• Proceedings of the Korea Information Processing Society Conference
• /
• 2023.11a
• /
• pp.442-445
• /
• 2023

최근 딥러닝 및 로봇기술의 발전으로 인해 대량의 데이터를 빠르게 수집하고 처리하는 연구 분야들로 확대되었다. 이와 관련된 한 가지 분야로써 다중 로봇을 이용한 분산학습 연구가 있으며, 이는 단일 에이전트를 이용할 때보다 대량의 데이터를 빠르게 수집 및 처리하는데 용이하다. 본 연구에서는 기존 Distributed Neural Network Optimization (DiNNO) 알고리즘에서 제안한 정적 분산 학습방법과 달리 단계적 분산학습 방법을 새롭게 제안하였으며, 모델 성능을 향상시키기 위해 원시 변수를 근사하는 단계수를 상수로 고정하는 기존의 방식에서 통신회차가 늘어남에 따라 점진적으로 근사 횟수를 높이는 방법을 고안하여 새로운 알고리즘을 제안하였다. 기존 알고리즘과 제안된 알고리즘의 정성 및 정량적 성능 평가를 수행하기 MNIST 분류와 2 차원 평면도 지도화 실험을 수행하였으며, 그 결과 제안된 알고리즘이 기존 DiNNO 알고리즘보다 동일한 통신회차에서 높은 정확도를 보임과 함께 전역 최적점으로 빠르게 수렴하는 것을 입증하였다.

• Computational Structural Engineering
• /
• v.9 no.1
• /
• pp.141-149
• /
• 1996

Efficient approximate reanalysis techniques based on substructuring are presented. In most optimal design problems, the analysis precedure must be repeated many times. In particular, one of the main obstacles in the structural optimization systems is high computational cost and time required for the repeated analysis of large-scale structural systems. The purpose of this paper is to show how to evaluate efficiently the sturctural behavior of new designs using information from the previous ones, instead of the multiple repeated analysis of basic equations for successive modification in the optimal design. The proposed reanalysis method is a combined Taylor series expansion and reduced basis method based on substructuring. Several numerical examples illustrate the effectiveness of the method.

• Journal of the Korean Society of Safety
• /
• v.12 no.3
• /
• pp.120-131
• /
• 1997

This paper presents an approximate reanalysis methods of structures based on substructuring for an effective optimization of large-scale structural systems. In most optimal design procedures the analysis of the structure must be repeated many times. In particular, one of the main obstacles in the optimization of structural systems are involved high computational cost and expended long time in the optimization of large-scale structures. The purpose of this paper is to evaluate efficiently the structural behavior of new designs using information from previous ones, without solving basic equations for successive modification in the optimal design. The proposed reanalysis procedure is combined Taylor series expansions which is a local approximation and reduced basis method which is a global approximation based on substructuring. This technique is to choose each of the terms of Taylor series expansions as the basis vector of reduced basis method in substructuring system which is one of the most effective analysis of large -scale structures. Several numerical examples illustrate the effectiveness of the solution process.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 2006.05a
• /
• pp.164-167
• /
• 2006

본 논문에서는 불확실한 비선형 시스템에 대한 적응 퍼지 슬라이딩 모드 제어기를 설계한다. 불확실한 비선형 시스템에서 발생할 수 있는 파라미터의 변화를 대처하기 위해서 적응 퍼지 이론을 이용하였고, 외란으로 인한 불확실성을 슬라이딩 모드의 제어기를 통해서 해결하였다. 또한 퍼지 튜닝을 통해 슬라이딩 조건을 가변화함으로써 기존의 슬라이딩 모드 제어기에 비해 빠르고 정확하게 추종 가능하도록 제어기의 성능을 향상시킨다. 제안하는 제어기는 정확한 동역학 모델의 구현이 어렵고 복잡한 비선형 시스템에 외란 특성이 우수한 슬라이딩 모드와 실제 시스템을 표현하는 범용 근사자로 유용성이 입증된 퍼지 시스템을 이용하여 간단하고 쉽게 제어할 수 있도록 하였다. Lyapunov이론을 통하여 전역적인 안정화를 보이며, 마지막으로 역진자 시스템에 적용하여 제안된 제어기의 성능을 검증한다.

• Journal of the Korean Institute of Telematics and Electronics C
• /
• v.35C no.12
• /
• pp.85-98
• /
• 1998

This paper proposes a new design method of neuro-FLC by the Lamarckian co-adaptation scheme that incorporates the backpropagation learning into the GA evolution in an attempt to find optimal design parameters (fuzzy rule base and membership functions) of application-specific FLC. The design parameters are determined by evolution and learning in a way that the evolution performs the global search and makes inter-FLC parameter adjustments in order to obtain both the optimal rule base having high covering value and small number of useful fuzzy rules and the optimal membership functions having small approximation error and good control performance while the learning performs the local search and makes intra-FLC parameter adjustments by interacting each FLC with its environment. The proposed co-adaptive design method produces better approximation ability because it includes the backpropagation learning in every generation of GA evolution, shows better control performance because the used COG defuzzifier computes the crisp value accurately, and requires small workspace because the optimization procedure of fuzzy rule base and membership functions is performed concurrently by an integrated fitness function on the same fuzzy partition. Simulation results show that the Lamarckian co-adapted FLC produces the most superior one among the differently generated FLCs in all aspects such as the number of fuzzy rules, the approximation ability, and the control performance.

