• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.37 no.1
• /
• pp.32-42
• /
• 2000

This paper presents an on-line learning algorithm for sequential construction of generalized radial basis function networks (GRBFNs) to model nonlinear systems from empirical data. The GRBFN, an extended from of standard radial basis function (RBF) networks with constant weights, is an architecture capable of representing nonlinear systems by smoothly integrating local linear models. The proposed learning algorithm has a two-stage learning scheme that performs both structure learning and parameter learning. The structure learning stage constructs the GRBFN model using two construction criteria, based on both training error criterion and Mahalanobis distance criterion, to assign new hidden units and the linear local models for given empirical training data. In the parameter learning stage the network parameters are updated using the gradient descent rule. To evaluate the modeling performance of the proposed algorithm, simulations and their results applied to two well-known benchmarks are discussed.

• Journal of the Korean Institute of Illuminating and Electrical Installation Engineers
• /
• v.23 no.2
• /
• pp.175-181
• /
• 2009

In this paper, it is suggested that the selection method of parameter of Power System Stabilizer(PSS) with robustness in low frequency oscillation for Static VAR Compensator(SVC) using a self tuning fuzzy controller for a synchronous generator excitation and SVC system. The proposed parameter self tuning algorithm of fuzzy controller is based on the steepest decent method using two direction vectors which make error between inference values of fuzzy controller and output values of the specially selected PSS reduce steepestly. Using input-output data pair obtained from PSS, the parameters in antecedent part and in consequent part of fuzzy inference rules are learned and tuned automatically using the proposed steepest decent method.

• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.20 no.6
• /
• pp.313-330
• /
• 2021

In an autonomous driving system equipped with a camera and lidar, depth completion techniques enable dense depth estimation. In particular, using self-supervised learning it is possible to train the depth completion network even without ground truth. In actual autonomous driving, such depth completion should have very short latency as it is the input of other algorithms. So, rather than complicate the network structure to increase the accuracy like previous studies, this paper focuses on network latency. We design a U-Net type network with RegNet encoders optimized for GPU computation. Instead, this paper presents several techniques that can increase accuracy during the process of self-supervised learning. The proposed techniques increase the robustness to unreliable lidar inputs. Also, they improve the depth quality for edge and sky regions based on the semantic information extracted in advance. Our experiments confirm that our model is very lightweight (2.42 ms at 1280x480) but resistant to noise and has qualities close to the latest studies.

• Journal of the Korean Geotechnical Society
• /
• v.16 no.1
• /
• pp.83-97
• /
• 2000

In the present study, artificial neural network based on the multi-layer perceptron is used and an optimum model is chosen through the process of efficiency evaluation in order to develop a system predicting maximum displacements of the earth retaining walls at various excavation stages. By analyzing the measured field data collected at various urban excavation sites in Korea, factors influencing on the behaviors of the excavation wall are examined. Among the measured data collected, reliable data are further selected on the basis of the performance ratio and are used as a data base. Data-based measurements are also utilized for both teaming and verifying the artificial neural network model. The learning is carried out by using the back-propagation algorithm based on the steepest descent method. Finally, to verify a validity of the formulated artificial neural network system, both the magnitude and the occurring position of the maximum horizontal displacement are predicted and compared with measured data at real excavation sites not included in the teaming process.

• Journal of the Economic Geographical Society of Korea
• /
• v.5 no.2
• /
• pp.277-291
• /
• 2002

The purpose of this study is to analyze the characteristics of the innovation system of a theme park, and to suggest the development strategies of a theme park through looking at Everland in Yongin, Korea. Everland had relatively the strong networks with in-house, customers, and suppliers, while it had the weak networks with competitors and universities. The innovation information network is constructed among in-house innovation actors; while, the level of interactive learning is low. So the innovation barriers exist; namely, the insufficiency of information exchange, the lack of roles of intermediate organizations, and the gap of R&D and practices. The cooperation and trust should be accumulated to overcome the barriers of innovations. Therefore, Everland should strengthen the networks with in-house innovation actors, and diffuse the cooperation and trust outwards. To maximize the synergies, Everland should construct the networks of innovation actors in a region (Regional Innovation System). To construct Regional Innovation System, first, Everland should construct the close and horizontal cooperation relationship with related firms, and intensify the innovation capacity through learning by interacting. Second, Everland should diffuse the principle of win-win through cooperation and competition.

