• Journal of the Korean Data and Information Science Society
• /
• v.24 no.3
• /
• pp.625-636
• /
• 2013

In this paper we study four kernel machines for estimating expected shortfall, which are constructed through combinations of support vector quantile regression (SVQR), restricted SVQR (RSVQR), least squares support vector machine (LS-SVM) and support vector expectile regression (SVER). These kernel machines have obvious advantages such that they achieve nonlinear model but they do not require the explicit form of nonlinear mapping function. Moreover they need no assumption about the underlying probability distribution of errors. Through numerical studies on two artificial an two real data sets we show their effectiveness on the estimation performance at various confidence levels.

• Geomechanics and Engineering
• /
• v.11 no.3
• /
• pp.361-372
• /
• 2016

The soil-concrete interface shear strength, although has been extensively studied, is still difficult to predict as a result of the dependence on many factors such as normal stresses, surface roughness, particle sizes, moisture contents, dilation angles of soils, etc. In this study, a well-known rigorous statistical learning approach, namely the least squares support vector machine (LS-SVM) realized in a ubiquitous spreadsheet platform is firstly used in estimating the soil-structure interface shear strength. Instead of studying the complicated mechanism, LS-SVM enables to explore the possible link between the fundamental factors and the interface shear strengths, via a sophisticated statistic approach. As a preliminary investigation, the authors study the expansive soils that are found extensively in most countries. To reduce the complexity, three major influential factors, e.g., initial moisture contents, initial dry densities and normal stresses of soils are taken into account in developing the LS-SVM models for the soil-concrete interface shear strengths. The predicted results by LS-SVM show reasonably good agreement with experimental data from direct shear tests.

• Journal of the Korean Data and Information Science Society
• /
• v.14 no.4
• /
• pp.965-974
• /
• 2003

In this paper we propose an incremental classification of multi-class data set by LS-SVM. By encoding the output variable in the training data set appropriately, we obtain a new specific output vectors for the training data sets. Then, online LS-SVM is applied on each newly encoded output vectors. Proposed method will enable the computation cost to be reduced and the training to be performed incrementally. With the incremental formulation of an inverse matrix, the current information and new input data are used for building another new inverse matrix for the estimation of the optimal bias and lagrange multipliers. Computational difficulties of large scale matrix inversion can be avoided. Performance of proposed method are shown via numerical studies and compared with artificial neural network.

• The Korean Journal of Applied Statistics
• /
• v.21 no.1
• /
• pp.159-168
• /
• 2008

Since the term structure of interest rates (TSIR) has longitudinal data, we should consider as input variables both time left to maturity and time simultaneously to get a more useful and more efficient function estimation. However, since the resulting data set becomes very large, we need to develop a fast and reliable estimation method for large data set. Furthermore, it tends to overestimate TSIR because data are correlated. To solve these problems we propose a mixture of weighted least squares support vector machines. We recognize that the estimate is well smoothed and well explains effects of the third stock market crash in USA through applying the proposed method to the US Treasury bonds data.

• Communications for Statistical Applications and Methods
• /
• v.13 no.3
• /
• pp.765-776
• /
• 2006

In this paper we propose an estimation method on the regression model with randomly censored observations of the training data set. The weighted least squares support vector machine regression is applied for the regression function estimation by incorporating the weights assessed upon each observation in the optimization problem. Numerical examples are given to show the performance of the proposed estimation method.

• Journal of the Korean Institute of Telematics and Electronics A
• /
• v.28A no.9
• /
• pp.671-679
• /
• 1991

In adaptive antenna, several models are known according to a prior knowledge about jammer signal. Among those, Frost model with contraint is generally used however it has the problem that convergence speed is slow and that stability is not good. To improve such problems, this paper proposes constrained NLMS algorithm using LS method. In addition, the result obtained by applying this algorithm to Duvall antenna model is compared with that of Frost model.

• 제어로봇시스템학회:학술대회논문집
• /
• 2004.08a
• /
• pp.383-388
• /
• 2004

This paper presents the aerodynamic coefficient modeling with a new model structure explored by Least Squares using Modulating Function Technique (LS/MFT) for an F-16XL airplane using wind tunnel data supplied by NASA/LRC. A new model structure for aerodynamic coefficient was proposed, one that considered all possible combination terms of angle of attack ${\alpha}$(t) and ${\alpha}$(t) given number of harmonics K, and was compared with Pearson's model, which has the same number of parameters as the new model. Our new model harmonic results show better agreement with the physical data than Pearson's model. The number of harmonics in the model was extended to 6 and its parameters were estimated by LS/MFT. The model output of lift coefficient with K=6 correspond reasonably well with the physical data. In particular, the estimation performances of four aerodynamic coefficients were greatly improved at high frequency by considering all harmonics included in the input${\alpha}$(t), and by using the new model. In addition, the importance of each parameter in the model was analyzed by parameter reduction errors. Moreover, the estimation of three parameters, i.e., amplitude, phase and frequency, for a pure sinusoid and a finite sum of sinusoids- using LS/MFT is investigated.

• Journal of the Institute of Electronics Engineers of Korea TC
• /
• v.49 no.8
• /
• pp.25-32
• /
• 2012

In this paper, we propose a method for improving the performance of low complexity LS channel estimation for OFDM in fast time varying channels. The CE-BEM channel model used for the low complexity LS channel estimation has a problem on its own and deteriorates channel estimation performance. In this paper, we first use time domain windowing in order to remove the effect of ICI caused by data symbols. Then samples are taken from the results of the LS channel estimation and the effects of the windowing are removed from them. For resolving the defect of CE-BEM, the channel responses are recovered by interpolating the resultant samples with DPSS employed as basis functions the characteristics of which is well matched to the time variation of the channel. Computer simulations show that the proposed channel estimation method gives rise to performance improvement over conventional methods especially when channel variation is very fast and confirm that not only which type of functions is selected for the basis but how many functions are used for the basis is another key factor to performance improvement.

• 제어로봇시스템학회:학술대회논문집
• /
• 1993.10b
• /
• pp.1-6
• /
• 1993

This report proposes a novel method of identification of continuous time-delay systems from sampled input-output data. By the aid of a digital pre-filter, an approximated discrete-time estimation model is first derived, in which the system parameters remain in their original form and the time delay need not be an integral multiple of th sampling period. Then an identification method combining the common linear least squares(LS) method or the instrumental variable(IV) method with the genetic algorithm(GA) is proposed. That is, the time-delay is selected by the GA, and the system parameters are estimated by the LS or IV method. Furthermore, the proposed method is extended to the case of multi-input multi-output systems where the time-delays in the individual input channels may differ each other. Simulation resutls show that our method yields consistent estimates even in the presence of high measurement noises.

• Proceedings of the Korea Institute of Convergence Signal Processing
• /
• 2003.06a
• /
• pp.268-271
• /
• 2003

There are several methods for the design of 2D filter. Notable among them is McClellan transform method. This transform allows us to obtain a high order 2D FIR filter through mapping the 1D frequency points of a 1D prototype FIR filter onto 2D frequency contours. We design 2D filter using this transform. Then we notice for mapping deviation of the 2D filter. In this paper, Quadratic programming (QP) method allows us to obtain coefficients of McClellan transform. Then we compare deviation of QP method with least-squares(LS) method. Elliptic filter is used for comparison. The optimal cutoff frequencies of a 1D filter are obtained directly from the QP method. Also several problem of LS method are solved.