• Proceedings of the KOR-KST Conference
• /
• 2002.02a
• /
• pp.17-42
• /
• 2002

전통적인 OD조사에 의한 OD추정의 여러 문제점들로 인해 링크관측교통량과 기존OD를 결합해 OD를 추정하고자 하는 연구들이 제시되고 있다. Yang(1995)은 일반화최소자승법을 풀기 위한 IEA와 SAB 알고리즘을 제시하였다. 그러나 두 알고리즘의 문제점은 첫째 실제 OD를 알기가 어렵기 때문에 기존 OD를 중요한 추정기준으로 설정한다는 것으로, 이러한 추정의 종속성으로 인해, 기존 OD와 실제 OD의 차이가 큰 경우 정확한 해를 도출하지 못한다. 두 번째 문제는 통행패턴 추정시 선형근사화를 가정하기 때문에 게임이론적 측면에서 전제로 설정한 완전한 Stackelberg 상황을 구현하지 못한다는 것이다. 이러한 문제점을 피하기 위해서는 기존 OD나 관측교통량의 오차에 일관적인 해도출 기법이 필요하다. OD추정 문제는 본질적으로 비선형이고 비볼록하여 전역해 탐색기법이 필요하기 때문에 전역최적화가 가능한 유전알고리즘을 이용한 OD추정모형(GAM)을 개발하였다. 사례네트워크 분석결과, GAM은 기존 OD의 오차에 대해 크게 종속적이지 않으며 OD구조가 변하는 경우에도 추정이 가능하여, 일반적으로 실제 OD를 알 수 없는(기존OD의 오차가 어느 정도인지를 알 수 없는) 도시부 네트워크에서 신뢰성있는 추정력을 보였다. 또한 기존 OD 추정모형은 비교적 용이하게 차종별로 관측할 수 있는 링크교통량을 차종구분 없이 단일차종으로 이용함으로써, 정보의 손실을 초래하여 결과적으로 모형의 추정력을 저하시켰다. 그렇지만 다차종 링크관측교통량으로부터 다차종 OD 추정연구는 거의 없었으며, 그 결과가 단일차종에 대한 추정결과와 어떻게 다른지에 대한 연구도 전무하였다. 본 연구에서는 유전알고리즘을 이용한 OD 추정모형을 다수단 OD 추정모형(GAMUC)으로 확대하였다. 사례 분석 결과 단일차종 OD추정기법은 심각한 추정오류를 범할 수 있으며, 그 적용성도 낮다는 것을 보였다. 다차종 OD 추정기법이 단일차종 OD 추정기법보다 양호한 추정력을 보였으며, 다차종 기법 중에서는 GAMUC가 IEAMUC보다 우수한 추정력을 보였다.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.38 no.1
• /
• pp.7-18
• /
• 2001

In this paper, we present a new method to generate obstacle-free path by using conformal mapping, when avatar navigates in virtual environment. First, we show that the proposed method generates a path to keep away from a circular obstacle. Then, we show that the method can be extended to an elliptical obstacle and multiple obstacles. For real applications, we combine the proposed local method with a global navigation method using sub-target to generate a global obstacle-free path by which avatar navigates naturally in virtual environment.

• Journal of Digital Contents Society
• /
• v.18 no.8
• /
• pp.1603-1613
• /
• 2017

This paper proposes a method for effectively segmenting the hand region using a widely popular RGB color-based webcam. This performs the empirical preprocessing method four times to remove the noise. First, we use Gaussian smoothing to remove the overall image noise. Next, the RGB image is converted into the HSV and the YCbCr color model, and global fixed binarization is performed based on the statistical value for each color model, and the noise is removed by the bitwise-OR operation. Then, RDP and flood fill algorithms are used to perform contour approximation and inner area fill operations to remove noise. Finally, ROI (hand region) is selected by eliminating noise through morphological operation and determining a threshold value proportional to the image size. This study focuses on the noise reduction and can be used as a base technology of gesture recognition application.

• Journal of Korean Society of Transportation
• /
• v.18 no.6
• /
• pp.89-99
• /
• 2000

To cope with the limits of conventional O-D trip matrix collecting methods, several approaches have been developed. One of them is bilevel Programming method Proposed by Yang(1995), which uses Sensitivity Analysis Based(SAB) algorithm to solve Generalized Least Square(GLS) problem. However, the SAB a1gorithm has revealed two critical short-comings. The first is that when there exists a significant difference between target O-D matrix and true O-D matrix, SAB algorithm may not produce correct solution. This stems from the heavy dependance on the historical O-D information, in special when gravel Patterns are dramatically changed. The second is the assumption of iterative linear approximation to original Problem. Because of the approximation, SAB algorithm has difficulty in converging to Perfect Stackelberg game condition. So as to avoid the Problems. we need a more robust and stable solution method. The main purpose of this Paper is to show the problem of the dependency of Previous models and to Propose an alternative solution method to handle it. The Problem of O-D matrix estimation is intrinsically nonlinear and nonconvex. thus it has multiple solutions. Therefore it is necessary to require a method for searching globa1 solution. In this paper, we develop a solution algorithm combined with genetic algorithm(GA) , which is widely used as probabilistic global searching method To compare the efficiency of the algorithm, SAB algorithm suggested by Yang et al. (1992,1995) is used. From the results of numerical example, the Proposed algorithm is superior to SAB algorithm irrespective of travel patterns.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.29 no.8C
• /
• pp.1175-1183
• /
• 2004

This paper propose a method to trace a moving object based on the information directly obtained from MPEG-2 compressed video stream without decoding process. In the proposed method, the motion flow is constructed from the motion vectors involved in compressed video and then we calculate the amount of pan, tilt, zoom associated with camera operations using generalized Hough transform. The local object motion can be extracted from the motion flow after the compensation with the parameters related to the global camera motion. The moving object is designated initially by a user via bounding box. After then automatic tracking is performed based on the mean-shift algorithm of the motion flows of the object. The proposed method can improve the computation speed because the information is directly obtained from the MPEG-2 compressed video, but the object boundary is limited by blocks rather than pixels.