• Proceedings of the Korean Information Science Society Conference
• /
• 2002.10d
• /
• pp.559-561
• /
• 2002

서포트 벡터 머신은 얼굴인식이나 문자인식과 같은 다양한 패턴인식 문제에서 좋은 성능을 보여준다. 그러나 이러한 문제는 Quadratic Programming(QP) 문제에 관하여 몇 가지 단점을 가지고 있다. 일반적으로 대용량의 QP 문제를 해결하기 위해 많은 계산비용이 요구되며, QP 기반 시스템을 효과적으로 구현하는 것이 쉽지 않은 문제이다. 또한 대규모 데이터의 처리 시에는 입출력을 맞추기 또한 쉽지 않은 단점이 있다. 본 논문에서는 위의 단점을 극복하기 위하여 단일부류 문제를 최소제곱 서포트 벡터 머신을 기반으로 하여 해결하였다. 제안한 방법은 QP 문제를 해결하는 과정이 없이 단일부류 문제를 표현하여 최소제곱 방법을 이용하는 알고리즘이다. 제안된 방법으로 쉽고, 계산 비용을 줄이는 결과를 얻었다. 또한 서포트 벡터 영역 표식자에 확장 적용하여 선형방정식으로 구현하여, 문제를 해결하였다. 제안된 방법의 효율성을 입증하기 위하여 패턴인식 분야 중에 얼굴 인증 방법과 바이오인포매틱스 분야 중에 전립선 암 분류 문제에 적용하였다. 우리의 실험결과는 적합한 성능과 좋은 Equal Error Rate(EER)를 보여준다. 제안된 방법은 알 수 없는 물체의 분류 방법의 효율성을 증대시켰고, 실시간 응용분야에 직접적으로 적용될 수 있을 것으로 기대 된다.

• Journal of Internet Computing and Services
• /
• v.6 no.3
• /
• pp.159-170
• /
• 2005

OC-SVM(One Class Support Vector Machine) avoids solving a full density estimation problem, and instead focuses on a simpler task, estimating quantiles of a data distribution, i.e. its support. OC-SVM seeks to estimate regions where most of data resides and represents the regions as a function of the support vectors, Although OC-SVM is powerful method for data description, it is difficult to incorporate human subjective importance into its estimation process, In order to integrate the importance of each point into the OC-SVM process, we propose a fuzzy version of OC-SVM. In FOC-SVM (Fuzzy One-Class Support Vector Machine), we do not equally treat data points and instead weight data points according to the importance measure of the corresponding objects. That is, we scale the kernel feature vector according to the importance measure of the object so that a kernel feature vector of a less important object should contribute less to the detection process of OC-SVM. We demonstrate the performance of our algorithm on several synthesized data sets, Experimental results showed the promising results.

• Journal of KIISE
• /
• v.42 no.1
• /
• pp.86-92
• /
• 2015

A level set tree provides a useful representation of a multidimensional density function. Visualizing the data structure as a tree offers many advantages for data analysis and clustering. In this paper, we present a level set tree estimation algorithm for use with a set of data points. The proposed algorithm creates a level set tree from a family of level sets estimated over a whole range of levels from zero to infinity. Instead of estimating density function then thresholding, we directly estimate the density level sets using one-class support vector machines (OC-SVMs). The level set estimation is facilitated by the OC-SVM solution path algorithm. We demonstrate the proposed level set tree algorithm on benchmark data sets.

• Annual Conference on Human and Language Technology
• /
• 2012.10a
• /
• pp.8-11
• /
• 2012

One-class SVM은 데이터가 존재하는 영역을 추출하고, 이 영역을 서포트 벡터로 표현하며 표현된 영역 밖의 데이터들은 아웃라이어(outlier)로 간주된다. 본 논문에서는 데이터 포인트마다 숨겨진 변수(hidden variable) 혹은 토픽이 있다고 가정하고, 이를 반영하기 위해 PoMEN에 기반한 Latent One-class SVM을 제안한다. 실험결과 Latent One-class SVM이 대부분의 구간에서 One-class SVM 보다 성능이 높았으며, 특히 높은 정확율을 요구하는 경우에 더욱 효과적임을 알 수 있었다.

• Journal of Internet Computing and Services
• /
• v.12 no.5
• /
• pp.87-100
• /
• 2011

In this paper, we address a video summarization task as generating both visually salient and semantically important video segments. In order to find salient data points, one can use the OC-SVM (One-class Support Vector Machine), which is well known for novelty detection problems. It is, however, hard to incorporate into the OC-SVM process the importance measure of data points, which is crucial for video summarization. In order to integrate the importance of each point in the OC-SVM process, we propose a fuzzy version of OC-SVM. The Importance-based Fuzzy OC-SVM weights data points according to the importance measure of the video segments and then estimates the support of a distribution of the weighted feature vectors. The estimated support vectors form the descriptive segments that best delineate the underlying video content in terms of the importance and salience of video segments. We demonstrate the performance of our algorithm on several synthesized data sets and different types of videos in order to show the efficacy of the proposed algorithm. Experimental results showed that our approach outperformed the well known traditional method.

