• Journal of KIISE:Software and Applications
• /
• v.31 no.3
• /
• pp.361-368
• /
• 2004

The purpose of this study is to propose a new on-line nonlinear PCA(OL-NPCA) method for a nonlinear feature extraction from the incremental data. Kernel PCA(KPCA) is widely used for nonlinear feature extraction, however, it has been pointed out that KPCA has the following problems. First, applying KPCA to N patterns requires storing and finding the eigenvectors of a N${\times}$N kernel matrix, which is infeasible for a large number of data N. Second problem is that in order to update the eigenvectors with an another data, the whole eigenspace should be recomputed. OL-NPCA overcomes these problems by incremental eigenspace update method with a feature mapping function. According to the experimental results, which comes from applying OL-NPCA to a toy and a large data problem, OL-NPCA shows following advantages. First, OL-NPCA is more efficient in memory requirement than KPCA. Second advantage is that OL-NPCA is comparable in performance to KPCA. Furthermore, performance of OL-NPCA can be easily improved by re-learning the data.

• Structural Engineering and Mechanics
• /
• v.27 no.5
• /
• pp.625-638
• /
• 2007

The damage detection process may appear difficult to be implemented for truss structures because not all degrees of freedom in the numerical model can be experimentally measured. In this context, the damage locating vector (DLV) method, introduced by Bernal (2002), is a useful approach because it is effective when operating with an arbitrary number of sensors, a truncated modal basis and multiple damage scenarios, while keeping the calculation in a low level. In addition, the present paper also evaluates the noise influence on the accuracy of the DLV method. In order to verify the DLV behavior under different damages intensities and, mainly, in presence of measurement noise, a parametric study had been carried out. Different excitations as well as damage scenarios are numerically tested in a continuous Warren truss structure subjected to five noise levels with a set of limited measurement sensors. Besides this, it is proposed another way to determine the damage locating vectors in the DLV procedure. The idea is to contribute with an alternative option to solve the problem with a more widespread algebraic method. The original formulation via singular value decomposition (SVD) is replaced by a common solution of an eigenvector-eigenvalue problem. The final results show that the DLV method, enhanced with the alternative solution proposed in this paper, was able to correctly locate the damaged bars, using an output-only system identification procedure, even considering small intensities of damage and moderate noise levels.

• Journal of the Korean Mathematical Society
• /
• v.46 no.3
• /
• pp.657-673
• /
• 2009

The fixed point algebra $C^*(E)^{\gamma}$ of a gauge action $\gamma$ on a graph $C^*$-algebra $C^*(E)$ and its AF subalgebras $C^*(E)^{\gamma}_{\upsilon}$ associated to each vertex v do play an important role for the study of dynamical properties of $C^*(E)$. In this paper, we consider the stability of $C^*(E)^{\gamma}$ (an AF algebra is either stable or equipped with a (nonzero bounded) trace). It is known that $C^*(E)^{\gamma}$ is stably isomorphic to a graph $C^*$-algebra $C^*(E_{\mathbb{Z}}\;{\times}\;E)$ which we observe being stable. We first give an explicit isomorphism from $C^*(E)^{\gamma}$ to a full hereditary $C^*$-subalgebra of $C^*(E_{\mathbb{N}}\;{\times}\;E)({\subset}\;C^*(E_{\mathbb{Z}}\;{\times}\;E))$ and then show that $C^*(E_{\mathbb{N}}\;{\times}\;E)$ is stable whenever $C^*(E)^{\gamma}$ is so. Thus $C^*(E)^{\gamma}$ cannot be stable if $C^*(E_{\mathbb{N}}\;{\times}\;E)$ admits a trace. It is shown that this is the case if the vertex matrix of E has an eigenvector with an eigenvalue $\lambda$ > 1. The AF algebras $C^*(E)^{\gamma}_{\upsilon}$ are shown to be nonstable whenever E is irreducible. Several examples are discussed.

• KIPS Transactions on Software and Data Engineering
• /
• v.10 no.3
• /
• pp.79-84
• /
• 2021

A feature extraction method capable of reflecting features well while mainaining the properties of data is required in order to process high-dimensional data. The principal component analysis method that converts high-level data into low-dimensional data and express high-dimensional data with fewer variables than the original data is a representative method for feature extraction of data. In this study, we propose a principal component analysis method based on adaptive correlation when selecting principal component variables in principal component analysis for data feature extraction when the data is high-dimensional. The proposed method analyzes the principal components of the data by adaptively reflecting the correlation based on the correlation between the input data. I want to exclude them from the candidate list. It is intended to analyze the principal component hierarchy by the eigen-vector coefficient value, to prevent the selection of the principal component with a low hierarchy, and to minimize the occurrence of data duplication inducing data bias through correlation analysis. Through this, we propose a method of selecting a well-presented principal component variable that represents the characteristics of actual data by reducing the influence of data bias when selecting the principal component variable.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.37 no.2
• /
• pp.31-43
• /
• 2000

