• Proceedings of the Korean Information Science Society Conference
• /
• 2005.07b
• /
• pp.715-717
• /
• 2005

Support vector machine(SVM)은 최근 각광받는 기계학습 방법 중 하나로서, kernel function 이라는 사상(mapping)을 이용하여 입력 공간의 벡터를 classification이 용이한 특징 (feature) 공간의 벡터로 변환하는 것을 근간으로 한다. SVM은 이러한 특징 공간에서 두 클래스를 구분 짓는 hyperplane을 일련의 최적화 방법론을 사용하여 찾아내며, 주어진 문제가 convex problem 인 경우 항상 global optimal solution 을 보장하는 등의 장점을 지닌다. 한편 bioinformatics 연구에서 주로 사용되는 데이터는 측정 오류 등 일련의 오류를 포함하고 있으며, 이러한 오류는 기계학습 방법론이 어떤 decision boundary를 찾아내는가에 영향을 끼치게 된다. 특히 SVM의 경우 이러한 오류는 특징 공간 벡터간의 관계를 나타내는 Gram matrix를 변화로 나타나게 된다. 본 연구에서는 입력 공간에 오류가 발생할 때 그것이 SVM 의 decision boundary를 어떻게 변화시키는가를 대표적인 두 가지 kernel function, 즉 linear kernel과 Gaussian kernel에 대해 분석하였다. Wisconsin대학의 유방암(breast cancer) 데이터에 대해 실험한 결과, 데이터의 오류에 따른 SVM 의 classification 성능 변화 양상을 관찰하여 커널의 종류에 따라 SVM이 어떠한 특성을 보이는가를 밝혀낼 수 있었다. 또 흥미롭게도 어떤 조건 하에서는 오류가 크더라도 오히려 SVM 의 성능이 향상되는 것을 발견했는데, 이것은 바꾸어 생각하면 Gram matrix 의 일부를 변경하여 SVM 의 성능 향상을 꾀할 수 있음을 나타낸다.

• Journal of Internet Computing and Services
• /
• v.11 no.5
• /
• pp.1-8
• /
• 2010

This paper presentsthe performance study of blind equalizer algorithms for impulsive-noise environments based on Gaussian kernel and constant modulus error(CME). Constant modulus algorithm(CMA) based on CME and mean squared error(MSE) criterion fails in impulsive noise environment. Correntropy blind method recently introduced for impulsive-noise resistance has shown in PAM system not very satisfying results. It is revealed in theoretical and simulation analysis that the maximization of zero-error probability based on CME(MZEP-CME) originally proposed for Gaussian noise environments produces superior performance in impulsive noise channels as well. Gaussian kernel of MZEP-CME has a strong effect of becoming insensitive to the large differences between the power of impulse-infected outputs and the constant modulus value.

• 한국가시화정보학회:학술대회논문집
• /
• 2004.12a
• /
• pp.47-55
• /
• 2004

Several interpolations for image deformation in PIV were evaluated. The tested interpolation methods are linear, quadratic, truncated sinc, windowed sinc, cubic, Lagrange, Gaussian $2^{nd}\;and\;6^{th}$ interpolators. Bias errors and random errors were evaluated in the range of $0\~3.0$ pixel uniform displacement using synthetic images. We also measured the time cost of each interpolator with respect to kernel size. The cubic interpolator with $6\times6$ kernel showed the best results in terms of the performance and time cost.

• International Journal of Control, Automation, and Systems
• /
• v.6 no.1
• /
• pp.109-118
• /
• 2008

We present two estimators for discrete non-Gaussian and nonstationary probability density estimation based on a dynamic Bayesian network (DBN). The first estimator is for off line computation and consists of a DBN whose transition distribution is represented in terms of kernel functions. The estimator parameters are the weights and shifts of the kernel functions. The parameters are determined through a recursive learning algorithm using maximum likelihood (ML) estimation. The second estimator is a DBN whose parameters form the transition probabilities. We use an asymptotically convergent, recursive, on-line algorithm to update the parameters using observation data. The DBN calculates the state probabilities using the estimated parameters. We provide examples that demonstrate the usefulness and simplicity of the two proposed estimators.

