• 전자공학회논문지B
• /
• 제33B권4호
• /
• pp.209-216
• /
• 1996

In this paper, we propose a learning method of multi-layer perceptrons (MLPs) with 8-bit data precision. The suggested method uses the cross-entropy cost function to remove the slope term of error signal in output layer. To decrease the possibility of overflows, we use 16-bit weighted sum results into the 8-bit data with appropriate range. In the forwared propagation, the range for bit-conversion is determined using the saturation property of sigmoid function. In the backwared propagation, the range for bit-conversion is derived using the probability density function of back-propagated signal. In a simulation study to classify hadwritten digits in the CEDAR database, our method shows similar generalization performance to the error back-propagation learning with 16-bit precision.

• 대한기계학회논문집
• /
• 제16권10호
• /
• pp.1816-1823
• /
• 1992

본 연구에서는 CMAC의 훈련에 관련된 문제점 뿐만 아니라 효율적인 CMAC 훈련 방법의 개발에 관한 문제를 연구하였으며, 학습간섭의 영향을 전혀 받지 않으면서 CM- AC의 학습일반화(learning generalization) 특성을 살린, 일반적으로 응용될 수 있는 이웃간훈련방법을 제안하였다. 이 훈련 방법을 2변수 연속함수를 위한 2차원 CMAC의 훈련모사에 적용하여 전체 입력점 수의 1.3% 정도의 훈련 회수로 그 연속함수의 최대 함수값 1.0에 대해 0.0025의 제곱 평균 제곱근 오차(root mean square error, 이하 RMS error라 함)를 갖는 수준의 훈련성과를 거둘 수 있다.

• 대한수학회지
• /
• 제42권6호
• /
• pp.1121-1136
• /
• 2005

Nonstandard finite difference schemes are considered for a generalization of the Cahn-Hilliard equation with Neumann boundary conditions and the Kuramoto-Sivashinsky equation with periodic boundary conditions, which are of the type $$U_t\;+\;\frac{{\partial}^2}{{\partial}x^2} g(u,\;U_x,\;U_{xx})\;=\;\frac{{\partial}^{\alpha}}{{\partial}x^{\alpha}}f(u,\;u_x),\;{\alpha}\;=\;0,\;1,\;2$$. Stability and error estimate of approximate solutions for the corresponding schemes are obtained using the extended Lax-Richtmyer equivalence theorem. Three examples are provided to apply the nonstandard finite difference schemes.

• 제어로봇시스템학회:학술대회논문집
• /
• 제어로봇시스템학회 1989년도 한국자동제어학술회의논문집; Seoul, Korea; 27-28 Oct. 1989
• /
• pp.653-662
• /
• 1989

Cerebellar Model Arithmetic Controller (CMAC) has been introduced as an adaptive control function generator. CMAC computes control functions referring to a distributed memory table storing functional values rather than by solving equations analytically or numerically. CMAC has a unique mapping structure as a coarse coding and supervisory delta-rule learning property. In this paper, learning aspects and a convergence of the CMAC were investigated. The efficient training algorithms were developed to overcome the limitations caused by the conventional maximum error correction training and to eliminate the accumulated learning error caused by a sequential node training. A nonlinear function generator and a motion generator for a two d.o.f. manipulator were simulated. The efficiency of the various learning algorithms was demonstrated through the cpu time used and the convergence of the rms and maximum errors accumulated during a learning process. A generalization property and a learning effect due to the various gains were simulated. A uniform quantizing method was applied to cope with various ranges of input variables efficiently.

• 대한수학회지
• /
• 제48권3호
• /
• pp.655-667
• /
• 2011

A class of functions involving divided differences of the psi and tri-gamma functions and originating from Kershaw's double inequality are proved to be completely monotonic. As applications of these results, the monotonicity and convexity of a function involving the ratio of two gamma functions and originating from the establishment of the best upper and lower bounds in Kershaw's double inequality are derived, two sharp double inequalities involving ratios of double factorials are recovered, the probability integral or error function is estimated, a double inequality for ratio of the volumes of the unit balls in $\mathbb{R}^{n-1}$ and $\mathbb{R}^n$ respectively is deduced, and a symmetrical upper and lower bounds for the gamma function in terms of the psi function is generalized.

• Communications for Statistical Applications and Methods
• /
• 제20권6호
• /
• pp.455-466
• /
• 2013

In nested case-control studies, the most common way to make inference under a proportional hazards model is the conditional logistic approach of Thomas (1977). Inclusion probability methods are more efficient than the conditional logistic approach of Thomas; however, the epidemiology research community has not accepted the methods as a replacement of the Thomas' method. This paper promotes the inverse probability weighting method originally proposed by Samuelsen (1997) in combination with an approximate jackknife standard error that can be easily computed using existing software. Simulation studies demonstrate that this approach yields valid type 1 errors and greater powers than the conditional logistic approach in nested case-control designs across various sample sizes and magnitudes of the hazard ratios. A generalization of the method is also made to incorporate additional matching and the stratified Cox model. The proposed method is illustrated with data from a cohort of children with Wilm's tumor to study the association between histological signatures and relapses.

