• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 1997.11a
• /
• pp.248-251
• /
• 1997

In this paper, We propose the concept of approximate Classification in the field of two group discriminan analysis. In our approach, an attribute space is divided into three subspaces. Two subspaces are for given two group and one subspace is for a boundary area between the two groups. We propose Approximate Pattern Classification with Rough set. We also propose learning procedures of neural networks for approximate classification. We propose two weighting methods which lead to possibility analysis and necessity analysis. We illustrate the proposed methods by numerical examples.

• Journal of the Institute of Electronics Engineers of Korea SC
• /
• v.48 no.5
• /
• pp.7-12
• /
• 2011

In this papwer, we propose a new feature extraction method, named as Class-augmented Kernel Principal Component Analysis (CA-KPCA), which can extract nonlinear features for classification. Among the subspace method that was being widely used for feature extraction, Class-augmented Principal Component Analysis (CA-PCA) is a recently one that can extract features for a accurate classification without computational difficulties of other methods such as Linear Discriminant Analysis (LDA). However, the features extracted by CA-PCA is still restricted to be in a linear subspace of the original data space, which limites the use of this method for various problems requiring nonlinear features. To resolve this limitation, we apply a kernel trick to develop a new version of CA-PCA to extract nonlinear features, and evaluate its performance by experiments using data sets in the UCI Machine Learning Repository.

• Bulletin of the Korean Mathematical Society
• /
• v.34 no.4
• /
• pp.667-678
• /
• 1997

This paper contains a convergence analysis of a flexible restarted FOM(k)(FFOM(k)), and its performance is compared with FGMRES(k). Performances of these two algorithms with variable preconditioners are also compared with those of preconditioned FOM(k) and GMRES(k). Numerical experiments show that FFOM(k) performs as well as, or better than for some problems, FGMRES(k).

• Proceedings of the Korean Information Science Society Conference
• /
• 2001.04b
• /
• pp.529-531
• /
• 2001

We present an expectation-maximization algorithm for principal component analysis via orthogonalization. The algorithm finds actual principal components, whereas previously proposed EM algorithms can only find principal subspace. New algorithm is simple and more efficient thant probabilistic PCA specially in noiseless cases. Conventional PCA needs computation of inverse of the covariance matrices, which makes the algorithm prohibitively expensive when the dimensions of data space is large. This EM algorithm is very powerful for high dimensional data when only a few principal components are needed.

• Transactions of the Korean Society of Mechanical Engineers A
• /
• v.35 no.10
• /
• pp.1211-1222
• /
• 2011

It is necessary to use prestressed modal analysis to calculate the modal frequencies and mode shapes of a prestressed structure such as a spinning blade, a preloaded structure, or a thermally deformed pipe, because the prestress effect sometimes causes significant changes in the frequencies and mode shapes. When the finite element model under consideration has a very large number of degrees of freedom, repeated prestressed modal analyses for investigating the prestress effects might become too computationally expensive to finish within a reasonable design-process time. To alleviate these computational difficulties, a Krylov subspace-based model order reduction, which reduces the number of degrees of freedom of the original finite element model and speeds up the necessary prestressed modal analysis with the reduced order models (ROMs), is presented. The numerical process for the moment-matching model reduction is performed directly on the full order models (FOMs) (modeled in ANSYS) by the Arnoldi process. To demonstrate the advantages of this approach for performing prestressed modal analysis, the prestressed wheel and the compressor impeller under their high-speed rotation are considered as examples.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.51 no.5
• /
• pp.177-187
• /
• 2014

This paper proposes a fast subspace tracking methods, which is called GVFF FAPI, based on FAPI (Fast Approximated Power Iteration) method and GVFF RLS (Gradient-based Variable Forgetting Factor Recursive Lease Squares). Since the conventional FAPI uses a constant forgetting factor for estimating covariance matrix of source signals, it has difficulty in applying to non-stationary environments such as continuously changing DOAs of source signals. To overcome the drawback of conventioanl FAPI method, the GVFF FAPI uses the gradient-based variable forgetting factor derived from an improved means square error (MSE) analysis of RLS. In order to achieve the decreased subspace error in non-stationary environments, the GVFF-FAPI algorithm used an improved forgetting factor updating equation that can produce a fast decreasing forgetting factor when the gradient is positive and a slowly increasing forgetting factor when the gradient is negative. Our numerical simulations show that GVFF-FAPI algorithm offers lower subspace error and RMSE (Root Mean Square Error) of tracked DOAs of source signals than conventional FAPI based MUSIC (MUltiple SIgnal Classification).

