• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.33B no.4
• /
• pp.191-200
• /
• 1996

This paper presetns a new fuzzy neural network for fuzzy modeling.The fuzzy neural network is composed of 4 layers and then odes of each layer represent the each step of the if-then fuzzy inference. A heuristic based on the back-propagation algorithm is proposed to ajdust the parameters of the fuzzy nerual network. We prove the feasibility of the network using the experiments on modeling a nonlinear mathematical system and the comparison with previous research.

• Communications for Statistical Applications and Methods
• /
• v.4 no.3
• /
• pp.827-832
• /
• 1997

Statistical relations between a system and empirical distribution are studied in terms of the concept of Akaike's Information Criterion. From this consideration we derive a bootstrap criterion for determining the optimal number of hidden units in nerual networks.

• Proceedings of the Korea Society for Industrial Systems Conference
• /
• 1996.10a
• /
• pp.211-216
• /
• 1996

This paper propose a fuzzy regression method using fuzzy neural networks when a membership value is attached to each input-output pair. First, an architecture of fuzzy nerual networks with fuzzy weights and fuzzy biases is shown. Next a cost function is defined using the fuzzy output from the fuzzy neural network and the corresponding target output with a membership value.A learning algorithm is derived from the cost function. The derived learning algorithm trains the fuzzy neural network so that the level set of the fuzzy output includes the target output. Last, the proposed method is applied to the quality evaluation problem of injection molding.

• Journal of the Korean Society for Precision Engineering
• /
• v.12 no.6
• /
• pp.27-36
• /
• 1995

This paper covers algorithms to determine a machine assignment strategy to locations on a single straight track by minimizing the total backtrack distance. Three different algorithms ar presented: an efficient heuristic procedure, the branch-and-bound algorithm, and the nerual network approach. Simulation results show that the proposed algorithms have potential power to design an on-line optimizer.

• The Journal of the Acoustical Society of Korea
• /
• v.40 no.2
• /
• pp.176-182
• /
• 2021

This paper evaluates and compares the performance of the Deep Nerual Network (DNN)-based speech enhancement models according to various loss functions. We used a complex network that can consider the phase information of speech as a baseline model. As the loss function, we consider two types of basic loss functions; the Mean Squared Error (MSE) and the Scale-Invariant Source-to-Noise Ratio (SI-SNR), and two types of perceptual-based loss functions, including the Perceptual Metric for Speech Quality Evaluation (PMSQE) and the Log Mel Spectra (LMS). The performance comparison was performed through objective evaluation and listening tests with outputs obtained using various combinations of the loss functions. Test results show that when a perceptual-based loss function was combined with MSE or SI-SNR, the overall performance is improved, and the perceptual-based loss functions, even exhibiting lower objective scores showed better performance in the listening test.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.11 no.8
• /
• pp.701-708
• /
• 2001

Recently, the research on intelligence of computer has actively been under way in various areas and gradually extended to adapt to uncertain and complex environments. In this paper, we propose the MVL-Neural Valued Logic. Also, we verify that the MVL-Automata can be implemented to Neural Network and the MVL-Neural Network Model can be a simulator by MVL-Automata. Therefore, we propose that the MVL-Neural Network Model can be widely used in such area, as intelligent system or modeling of brain. In particular, the MVL-Neural Network is expected to be used as core technology of next generation computer.

• Journal of the Korean Institute of Navigation
• /
• v.22 no.2
• /
• pp.47-52
• /
• 1998

Recently , there have been considerable researches about the fusion of fuzzy logic and neural networks. The propose of thise researches is to combine the advantages of both. After the function of approximation using GMDP (Generalized Multi-Denderite Product)neural network for defuzzification operation of fuzzy controller, a new fuzzy-neural network is proposed. Fuzzy membership function of the proposed fuzzy-neural network can be adjusted by learning in order to be adaptive to the variations of a parameter or the external environment. To show the applicability of the proposed fuzzy-nerual network, the proposed model is applied to a speed control o fDC sevo motor. By the hardware implementation, we obtained the desriable results.

• Journal of the Korean Society for Precision Engineering
• /
• v.10 no.3
• /
• pp.141-151
• /
• 1993

For navigation of a mobile robot, it is one of the essential tasks to find out its current position. Dead reckonining is the most frequently used method to estimate its position. Hpwever conventional dead reckoner is prone to give us false information on the robot position especially when the wheels are slipping. This paper proposes an improved dead reckoning scheme using neural networks. The network detects the instance of wheel slopping and estimates the linear velocity of the wheel; thus it calculates current position and heading angle of a mobile robot. The structure and variables of the nerual network are chosen in consideration of slip motion characteristics. A series of experiments are performed to train the networks and to investigate the performance of the improved dead reckoning system.

• The Journal of the Convergence on Culture Technology
• /
• v.3 no.4
• /
• pp.165-169
• /
• 2017

CNN(Convolutional Nerual Network) is one of the algorithms that show superior performance in image recognition and classification among machine learning algorithms. CNN is simple, but it has a large amount of computation and it takes a lot of time. Consequently, in this paper we performed an parallel processing unit for the convolution layer, pooling layer and the fully connected layer, which consumes a lot of handling time in the process of CNN, through the SIMT(Single Instruction Multiple Thread)'s structure of GPGPU(General-Purpose computing on Graphics Processing Units).And we also expect to improve performance by reducing the number of memory accesses and directly using the output of convolution layer not storing it in pooling layer. In this paper, we use MNIST dataset to verify this experiment and confirm that the proposed CNN structure is 12.38% better than existing structure.

• 제어로봇시스템학회:학술대회논문집
• /
• 1997.10a
• /
• pp.1617-1620
• /
• 1997

This paper describes the camera caliberation based-neural network with a camera modeling that accounts for major sources of camera distortion, namely, radial, decentering, and thin prism distortion. Radial distoriton causes an inward or outward displacement of a given image point from its ideal location. Actual optical systems are subject to various degrees of decentering, that is the optical centers of lens elements are not strictly collinear. Thin prism distortion arises from imperfection in lens design and manufacturing as well as camera assembly. It is our purpose to develop the vision system for the pattern recognition and the automatic test of parts and to apply the line of manufacturing. The performance of proposed camera aclibration is illustrated by simulation and experiment.