• The Journal of the Korea Contents Association
• /
• v.21 no.3
• /
• pp.616-625
• /
• 2021

Deep neural networks are an approximation method that approximates an arbitrary function to a linear model and then repeats additional approximation using a nonlinear active function. In this process, the method of evaluating the performance of approximation uses the loss function. Existing in-depth learning methods implement approximation that takes into account loss functions in the linear approximation process, but non-linear approximation phases that use active functions use non-linear transformation that is not related to reduction of loss functions of loss. This study proposes parametric activation functions that introduce scale parameters that can change the scale of activation functions and location parameters that can change the location of activation functions. By introducing parametric activation functions based on scale and location parameters, the performance of nonlinear approximation using activation functions can be improved. The scale and location parameters in each hidden layer can improve the performance of the deep neural network by determining parameters that minimize the loss function value through the learning process using the primary differential coefficient of the loss function for the parameters in the backpropagation. Through MNIST classification problems and XOR problems, parametric activation functions have been found to have superior performance over existing activation functions.

• Korean journal of food and cookery science
• /
• v.8 no.4
• /
• pp.371-378
• /
• 1992

This experiment was carried out in order to investigate effect of rice varieties, milling methods and particle sizes of rice flours on characters of songe cake. The expansion height of batters peaked from in 14 to 18 minutes during baking for 34 minutes at $150^{\circ}C$, and then was gradually decreased. The decrease in the height ranged from 6 to 50% compared to the expansion peak of batters at one day after baking. The inner temperature of the batters showed a tendency to increase slowly and was greatly increased for 8 to 10 minutes. The highest temperature was 94 to $96^{\circ}C$ in 24 to 26 minutes during baking. The volume of the rice cake was not affected by the particle sizes of the rice flours and milling treatment, and ranged from 89 to 95% compared the volume of wheat cake. Therefore, this result indicated the great possibility baking sponge cake used with rice flours. The baked sponge cake by the hydrated flours showed lower stregth than that by the non-hydrated rice flours without correlation with kinds and particle sizes of rice flours, and there were not significant differences in the cohesiveneness of the cakes. The sponge cake of the hydrated rice flour of Taebaek variety was generally superior in texture and taste to the cake of wheat flour as the result of sensory evaluation.

• Computational Structural Engineering
• /
• v.8 no.4
• /
• pp.129-136
• /
• 1995

This paper presents a boundary element method for obtaining approximate solutions of 2-dimensional consolidation problems based on the Biot's linear theory. Laplace transform is applied to differential equation system in order to eliminate the time dependency. The boundary integral equations in transformed space are formulated and the fundamental solutions are shown in a closed form. In order to convert the transformed solutions to the ones in real space, the Hosono's numerical Laplace transform inversion method is applied. As a numerical example, a half-space consolidation problem subjected to a strip local load is selected and the applicability of the method is demonstrated through the comparison with the exact solutions.

• Computational Structural Engineering
• /
• v.3 no.1
• /
• pp.83-95
• /
• 1990

Many structural optimization problems are solved by numerical algorithms since these are complicated and nonlinear. To provide a wider base and popular it to structual design optimization, reliable, accurate and superlinearly convergent nonlinear programming algorithm with active-set strategy have been developed. One of these is RQP(recursive quadratic programming method). This algorithm has several parameters and its performance is influenced by variations of these key parameters. Therefore, an RQP algorithm is selected to enhance its numerical performances by choosing proper parameters. The paper persents these influences on its numerical performance. For comparison of performances, a structural design software for minimum weight of truss subjected to displacement, stress, and lower and upper bounds on design variables is also implemented.

• Geophysics and Geophysical Exploration
• /
• v.25 no.1
• /
• pp.44-49
• /
• 2022

Closed-form expressions of vector gravity and gravity gradient tensor based on a line segment are derived. If a cylindrical object with axial symmetry is observed from a distance, it is possible to approximate it as a line segment; therefore, it is necessary to compute the gravity and the gravity gradient tensor due to a line source by using closed-form expressions. The gravitational potential for a line segment is defined as a one-dimensional integral, and this integral is differentiated with respect to the Cartesian coordinate system to derive the vector gravity. The expressions of the gravity gradient tensor are derived by differentiating the vector gravity once more in the same coordinate system.