This paper presents a novel algorithm lot extraction of the eye and mouth fields (facial features) from 2D gray level face images. First of all, it has been found that Eigenfeatures, derived from the eigenvalues and the eigenvectors of the binary edge data set constructed from the eye and mouth fields are very good features to locate these fields. The Eigenfeatures, extracted from the positive and negative training samples for the facial features, ate used to train a MultiLayer Perceptron(MLP) whose output indicates the degree to which a particular image window contains the eye or the mouth within itself. Second, to ensure robustness, the ensemble network consisting of multiple MLPs is used instead of a single MLP. The output of the ensemble network becomes the average of the multiple locations of the field each found by the constituent MLPs. Finally, in order to reduce the computation time, we extracted the coarse search region lot eyes and mouth by using prior information on face images. The advantages of the proposed approach includes that only a small number of frontal faces are sufficient to train the nets and furthermore, lends themselves to good generalization to non-frontal poses and even to other people's faces. It was also experimentally verified that the proposed algorithm is robust against slight variations of facial size and pose due to the generalization characteristics of neural networks.

• Journal of Korean Tunnelling and Underground Space Association
• /
• v.5 no.1
• /
• pp.71-88
• /
• 2003

The DRASTIC system is widely used for assessing regional groundwater pollution susceptibility by using hydrogeological factors such as depth to water, net recharge, aquifer media, soil media, topography, vadose zone media, hydraulic conductivity. This study is providing Modified Drastic Model to which lineament density, land use, influence of groundwater drawdown caused by tunnel excavation are added as additional factors using geographic information system, and then to evaluate groundwater contamination potential of ${\bigcirc}{\bigcirc}$ area. For statistical analysis, vector coverage per each factor is converted to grid layer and after each correlation coefficient between factors, covariance, variance, eigenvalue and eigenvector by principal component analysis of 3 direction, are calculated, correlation between factors is analyzed. Also after correlation coefficients between general DRASTIC layer and rated lineament density layer, between general DRASTIC layer and rated land use layer, between general DRASTIC layer and rated tunnel excavation influence layer are calculated, final modified DRASTIC model is constructed by using them with each weighting. When modified DRASTIC model was compared with general DRASTIC model, contamination potential in modified DRASTIC model is fairly detailed and consequently, vulnerable area which has high contamination potential could be presented concretly.

• Proceedings of the KSMRM Conference
• /
• 2002.11a
• /
• pp.133-133
• /
• 2002

Purpose： The purpose of paper is to implement software to visualize brain fiber tract using a 24-bit color coding scheme and to test its feasibility. Materials and Methods： MR imaging was performed on GE 1.5 T Signa scanner. For diffusion tensor image, we used a single shot spin-echo EPI sequence with 7 non-colinear pulsed-field gradient directions： (x, y, z)：(1,1,0),(-1,1,0),(1,0,1),(-1,0,1),(0,1,1),(0,1,-1) and without diffusion gradient. B-factor was 500 sec/$\textrm{mm}^2$. Acquisition parameters are as follows： TUTE=10000ms/99ms, FOV=240mm, matrix=128${\times}$128, slice thickness/gap=6mm/0mm, total slice number=30. Subjects consisted of 10 normal young volunteers (age：21∼26 yrs, 5 men, 5 women). All DTI images were smoothed with Gaussian kernel with the FWHM of 2 pixels. Color coding schemes for visualization of directional information was as follows. HSV(Hue, Saturation, Value) color system is appropriate for assigning RGB(Red, Green, and Blue) value for every different directions because of its volumetric directional expression. Each of HSV are assigned due to (r,$\theta$,${\Phi}$) in spherical coordinate. HSV calculated by this way can be transformed into RGB color system by general HSV to RGB conversion formula. Symmetry schemes： It is natural to code the antipodal direction to be same color(antipodal symmetry). So even with no symmetry scheme, the antipodal symmetry must be included. With no symmetry scheme, we can assign every different colors for every different orientation.(H =${\Phi}$, S=2$\theta$/$\pi$, V=λw, where λw is anisotropy). But that may assign very discontinuous color even between adjacent yokels. On the other hand, Full symmetry or absolute value scheme includes symmetry for 180$^{\circ}$ rotation about xy-plane of color coordinate (rotational symmetry) and for both hemisphere (mirror symmetry). In absolute value scheme, each of RGB value can be expressed as follows. R=λw｜Vx｜, G=λw｜Vy｜, B=λw｜Vz｜, where (Vx, Vy, Vz) is eigenvector corresponding to the largest eigenvalue of diffusion tensor. With applying full symmetry or absolute value scheme, we can get more continuous color coding at the expense of coding same color for symmetric direction. For better visualization of fiber tract directions, Gamma and brightness correction had done. All of these implementations were done on the IDL 5.4 platform.