• Proceedings of the Korean Information Science Society Conference
• /
• 2005.11b
• /
• pp.835-837
• /
• 2005

최근 얼굴 인식은 사용자의 편의성을 포함한 다양한 장점으로 인하여 생체 인식 시장에서 주요 기술로 대두되고 있다. 그러나 조명 변화에 기인한 얼굴 인식 성능의 저하는 실용화에 걸림돌이 되고 있는 실정이다. 따라서 조명 변화에 따른 얼굴의 외형 변화를 분석하는 연구들이 세계적으로 활발히 진행되고 있다. 그러나 기존 방법들은 다수의 등록 영상이나 조명에 대한 사전 정보가 필요하거나 실시간으로 구현되기 어렵기 때문에 실용 시스템에 적용하기는 어려운 실정이다. 따라서, 본 논문에서는, 여러 조명 영상들로 구성된 학습 데이터를 이용하여, 조명에 대한 정보가 없는 한 장의 입력 영상을 분석하는 방법을 제안한다. 제안된 방법은 SVDD를 이용하여 학습 데이터의 여러 조면 영상들로부터 입력 영상의 조명과 같은 대표영상을 합성하고 이 대표영상들의 선형 조합을 이용하여 입력 영상을 표현한다. 제안 방법의 효율성을 검증하기 위하여 공인 얼굴 데이터베이스들을 이용하여, 기존 방법들과 비교 실험을 수행하였으며, 조명 변화가 큰 영상에서도 안정된 조명 변화의 분석이 가능하였다.

• The Korean Journal of Applied Statistics
• /
• v.26 no.6
• /
• pp.903-913
• /
• 2013

Social network services generate a considerable amount of social data every day on personal feelings or thoughts. This social data provides changing patterns of information production and consumption but are also a tool that reflects social phenomenon. We analyze negative emotional words from daily blogs to detect the change in blooger sentiment using multivariate control charts. We used the all the blogs produced between 1 January 2008 and 31 December 2009. Hotelling's T-square control chart control chart is commonly used to monitor multivariate quality characteristics; however, it assumes that quality characteristics follow multivariate normal distribution. The performance of a multivariate control chart is affected by this assumption; consequently, we introduce the support vector data description and its extension (K-control chart) suggested by Sun and Tsung (2003) and they are applied to detect the chage in blogger sentiment.

• Journal of Korea Multimedia Society
• /
• v.10 no.3
• /
• pp.391-400
• /
• 2007

According to the recent advances in multimedia computing, storage and searching technology have made large volume of music contents become prevalent. Also there has been increasing needs for the study on efficient categorization and searching technique for music contents management. In this paper, a new classifying method using the local information of music content and music tone feature is proposed. While the conventional classifying algorithms are based on entire information of music content, the algorithm proposed in this paper focuses on only the specific local information, which can drastically reduce the computing time without losing classifying accuracy. In order to improve the classifying accuracy, it uses a new classification feature based on music tone. The proposed method has been implemented as a part of MuSE (Music Search/Classification Engine) which was installed on various systems including commercial PDAs and PCs.

• Analytical Science and Technology
• /
• v.28 no.3
• /
• pp.187-195
• /
• 2015

The heat of sublimation (HOS) is an essential parameter used to resolve environmental problems in the transfer of organic contaminants to the atmosphere and to assess the risk of toxic chemicals. The experimental measurement of the heat of sublimation is time-consuming, expensive, and complicated. In this study, quantitative structural property relationships (QSPR) were used to develop a simple and predictive model for measuring the heat of sublimation of organic compounds. The population-based forward selection method was applied to select an informative subset of descriptors of learning algorithms, such as by using multiple linear regression (MLR) and the support vector machine (SVM) method. Each individual model and consensus model was evaluated by internal validation using the bootstrap method and y-randomization. The predictions of the performance of the external test set were improved by considering their applicability to the domain. Based on the results of the MLR model, we showed that the heat of sublimation was related to dispersion, H-bond, electrostatic forces, and the dipole-dipole interaction between inter-molecules.

• Analytical Science and Technology
• /
• v.24 no.6
• /
• pp.544-555
• /
• 2011

The acute toxicity in the rainbow trout (Oncorhynchus mykiss) was analyzed and predicted using quantitative structure-activity relationships (QSAR). The aquatic toxicity, 96h $LC_{50}$ (median lethal concentration) of 275 organic pesticides, was obtained from EU-funded project DEMETRA. Prediction models were derived from 558 2D molecular descriptors, calculated in PreADMET. The linear (multiple linear regression) and nonlinear (support vector machine and artificial neural network) learning methods were optimized by taking into account the statistical parameters between the experimental and predicted p$LC_{50}$. After preprocessing, population based forward selection were used to select the best subsets of descriptors in the learning methods including 5-fold cross-validation procedure. The support vector machine model was used as the best model ($R^2_{CV}$=0.677, RMSECV=0.887, MSECV=0.674) and also correctly classified 87% for the training set according to EU regulation criteria. The MLR model could describe the structural characteristics of toxic chemicals and interaction with lipid membrane of fish. All the developed models were validated by 5 fold cross-validation and Y-scrambling test.