• Journal of the Korean Institute of Telematics and Electronics
• /
• v.11 no.4
• /
• pp.1.1-11
• /
• 1974

A general and systematic method of organizing two-dimensional flexible cellular array which is capable of reclizing arbitrary combinational switching function is developed. A set of n functions of n variables is transformed to revalued functions of one variable. This set of functions form a semigroup under the normal operation which is defined in this paper. A systematic method of generating any functions using three base functions is presented. Three basic networks which are capable of realizing three base functions are designed using only one one-dimensional array. The algorithm is presented for lealizing arbitrary combinational switching functions by organizing this basic array in two.dimensional cellular array and by appropriately setting the parameters or the edge of the array.

• Proceedings of the KIEE Conference
• /
• 2003.07d
• /
• pp.2576-2578
• /
• 2003

RBF 네트워크에서 기저함수의 위치는 네트워크의 성능에 매우 큰 영향을 미친다. 몇몇 응용들에서 교사학습을 이용한 기저함수의 위치 선정이 비교사학습에 비해 우수함을 보인다. 그러나 교사학습에 의한 네트워크는 시그모이드 네트워크와 같은 긴 학습시간을 필요로 한다. 본 논문에서는 오차함수의 gradient와 Hessian을 이용해 교사학습에서 요구하는 학습시간을 단축시키면서 기저함수의 최적위치를 추정하였다.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 2003.05a
• /
• pp.81-84
• /
• 2003

$\varepsilon$-SVR(e-Support Vector Regression)학습방법은 SV(Support Vector)들을 이용하여 함수 근사(Regression)하는 방법으로 최근 주목받고 있는 기법이다. SVM(SV machine)의 한 가지 방법으로, 신경망을 기반으로 한 다른 알고리즘들이 학습과정에서 지역적 최적해로 수렴하는 등의 문제를 한계로 갖는데 반해, 이러한 구조들을 대체할 수 있는 학습방법으로 사용될 수 있다. 일반적인 $\varepsilon$-SVR에서는 학습 데이터와 관사 함수 f사이에 허용 가능한 에러범위 $\varepsilon$값이 학습하기 전에 정해진다. 그러나 Nu-SVR(ν-version SVR)학습방법은 학습의 결과로 최적화 된 $\varepsilon$값을 얻을 수 있다. 정해진 기저함수가 포함되는 $\varepsilon$-SVR 학습방법(Sermparametric SVR)은 정해진 독립 기저함수를 사용하여 함수를 근사하는 방법으로, 일반적인 $\varepsilon$-SVR 학습방범에 비해 우수한 결과를 나타내는 것이 성공적으로 입증된 바 있다. 이에 따라, 본 논문에서는 정해진 기저함수가 포함된 ν-SVR 학습 방법을 제안하고, 이에 대한 수식을 유도하였다. 그리고, 모의 실험을 통하여 제안된 Sermparametric ν-SVR 학습 방법의 적용 가능성을 알아보았다.

• Journal of the Computational Structural Engineering Institute of Korea
• /
• v.36 no.6
• /
• pp.391-398
• /
• 2023

In this paper, a penalized maximum likelihood estimation (PMLE) method that applies a penalty to increase the accuracy of a basis-screening-based Kriging model (BSKM) is introduced. The maximum order and set of basis functions used in the BSKM are determined according to their importance. In this regard, the cross-validation error (CVE) for the basis functions is employed as an indicator of importance. When constructing the Kriging model (KM), the maximum order of basis functions is determined, the importance of each basis function is evaluated according to the corresponding maximum order, and finally the optimal set of basis functions is determined. This optimal set is created by adding basis functions one by one in order of importance until the CVE of the KM is minimized. In this process, the KM must be generated repeatedly. Simultaneously, hyper-parameters representing correlations between datasets must be calculated through the maximum likelihood evaluation method. Given that the optimal set of basis functions depends on such hyper-parameters, it has a significant impact on the accuracy of the KM. The PMLE method is applied to accurately calculate hyper-parameters. It was confirmed that the accuracy of a BSKM can be improved by applying it to Branin-Hoo problem.