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.47 no.1
• /
• pp.35-43
• /
• 2010

1D PCA utilized in the face appearance-based face recognition methods such as eigenface-based face recognition method may lead to less face representative power and more computational cost due to the resulting 1D face appearance data vector of high dimensionality. To resolve such problems of 1D PCA, 2D PCA-based face recognition methods had been developed. However, the face representation model obtained by direct application of 2D PCA to a face image set includes both face common features and face distinctive identity features. Face common features not only prevent face recognizability but also cause more computational cost. In this paper, we first develope a model of a face distinctive identity feature subspace separated from the effects of face common features in the face feature space obtained by application of 2D PCA analysis. Then, a novel robust face recognition based on the face distinctive identity feature subspace model is proposed. The proposed face recognition method based on the face distinctive identity feature subspace shows better performance than the conventional PCA-based methods (1D PCA-based one and 2D PCA-based one) with respect to recognition rate and processing time since it depends only on the face distinctive identity features. This is verified through various experiments using Yale A and IMM face database consisting of face images with various face poses under various illumination conditions.

• Proceedings of the Korea Institutes of Information Security and Cryptology Conference
• /
• 2006.06a
• /
• pp.251-257
• /
• 2006

The performance of face recognition methods using subspace projection is directly related to the characteristics of their basis images, especially in the cases of local distortion or partial occlusion. In order for a subspace projection method to be robust to local distortion and partial occlusion, the basis images generated by the method should exhibit a part-based local representation. We propose an effective part-based local representation method named locally salient ICA (LS-ICA) method for face recognition that is robust to local distortion and partial occlusion. The LS-ICA method only employs locally salient information from important facial parts in order to maximize the benefit of applying the idea of 'recognition by parts.' It creates part-based local basis images by imposing additional localization constraint in the process of computing ICA architecture I basis images. We have contrasted the LS-ICA method with other part-based representations such as LNMF (Localized Non-negative Matrix Factorization)and LFA (Local Feature Analysis). Experimental results show that the LS-ICA method performs better than PCA, ICA architecture I, ICA architecture II, LFA, and LNMF methods, especially in the cases of partial occlusions and local distortion

• Wind and Structures
• /
• v.11 no.3
• /
• pp.209-220
• /
• 2008

Flutter derivatives provide the basis of predicting the critical wind speed in flutter and buffeting analysis of long-span cable-supported bridges. Many studies have been performed on the methods and applications of identification of flutter derivatives of bridge decks under wind action. In fact, strong wind, especially typhoon, is always accompanied by heavy rain. Then, what is the effect of rain on flutter derivatives and flutter critical wind speed of bridges? Unfortunately, there have been no studies on this subject. This paper makes an initial study on this problem. Covariance-driven Stochastic Subspace Identification (SSI in short) which is capable of estimating the flutter derivatives of bridge decks from their steady random responses is presented first. An experimental set-up is specially designed and manufactured to produce the conditions of rain and wind. Wind tunnel tests of a quasi-streamlined thin plate model are conducted under conditions of only wind action and simultaneous wind-rain action, respectively. The flutter derivatives are then extracted by the SSI method, and comparisons are made between the flutter derivatives under the two different conditions. The comparison results tentatively indicate that rain has non-trivial effects on flutter derivatives, especially on and $H_2$ and $A_2$thus the flutter critical wind speeds of bridges.

• Smart Structures and Systems
• /
• v.26 no.4
• /
• pp.419-433
• /
• 2020

Investigation of structural integrity has been a critical issue in the field of civil engineering for years. Visual inspection is one of the most available methods to explore deteriorative components in structures. Still, this method is not applicable to invisible damage of structures. Alternatively, system identification methods are capable of tracking modal properties of structures over time. The deviation of these dynamic properties can serve as indicators to access structural integrity. In this study, a modal tracking technique using frequency-domain system identification from seismic responses of structures is proposed. The method first segments the measured signals into overlapped sequential portions and then establishes multiple Hankel matrices. Each Hankel matrix is then converted to the frequency domain, and a temporal-average frequency-domain Hankel matrix can be calculated. This study also proposes the frequency band selection that can divide the frequency-domain Hankel matrix into several portions in accordance with referenced natural frequencies. Once these referenced natural frequencies are unavailable, the first few right singular vectors by the singular value decomposition can offer these references. Finally, the frequency-domain stochastic subspace identification tracks the natural frequencies and mode shapes of structures through quick stabilization diagrams. To evaluate performance of the proposed method, a numerical study is carried out. Moreover, the long-term monitoring strong motion records at a specific site are exploited to assess the tracking performance. As seen in results, the proposed method is capable of tracking modal properties through seismic responses of structures.