• Korean Journal of Cognitive Science
• /
• v.4 no.2
• /
• pp.181-200
• /
• 1994

In the learning phase of multilyer feedforword neural network,there are problems such that local minimum,learning praralysis and slow learning speed when backpropagation algorithm used.To overcome these problems, the genetic algorithm has been used as learing method in the multilayer feedforword neural network instead of backpropagation algorithm.However,because the genetic algorith, does not have any mechanism for fine-tuned local search used in backpropagation method,it takes more time that the genetic algorithm converges to a global optimal solution.In this paper,we suggest a new GA-BP method which provides a fine-tunes local search to the genetic algorithm.GA-BP method uses gradient descent method as one of genetic algorithm's operators such as mutation or crossover.To show the effciency of the developed method,we applied it to the 3-parity bit problem with analysis.

• Proceedings of the KIEE Conference
• /
• 2004.07d
• /
• pp.2613-2615
• /
• 2004

본 논문에서는 광학 디스크 기기의 주기적인 외란인 편심 보상을 위해 웨이블릿 신경 회로망 기반 외란 모델로 구성된 순방향 오차 제거(feedforward error rejection) 방법을 제안한다. 신호 모델링 방법으로 사용되어진 신경 회로망 모델의 단점인 실시간 처리 능력 및 국부 최소치로의 가능성 등을 극복하며 주파수와 시간 영역에서의 우수한 신호 해석 능력을 가진 웨이블릿 변환의 장점을 가진 웨이블릿 신경 회로망을 이용하여 디스크의 외란을 모델링 한다. 웨이블릿 신경회로망은 경사 강하법 (gradient descent method)을 이용하여 학습하며, 본 논문에서 제안한 방법의 효율성을 검증하기 위해 실제 광학 디스크 기기의 외란 데이터를 이용한 컴퓨터 모의 실험을 수행한다.

• Journal of Biomedical Engineering Research
• /
• v.21 no.5
• /
• pp.487-494
• /
• 2000

본 논문에서는 배경잡음이 섞여 있는 QRS 파의 증대를 위해 신경망에 근거한 적응라인증대기(ALE) 적용을 다루고 있다. Elman과 Jordan RNN 구조의 합성형태를 갖는 수정된 완전연결 리커런트 신경망이 ALE의 비션형 적응필터로 사용되고 있다. 신경망 노드사이의 연결계수와 이득, 기울기, 지연과 같은 노드 활성함수의 변수들이 기울기 강하 알고리즘을 사용하여 학습이 반복될 때마다 갱신된다. 수정된 신경망은 먼저 미지의 선형과 비선형 시스템 identification을 수행함으로써 평가하였다. 그리고 미약한 QRS를 증대시키기 위해서 적당한 크기의 잡음과 매우 심한 잡음이 포함된 실제의 ECG 신호를 비선형 신경망 적응필처를 사용하는 ALE에 입력하였다. 수정된 신경망은 시스템 identification에 사용하기가 적합함을 확인하였으며, 시뮬레이션 결과에 의하면 신경망 ALE는 잡음 ECG 신호로부터 QRS 파를 증대를 잘 수행하였다.

• The Journal of Korean Institute of Electromagnetic Engineering and Science
• /
• v.29 no.1
• /
• pp.68-76
• /
• 2018

Target prioritization is necessary for a multifunction radar(MFR) to track an important target and manage the resources of the radar platform efficiently. In this paper, we consider an artificial neural network(ANN) model that calculates the priority of the target. Furthermore, we propose a neural network learning algorithm based on the steepest descent method, which is more suitable for target prioritization by combining the conventional gradient descent method. Several simulation results show that the proposed scheme is much more superior to the traditional neural network model from analyzing the training data accuracy and the output priority relevance of the test scenarios.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.25 no.12
• /
• pp.1749-1755
• /
• 2021

Virtual reality is actively applied in the fields of games, entertainment, communication, sports, and architecture. In particular, many virtual reality-based education systems are being developed in the field of education, creating efficient learning effects. In addition, virtual reality-based education is used in areas such as maintenance, fighter control, medical care, and firefighting as it can maximize the educational effect through the mastery process of the function itself through the curriculum as well as indirect experiences of dangerous situations that are difficult to experience. However, due to technical limitations, lack of contents, and lack of theoretical research, the level of application of military education and training is still insufficient. This paper aim to contribute to the development of a virtual reality-based education system as a military training system by developing a high-quality drop training system applicable to military group descent training, studying key technologies and implementation methods necessary for development.