• 말소리와 음성과학
• /
• 제8권2호
• /
• pp.23-30
• /
• 2016

This paper describes some implementation schemes of CNN in view of mini-batch DNN training for efficient second order optimization. This uses same procedure updating parameters of DNN to train parameters of CNN by simply arranging an input image as a sequence of local patches, which is actually equivalent with mini-batch DNN training. Through this conversion, second order optimization providing higher performance can be simply conducted to train the parameters of CNN. In both results of image recognition on MNIST DB and syllable automatic speech recognition, our proposed scheme for CNN implementation shows better performance than one based on DNN.

• Journal of the Korean Statistical Society
• /
• 제30권4호
• /
• pp.585-595
• /
• 2001

A second-order response surface model is often used to approximate the relationship between a response factor and a set of explanatory factors. In this article, we deal with canonical analysis in response surface models. For the interpretation of the geometry of second-order response surface model, standard errors and confidence intervals for the eigenvalues of the second-order coefficient matrix play an important role. If the confidence interval for some eigenvalue includes 0 or the estimate of some eigenvalue is very small (near to 0) with respect to other eigenvalues, then we are able to delete the corresponding canonical factor. We propose a formulation of criterion which can be used to select canonical factors. This criterion is based on the IMSE(=Integrated Mean Squared Error). As a result of this method, we may approximately write the canonical factors as a set of some important explanatory factors.

• Communications for Statistical Applications and Methods
• /
• 제18권2호
• /
• pp.189-199
• /
• 2011

In this paper, we develop the noninformative priors for the common intraclass correlation coefficient when independent samples drawn from multivariate normal populations. We derive the first and second order matching priors. We reveal that the second order matching prior dose not match alternative coverage probabilities up to the second order and is not a HPD matching prior. It turns out that among all of the reference priors, one-at-a-time reference prior satisfies a second order matching criterion. Our simulation study indicates that one-at-a-time reference prior performs better than the other reference priors in terms of matching the target coverage probabilities in a frequentist sense.

• Journal of Mechanical Science and Technology
• /
• 제20권10호
• /
• pp.1662-1669
• /
• 2006

Since randomness and uncertainties of design parameters are inherent, the robust design has gained an ever increasing importance in mechanical engineering. The robustness is assessed by the measure of performance variability around mean value, which is called as standard deviation. Hence, constraints in robust optimization problem can be approached as probability constraints in reliability based optimization. Then, the FOSM (first order second moment) method or the AFOSM (advanced first order second moment) method can be used to calculate the mean values and the standard deviations of functions describing constraints and object. Among two methods, AFOSM method has some advantage over FOSM method in evaluation of probability. Nevertheless, it is difficult to obtain the mean value and the standard deviation of objective function using AFOSM method, because it requires that the mean value of function is always positive. This paper presented a special technique to overcome this weakness of AFOSM method. The mean value and the standard deviation of objective function by the proposed method are reliable as shown in examples compared with results by FOSM method.

• 한국정밀공학회:학술대회논문집
• /
• 한국정밀공학회 2004년도 춘계학술대회 논문요약집
• /
• pp.284-284
• /
• 2004

기존의 모빌리티 분석 방법은 크게 Grubler의 모빌리티 공식에 의한 zeroth-order 모빌리티 분석 방법, 관절 스크류를 활용한 일차 기구학 성질을 관절 스크류로서 활용한 first-order 모빌리티 분석 방법, 그리고 이차 기구학 특성을 활용하는 second-order 모빌리티 분석 방법으로 구분된다. 그러나, 많은 과다 구속 메커니즘의 경우 zeroth-order 또는 first-order 모빌리티 분석 방법에 의해서는 모빌리티의 분석이 불가능하며 현재까지 문헌에 소개된 second-order 모빌리티 분석 방법은 수치적인 방법에 의존하여 전체 메카니즘의 이차 기구학 특성을 분석하게 되므로 매우 복잡하여 활용하는데 매우 어려움이 있다.(중략)

• 대한수학회보
• /
• 제42권1호
• /
• pp.203-211
• /
• 2005

If a meromorphic solution of second order homogeneous linear differential equation is factorizable, then the right factor of the factorization of the solution has order not more than the coefficient's. And some asympotic properties of solutions are studied.

• 대한수학회보
• /
• 제40권2호
• /
• pp.235-241
• /
• 2003

It is proved that the product of any two linearly independent meromorphic solutions of second order linear differential equations with coefficients of small lower growth must have infinite exponent of convergence of its zero-sequences, under some suitable conditions.

• Journal of the Korean Society for Industrial and Applied Mathematics
• /
• 제5권2호
• /
• pp.1-6
• /
• 2001

Multiplicity of nonlinear second order nonlinear ordinary differential equations will be discussed.

• Ocean Systems Engineering
• /
• 제9권1호
• /
• pp.97-109
• /
• 2019

Accurate calculation of the mean drift forces and moments is necessary when studying the higher order excitations on the floater in waves. When taking the time average of the second order forces and moments, the second order potential and motion diminish with only the first order terms remained. However, in the results of the first order forces or motions, the irregular frequency effects are often observed in higher frequencies, which will affect the accuracy of the calculation of the second order forces and moments. Therefore, we need to pay close attention to the irregular frequency effects in the mean drift forces. This paper will discuss about the irregular frequency effects in the calculations of the mean drift forces and validate our in-house program MDL Multi DYN using some examples which are known to have irregular frequency effects. Finally, we prove that it is necessary to remove the effects and demonstrate that the effectiveness of the formula and methods adopted in the development of our program.

• Progress in Superconductivity
• /
• 제4권2호
• /
• pp.109-113
• /
• 2003

We have developed a planar-type single layer second-order $high-T_{c}$ SQUID gradiometer, which can detect the $d^2$$B_{z/}$dxdy of the second-order field gradient. This SQUID gradiometer consists of four-way 'clover-leaf' pick-up loops and is coupled directly to a 4-junction dc SQUID in such a way that the coupling polarity of the two diagonal loops is opposite to that of the other two loops. The pickup loops are intrinsically balanced for both uniform field and the 1 st-order field gradient. The $YBa_2$$Cu_3$$O_{7}$ thin film was made by pulsed laser deposition method on $SrTiO_3$ single crystal substrate and patterned by photolithography with Ar ion milling technique. Response of this gradiometer was tested for both uniform field and the 2nd-order field gradient. Details of the design, fabrication, and results will be discussed.