• 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.

• The Korean Journal of Applied Statistics
• /
• v.22 no.1
• /
• pp.153-161
• /
• 2009

We consider a logistic model with random effects as the superpopulation for estimating the small area pro-portions. The best linear unbiased predictor under linear mired model is popular in small area estimation. We use this type of estimator under logistic mixed motel for the small area proportions, on which the estimation of mean squared error is also discussed. Two kinds of estimation methods, the parametric bootstrap and the linear approximation will be compared through a Monte Carlo study in the respects of the normality assumption on the random effects distribution and also the magnitude of sample sizes on the approximation.

• Journal of Korean Society of Industrial and Systems Engineering
• /
• v.38 no.3
• /
• pp.108-116
• /
• 2015

A missile defense system is composed of radars detecting incoming missiles aiming at defense assets, command control units making the decisions on weapon target assignment, and artillery batteries firing of defensive weapons to the incoming missiles. Although, the technology behind the development of radars and weapons is very important, effective assignment of the weapons against missile threats is much more crucial. When incoming missile targets toward valuable assets in the defense area are detected, the asset-based weapon target assignment model addresses the issue of weapon assignment to these missiles so as to maximize the total value of surviving assets threatened by them. In this paper, we present a model for an asset-based weapon assignment problem with shoot-look-shoot engagement policy and fixed set-up time between each anti-missile launch from each defense unit. Then, we show detailed linear approximation process for nonlinear portions of the model and propose final linear approximation model. After that, the proposed model is applied to several ballistic missile defense scenarios. In each defense scenario, the number of incoming missiles, the speed and the position of each missile, the number of defense artillery battery, the number of anti-missile in each artillery battery, single shot kill probability of each weapon to each target, value of assets, the air defense coverage are given. After running lpSolveAPI package of R language with the given data in each scenario in a personal computer, we summarize its weapon target assignment results specified with launch order time for each artillery battery. We also show computer processing time to get the result for each scenario.

• Journal of Broadcast Engineering
• /
• v.23 no.2
• /
• pp.302-315
• /
• 2018

Highly non-linear electro-optical transfer function of the Perceptual Quantizer was approximated by a truncated Taylor series, resulting in a closed form solution for luma adjustment. This previous solution is fast and quite suitable for the hardware implementation of luma adjustment, but the approximation error becomes relatively large in the range of 600~3,900 cd/m2 linear light. In order to reduce such approximation error, we propose a new linear model, for which a correction is performed on the position and the slope of line based on the scope of approximation. In order to verify the approximation capability of the proposed linear model, a comparative study on the luma adjustment schemes was conducted using various high dynamic range test video sequences. Via the comparative study, we identified a significant performance enhancement over the previous fast luma adjustment scheme, where a 4.65dB of adjusted luma t-PSNR gain was obtained for a test sequence having a large portion of saturated color pixels.

• Communications of Mathematical Education
• /
• v.23 no.3
• /
• pp.625-642
• /
• 2009

In the school mathematics the concept of tangent is introduced in several steps in suitable contexts. Students are required to reflect and revise their concepts of tangent in order to apply the improved concept to wider range of contexts. In this paper we consider the tangent as the optimal linear approximation to a curve at a given point and make three discussions on pedagogical aspects of it. First, it provides a method of finding roots of real numbers which can be used as an application of tangent. This may help students improve their affective variables such as interest, attitude, motivation about the learning of tangent. Second, this concept reflects the modern point of view of tangent, the linear approximation of nonlinear problems. Third, it gives precise meaning of two tangent lines appearing two sides of a cusp point of a curve.

• Journal of the Korean Mathematical Society
• /
• v.31 no.3
• /
• pp.467-475
• /
• 1994

In 1983 Harmand and Lima [5] proved that if X is a Banach space for which K(X), the space of compact linear operators on X, is an M-ideal in L(X), the space of bounded linear operators on X, then it has the metric compact approximation property. A strong converse of the above result holds if X is a closed subspace of either $\elll_p(1 < p < \infty) or c_0 [2,15]$. In 1979 J. Johnson [7] actually proved that if X is a Banach space with the metric compact approximation property, then the annihilator K(X)^\bot$ of K(X) in $L(X)^*$ is the kernel of a norm-one projection in $L(X)^*$, which is the case if K(X) is an M-ideal in L(X).

• Communications for Statistical Applications and Methods
• /
• v.16 no.6
• /
• pp.971-978
• /
• 2009

Poisson generalized linear mixed models(GLMMs) have been widely used for the analysis of clustered or correlated count data. For the inference marginal likelihood, which is obtained by integrating out random effects is often used. It gives maximum likelihood(ML) estimator, but the integration is usually intractable. In this paper, we propose how to obtain the ML estimator via Laplace approximation based on hierarchical-likelihood (h-likelihood) approach under the Poisson GLMMs. In particular, the h-likelihood avoids the integration itself and gives a statistically efficient procedure for various random-effect models including GLMMs. The proposed method is illustrated using two practical examples and simulation studies.

• Kyungpook Mathematical Journal
• /
• v.58 no.4
• /
• pp.733-746
• /
• 2018

The concept of f-statistical convergence which is, in fact, a generalization of statistical convergence, has been introduced recently by Aizpuru et al. (Quaest. Math. 37: 525-530, 2014). The main object of this paper is to prove an f-statistical analog of the classical Korovkin type approximation theorem of Boyanov and Veselinov. It is shown that the f-statistical analog is intermediate between the classical theorem and its statistical analog. As an application, we estimate the rate of f-statistical convergence of the sequence of positive linear operators defined from $C^*[0,{\infty})$ into itself.

• Bulletin of the Korean Mathematical Society
• /
• v.33 no.3
• /
• pp.367-376
• /
• 1996

We characterize best simultaneous approximations from a finite-dimensional subspace of a normed linear space. In the characterization we reveal usefulness of a minimax theorem presented in [2,4].

• Journal of the Korean Statistical Society
• /
• v.34 no.3
• /
• pp.197-208
• /
• 2005

We consider the problem for approximate solution of linear stochastic evolution equations driven by infinite-dimensional fractional Brownian motion with Hurst parameter $H\;\in$ (0,1/2). The error of the approximate solution for the explicit Euler scheme is investigated.