• 한국기계연구소 소보
• /
• 통권19호
• /
• pp.93-115
• /
• 1989

Cerebellar Model Arithmetic Controller(CMAC) has been introduced as an adaptive control function generator. CMAC computes control functions referring to a distributed memory table storing functional values rather than by solving equations analytically or numerically. CMAC has a unique mapping structure as a coarse coding and supervisory delta-rule learning property. In this paper, learning aspects and a convergence of the CMAC were investigated. The efficient training algorithms were developed to overcome the limitations caused by the conventional maximum error correction training and to eliminate the accumulated learning error caused by a sequential node training. A nonlinear function generator and a motion generator for a two d. o. f. manipulator were simulated. The efficiency of the various learning algorithms was demonstrated through the cpu time used and the convergence of the rms and maximum errors accumulated during a learning process; A generalization property and a learning effect due to the various gains were simulated. A uniform quantizing method was applied to cope with various ranges of input variables efficiently.

• 정보처리학회논문지D
• /
• 제8D권2호
• /
• pp.175-180
• /
• 2001

본 논문은 미래의 소프크웨어 공장 수나 고장시간 예측 정확성을 얻기 위해, 뉴로-피지 시스템을 이용할 경우 최적의 검증 데이터 할당 비율에 대한 연구이다. 훈련 데이터가 주어졌을 때, 과소 적합과 과잉 적합을 회피하면서 최적의 일반화 능력을 얻기 취해 Early Stopping 방법이 일반적으로 사용되고 있다. 그러나 훈련과 검증 데이터로 얼마나 많은 데이터를 할당갈 것인가는 시행착오법을 이용해 경험적으로 해를 구해야만 하며, 과다한 시간이 소요된다. 최적의 검증 데이터 양을 구하기 위해 규칙 수를 증가시키면서 다양한 검증 데이터 양을 할당하였다. 실험결과 최소의 검증 데이터로도 좋은 예측 능력을 보였다. 이 결과는 뉴로-퍼지 시스템을 소프트웨어 신뢰성 분야에 적용시 실질직언 지침을 제공할 수 있는 것이다.

• 한국농공학회지
• /
• 제32권4호
• /
• pp.27-43
• /
• 1990

This study aimed at developing a generalized model on the estimation of the long - term run - off volume for practical purpose. During the research period of last 3 years( 1986-1988), 3 types of estimation model on the long - term run - off volume(Effective rainfall model, unit hydrograph model and barne's model for dry season) had been developed by the author. In this study, through regressional analysis between determinant factors (bi of effective rainfall model, ai of unit hydrograph model and Wi of barne's model) and catchment characteris- tics(catchment area, distance round the catchment area, massing degree coefficient, river - exte- nsion, river - slope, river - density, infiltration of Watershed) of 11 test case areas by multiple regressional method, a new methodology on the derivation of determinant factors from catchment characteristics in the watershed areas having no hydrological station was developed. Therefore, in the resulting step, estimation equations on run - off volume for practical purpose of which input facor is only rainfall were developed. In the next stage, the derived equations were applied on the Kang - and Namgye - river catchment areas for checking of their goodness. The test results were as follows ; 1. In Kang - river area, average relative estimation errors of 72 hydrographs and of continuous daily run - off volume for 245 days( 1/5/1982 - 31/12) were calculated as 6.09%, 9.58% respectively. 2. In Namgye - river area, average relative estimation errors of 65 hydrographs and of conti- nuous daily run - off volume for 2fl days(5/4/1980-31/12) were 5.68%, 10.5% respectively. In both cases, relative estimation error was averaged as 7.96%, and so, the methodology in this study might be hetter organized than Kaziyama's formula when comparing with the relative error of the latter, 24~54%. However, two case studies cannot be the base materials enough for the full generalization of the model. So, in the future studies, many test case studies of this model should he carries out in the various catchment areas for making its generalization.

• 한국인공지능학회지
• /
• 제5권1호
• /
• pp.18-28
• /
• 2017

Recently, there are many companies that use systems based on artificial intelligence. The accuracy of artificial intelligence depends on the amount of learning data and the appropriate algorithm. However, it is not easy to obtain learning data with a large number of entity. Less data set have large generalization errors due to overfitting. In order to minimize this generalization error, this study proposed DGA which can expect relatively high accuracy even though data with a less data set is applied to machine learning based genetic algorithm to deep learning based dropout. The idea of this paper is to determine the active state of the nodes. Using Gradient about loss function, A new fitness function is defined. Proposed Algorithm DGA is supplementing stochastic inconsistency about Dropout. Also DGA solved problem by the complexity of the fitness function and expression range of the model about Genetic Algorithm As a result of experiments using MNIST data proposed algorithm accuracy is 75.3%. Using only Dropout algorithm accuracy is 41.4%. It is shown that DGA is better than using only dropout.