• Journal of the Korean Institute of Telematics and Electronics
• /
• v.22 no.5
• /
• pp.83-86
• /
• 1985

An alternative technique (or the numerical solution of Fredholm integral equations of second kind is presented. The approximate solution is obtained by fitting the data in mixed form at knots in the region of the problem. To decrease the error in the numerical solution, cubic B-spline functions which are twice continuously differentiable at knots are employed as basis function. For a given example, the results of this technique are compared with those of Moment method employing pulse functions for basis function and delta functions for test function and found to br in good agreement.

• Transactions of the Korean Society for Noise and Vibration Engineering
• /
• v.17 no.7 s.124
• /
• pp.638-648
• /
• 2007

A principal components analysis of the entire median HRIRs in the CIPIC HRTF database reveals that the individual HRIRs can be adequately reconstructed by a linear combination of several orthonormal basis functions. The basis functions represent the inter-individual and inter-elevation variations in median HRIRs. There exist elevation-dependent tendencies in the weights of basis functions, and the basis functions can be ordered according to the magnitude of standard deviation of the weights at each elevation. We propose a HRIR customization method via tuning of the weights of 3 dominant basis functions corresponding to the 3 largest standard deviations at each elevation. Subjective listening test results show that both front-back reversal and vertical perception can be improved with the customized HRIRs.

• The KIPS Transactions:PartB
• /
• v.11B no.3
• /
• pp.303-308
• /
• 2004

The Gabor cosine and sine transform can be applied to image and video compression algorithm by representing image frequency components locally The computational complexity of forward and inverse matrix transforms used in the compression and decompression requires O($N^3$)operations. In this paper, the length of basis functions is truncated to produce a sparse basis matrix, and the computational burden of transforms reduces to deal with image compression and reconstruction in a real-time processing. As the length of basis functions is decreased, the truncation effects to the energy of basis functions are examined and the change in various Qualify measures is evaluated. Experiment results show that 11 times fewer multiplication/addition operations are achieved with less than 1% performance change.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.13 no.3
• /
• pp.316-321
• /
• 2003

Recently, support vector learning attracts great interests in the areas of pattern classification, function approximation, and abnormality detection. It is well-known that among the various support vector learning methods, the so-called no-versions are particularly useful in cases that we need to control the total number of support vectors. In this paper, we consider the problem of function approximation utilizing both predetermined basis functions and a no-version support vector learning called $\nu-SVR$. After reviewing $\varepsilon-SVR$, $\nu-SVR$, and a semi-parametric approach, this paper presents an extension of the conventional $\nu-SVR$ method toward the direction that can utilize Predetermined basis functions. Moreover, the applicability of the presented method is illustrated via an example.

• Proceedings of the Acoustical Society of Korea Conference
• /
• 1994.06c
• /
• pp.55-60
• /
• 1994

본 논문에서는 음원을 모델링하기 위한 새로운 음원 모델로서 기저함수합계 모델을 제안하고 그 모델의 변수를 추정하는 방법에 관하여 설명한다. 기존 모델들이 다양한 음원신호를 표현하는데 부족함이 많았던데 비해 기저함수합계 모델은 다양한 음원신호를 표현하기에 적합하며 ML 이라는 통일된 추정 방법을 통해 모델의 변수들을 구할 수 있다. 또한 기저함수합계 모델은 기존의 모델들을 포함하는 일반화된 음원 모델이 됨을 보인다.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.39A no.3
• /
• pp.140-148
• /
• 2014

In this paper, We proposed Gaussian function as a basis of hybrid method. Hybrid method is to extrapolate late time and high frequency data using early time and low frequency data. This method takes advantages of both MOT and MOM as well as having shorter running time and smaller error. For this method a better basis function is required. We compared the performance of the result with proposed function and conventional basis including Hermite and Laguerre polynomial.