• Journal of Communications and Networks
• /
• v.14 no.5
• /
• pp.540-545
• /
• 2012

In this paper, the lagged cross-correlation of two probability density functions constructed by kernel density estimation is proposed, and by maximizing the proposed function, adaptive filtering algorithms for supervised and unsupervised training are also introduced. From the results of simulation for blind equalization applications in multipath channels with impulsive and slowly varying direct current (DC) bias noise, it is observed that Gaussian kernel of the proposed algorithm cuts out the large errors due to impulsive noise, and the output affected by the DC bias noise can be effectively controlled by the lag ${\tau}$ intrinsically embedded in the proposed function.

• Korean Journal of Remote Sensing
• /
• v.26 no.6
• /
• pp.705-719
• /
• 2010

We investigated the speckle filtering effect in supervised classification of the C-band polarimetric Ground Based SAR image data. Wishart classification method was used for the supervised classification of the polarimetric GB-SAR image data and total of 6 kinds of speckle filters were applied before supervised classification, which are boxcar, Gaussian, Lopez, IDAN, the refined Lee, and the refined Lee sigma filters. For each filters, we changed the filtering kernel size from $3{\times}3$ to $9{\times}9$ to investigate the filtering size effect also. The refined Lee filter with the kernel size of bigger than $5{\times}5$ showed the best result for the Wishart supervised classification of polarimetric GB-SAR image data. The result also showed that the type of trees could be discriminated by Wishart supervised classification of polarimetric GB-SAR image data.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.17 no.5
• /
• pp.679-683
• /
• 2007

Statistical learning theory by Vapnik consists of support vector machine(SVM), support vector regression(SVR), and support vector clustering(SVC) for classification, regression, and clustering respectively. In this algorithms, SVC is good clustering algorithm using support vectors based on Gaussian kernel function. But, similar to SVM and SVR, SVC needs to determine kernel parameters and regularization constant optimally. In general, the parameters have been determined by the arts of researchers and grid search which is demanded computing time heavily. In this paper, we propose a differential evolution based SVC(DESVC) which combines differential evolution into SVC for efficient selection of kernel parameters and regularization constant. To verify improved performance of our DESVC, we make experiments using the data sets from UCI machine learning repository and simulation.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.13 no.6
• /
• pp.3055-3073
• /
• 2019

In this paper, a new user identification method is presented using real environmental human-computer-interaction (HCI) behavior data to improve method usability. User behavior data in this paper are collected continuously without setting experimental scenes such as text length, action number, etc. To illustrate the characteristics of real environmental HCI data, probability density distribution and performance of keyboard and mouse data are analyzed through the random sampling method and Support Vector Machine(SVM) algorithm. Based on the analysis of HCI behavior data in a real environment, the Multiple Kernel Learning (MKL) method is first used for user HCI behavior identification due to the heterogeneity of keyboard and mouse data. All possible kernel methods are compared to determine the MKL algorithm's parameters to ensure the robustness of the algorithm. Data analysis results show that keyboard data have a narrower range of probability density distribution than mouse data. Keyboard data have better performance with a 1-min time window, while that of mouse data is achieved with a 10-min time window. Finally, experiments using the MKL algorithm with three global polynomial kernels and ten local Gaussian kernels achieve a user identification accuracy of 83.03% in a real environmental HCI dataset, which demonstrates that the proposed method achieves an encouraging performance.

• Proceedings of the IEEK Conference
• /
• 2002.07a
• /
• pp.393-396
• /
• 2002

In this paper, we give the definitions of the q-Haar and q-Gabor wavelet. Instead of using the conventional Gaussian distribution as a kernel of the Gabor wavelet, if the q-normal distribution is used, we can get the q-Gabor wavelet as a possible generalization of the Gabor wavelet. The q-normal distribution, which is given by the author, is one of the generalized Gaussian distribution. On the other hand, if two sets of the q-normal distribution are connected anti-symmetrically, we can get the q-Haar wavelet as a possible generalization of the Haiw wavelet. We give experiments on the q-eabor and q-Haar wavelet and discuss about the q-Gabor and q-Haar wavelet.

• Journal of Korean Society for Geospatial Information Science
• /
• v.4 no.1 s.6
• /
• pp.67-73
• /
• 1996

The objective of this research is to investigate the behavior of the $F\"{o}rstner$ interest operator, which has been widely used for detecting distinct points in the field of digital photogrammetry and computer vision, in scale space. Considering the hugh volume of digital image utilized in digital photogrammetry, the scale space (image pyramid) approach which appears to be a solution for enhancing image processing, began to gain its attention. The investigation of the $F\"{o}rstner$ interest operator in scale space generated by the Gaussian kernel shows its behavior and feasibility for being used in practice.