• Proceedings of the Acoustical Society of Korea Conference
• /
• spring
• /
• pp.129-132
• /
• 1999

스테레오 음향 반향 제거기는 입력 신호로 사용되는 두 채널간의 강한 상관관계로 인하여 수렴특성이 악화되는 문제점을 갖는다. 따라서 수렴속도를 향상시키기 위해 RLS또는 Affine Projection(AP) 알고리즘 같은 Least Square (LS)계 열의적응 알고리즘을 사용하는 것이 필요하다. 그러나, 이러한 알고리즘은 LMS 알고리즘과 같은 통계적 미분계열 알고리 즘에 비하여 과도한 계산량을 요구하므로 고속 알고리즘에 대한 연구가 진행되어왔다. 본 논문에서는 스테레오 환경에서 Gram-Schmidt(GS) 직교화를 이용하여 LMS 알고리즘 수준의 계산량을 갖는 근사화된 AP 알고리즘을 제안하였다. 제안한 알고리즘은 AP 알고리즘의 후행오차 성질을 이용하여 GS 직교화 구조로 구성되며, 계산량 감소를 위해 샘플단위의 재귀적 GS 직교화를 사용하였다. 또한 GS직교화를 스테레오 채널에 적용함으로써 적은 계산량으로 AS 알고리즘과 대등한 수렴 성능을 갖는다.

• Journal of the Korean Magnetics Society
• /
• v.23 no.3
• /
• pp.110-116
• /
• 2013

This paper proposes a sampling-based sensitivity method for dealing with electromagnetic coupled design problems effectively. The black-box modeling technique is basically applied to obtain an optimum regardless of how strong the electromagnetic, thermal and structural analyses are coupled with each other. To achieve this, Kriging surrogate models are produced in a hyper-cubic local window with the center of a current design point. Then design sensitivity values are extracted from the differentiation of basis functions which consist of the models. The proposed method falls under a hybrid optimization method which takes advantages of the sampling-based and the sensitivity-based methods. Owing to the aforementioned feature, the method can be applied even to electromagnetic problems of which the material properties are strongly coupled with thermal or structural outputs. To examine the accuracy and validity of the proposed method, a strongly nonlinear mathematical example and a coil design problem for local induction heating are tested.

• Tunnel and Underground Space
• /
• v.30 no.3
• /
• pp.181-192
• /
• 2020

Pin-on-disk test is a suggested abrasion testing method by ASTM (American Society for Testing and Materials). This briefly illustrated the Archimedean spiral motion of a pin type specimen on a disk. To apply this method to rock cutting tools, a constant linear velocity (CLV) is precisely maintained during the test. We defined the two velocity vectors (RPM and horizontal speed) which connected to the resultatnt velocity. We derived a differential equations for the two parameters under CLV condition. It was difficult to find a exact solution. Previous literatures had been reviewed, and an approximate solution was investigated. We mathematically simulated the result for a certain parameter, and examine the accuracy of the solution.

• Proceedings of the Korean Information Science Society Conference
• /
• 1999.10b
• /
• pp.464-466
• /
• 1999

필기 한글 문자 인식을 위해서는 패턴을 구성하는 획 성분을 분석하는 작업이 필수적이다. 기존 인식 방법들은 세선화와 직선 근사에 기반한 방법을 사용하였다. 하지만 세선화는 필기 패턴을 크게 왜곡하는 단점을 안고 있기 때문에 새로운 방법론의 필요성이 대두되고 있다. 본 논문에서는 필기 한글 패턴의 영역-기반 모양 분해 알고리즘을 제안한다. 외곽선 분석을 이용한 기존의 한 단계 알고리즘의 한계를 지적하고, 이 한계를 극복할 수 있는 두 단계 알고리즘을 기술한다. 첫 번째 단계에서는 우세점을 찾아 B접점과 T접점을 탐지한다. 두 번째 단계에서는 볼록 헐(convex hull) 연산을 적용하여 미분할된 부분에 대해 두 번째 분할 작업을 수행한다. PE92 데이터베이스에 대해 실험 한 결과는 세선화 방법보다 우수함을 보인다.

• Journal of Korean Society of Coastal and Ocean Engineers
• /
• v.8 no.3
• /
• pp.246-255
• /
• 1996

A mild slope equation of parabolic type is derived with the revision of the 2nd order differential term and a new approach for the application to large area is presented. The replacement with long waves can overcome the numerical difficulty due to small waves over the system of large grid sizes. No matter how long the replaced wave length is, the refraction and shoaling are maintained by toeing its own wave speed and group velocity, respectively. However, the diffraction effect is modified by means of Eikonal equation. The developed numerical model was applied to the shoal of Ito and Tanimoto (1972) to yield the satisfactory results.