of the rule base is provided to level control of steam generator at low power level (0%∼30% of full power). Response time of steam generator level control at low power level with this rule base is shown generator level control at low power level with this rule base is shown to be shorter than that of fuzzy controller with direct inference.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.17 no.8
• /
• pp.846-855
• /
• 1992

In this paper, the decrease multipoint search method and Its learning algorithm for optimum value computatlon method of object function Is proposed. Using this method, the number of evaluation point according to searching time can t)e reduced multipoint of the direct search method by applying the unlivarlate method. And the learning algorithm can reprat the same search method in a new established boundary by using the searched result. In order to Investigate the efficience of algorithm, this method this method is applied to Rosenbrock and Powell, Colvelle function that are Impossible or uncertain in traditional direct search method. And the result of application, the optimum value searching oil every function Is successful. Especially, the algorithm is certified as a good calculation method for producing global(absolute) optimum value.

• The Mathematical Education
• /
• v.49 no.4
• /
• pp.423-436
• /
• 2010

In this paper we discuss the Socratic method from the aspects of subject education and examine the meaning of the method in mathematics education that is the most suitable subject for the realization of the Socratic method. In addition, as a realization of the Socratic method, we conducted a teaching-learning experiment of complex natural exponential function with a 2nd year college student. The results of the experiment are analyzed with the intention of improving instruction of the complex analysis that is one of the college mathematics courses.