• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 2004.10a
• /
• pp.313-316
• /
• 2004

Model selection is the process that sets up the regularization parameter in the support vector machine or regularization network by using the external methods such as general cross validation or L-curve criterion. This paper suggests that the regularization parameter can be obtained simultaneously within the learning process of neural networks without resort to separate selection methods. In this paper, extended kernel method is introduced. The relationship between regularization parameter and the bias term in the extended kernel is established. Experimental results show the effectiveness of the new model selection method.

• Journal of Information Processing Systems
• /
• v.13 no.5
• /
• pp.1168-1182
• /
• 2017

In the past decades, various image regularization methods have been introduced. Among them, total variation model has drawn much attention for the reason of its low computational complexity and well-understood mathematical behavior. However, regularization parameter estimation of total variation model is still an open problem. To deal with this problem, a novel adaptive regularization parameter selection scheme is proposed in this paper, by means of using the local spectral response, which has the capability of locally selecting the regularization parameters in a content-aware way and therefore adaptively adjusting the weights between the two terms of the total variation model. Experiment results on simulated and real noisy image show the good performance of our proposed method, in visual improvement and peak signal to noise ratio value.

• Structural Engineering and Mechanics
• /
• v.22 no.4
• /
• pp.401-418
• /
• 2006

Solutions to the problems of structural parameter estimation from modal response using leastsquares minimization of force or displacement residuals are generally sensitive to noise in the response measurements. The sensitivity of the parameter estimates is governed by the physical characteristics of the structure and certain features of the noisy measurements. It has been shown that the regularization method can be used to reduce effects of the measurement noise on the estimation error through adding a regularization function to the parameter estimation objective function. In this paper, we adopt the regularization function as the Euclidean norm of the difference between the values of the currently estimated parameters and the a priori parameter estimates. The effect of the regularization function on the outcome of parameter estimation is determined by a regularization factor. Based on a singular value decomposition of the sensitivity matrix of the structural response, it is shown that the optimal regularization factor is obtained by using the maximum singular value of the sensitivity matrix. This selection exhibits the condition where the effect of the a priori estimates on the solutions to the parameter estimation problem is minimal. The performance of the proposed algorithm is investigated in comparison with certain algorithms selected from the literature by using a numerical example.

• Communications for Statistical Applications and Methods
• /
• v.27 no.5
• /
• pp.535-546
• /
• 2020

In genetic association studies, pleiotropy is a phenomenon where a variant or a genetic region affects multiple traits or diseases. There have been many studies identifying cross-phenotype genetic associations. But, most of statistical approaches for detection of pleiotropy are based on individual tests where a single variant association with multiple traits is tested one at a time. These approaches fail to account for relations among correlated variants. Recently, multivariate regularization methods have been proposed to detect pleiotropy in analysis of high-dimensional genomic data. However, they suffer a problem of tuning parameter selection, which often results in either too many false positives or too small true positives. In this article, we applied selection probability to multivariate regularization methods in order to identify pleiotropic variants associated with multiple phenotypes. Selection probability was applied to individual elastic-net, unified elastic-net and multi-response elastic-net regularization methods. In simulation studies, selection performance of three multivariate regularization methods was evaluated when the total number of phenotypes, the number of phenotypes associated with a variant, and correlations among phenotypes are different. We also applied the regularization methods to a wild bean dataset consisting of 169,028 variants and 17 phenotypes.

• Journal of IKEEE
• /
• v.19 no.1
• /
• pp.33-40
• /
• 2015

Inverse problem in electrical impedance tomography (EIT) is highly ill-posed therefore prior information is used to mitigate the ill-posedness. Regularization methods are often adopted in solving EIT inverse problem to have satisfactory reconstruction performance. In solving the EIT inverse problem, iterative Gauss-Newton method is generally used due to its accuracy and fast convergence. However, its performance is still suboptimal and mainly depends on the selection of regularization parameter. Although, there are few methods available to determine the regularization parameter such as L-curve method they are sometimes not applicable for all cases. Moreover, regularization parameter is a scalar and it is fixed during iteration process. Therefore, in this paper, a novel method is used to determine the regularization parameter to improve reconstruction performance. Conductivity norm is calculated at each iteration step and it used to obtain the regularization parameter which is a diagonal matrix in this case. The proposed method is applied to human thorax imaging and the reconstruction performance is compared with traditional methods. From numerical results, improved performance of proposed method is seen as compared to conventional methods.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.15 no.1
• /
• pp.161-166
• /
• 2020

This paper suggests an extension of the unified Bayesian Tikhonov regularization method. The unified method establishes the relationship between Tikhonov regularization parameter and Bayesian hyper-parameters, and presents a formula for obtaining the regularization parameter using the maximum posterior probability and the evidence framework. When the dimension of the data matrix is m by n (m >= n), we derive that the total misfit has the range of m ± n instead of m. Thus the search range is extended from one to 2n + 1 integer points. Golden section search rather than linear one is applied to reduce the time. A new benchmark for optimizing relative error and new model selection criteria to target it are suggested. The experimental results show the effectiveness of the proposed method in the image restoration problem.

• Journal of Biomedical Engineering Research
• /
• v.42 no.6
• /
• pp.268-276
• /
• 2021

Functional near-infrared spectroscopy-based brain-computer interface (fNIRS-based BCI) has been receiving much attention. However, we are practically constrained to obtain a lot of fNIRS data by inherent hemodynamic delay. For this reason, when employing machine learning techniques, a problem due to the high-dimensional feature vector may be encountered, such as deteriorated classification accuracy. In this study, we employ an elastic net-based feature selection which is one of the embedded methods and demonstrate the utility of which by analyzing the results. Using the fNIRS dataset obtained from 18 participants for classifying brain activation induced by mental arithmetic and idle state, we calculated classification accuracies after performing feature selection while changing the parameter α (weight of lasso vs. ridge regularization). Grand averages of classification accuracy are 80.0 ± 9.4%, 79.3 ± 9.6%, 79.0 ± 9.2%, 79.7 ± 10.1%, 77.6 ± 10.3%, 79.2 ± 8.9%, and 80.0 ± 7.8% for the various values of α = 0.001, 0.005, 0.01, 0.05, 0.1, 0.2, and 0.5, respectively, and are not statistically different from the grand average of classification accuracy estimated with all features (80.1 ± 9.5%). As a result, no difference in classification accuracy is revealed for all considered parameter α values. Especially for α = 0.5, we are able to achieve the statistically same level of classification accuracy with even 16.4% features of the total features. Since elastic net-based feature selection can be easily applied to other cases without complicated initialization and parameter fine-tuning, we can be looking forward to seeing that the elastic-based feature selection can be actively applied to fNIRS data.

• Journal of the Computational Structural Engineering Institute of Korea
• /
• v.21 no.1
• /
• pp.73-82
• /
• 2008

In this paper, a new regularization method by parameter subset selection method is proposed based on the residual force vector for damage localization. Although subset selection using the fundamental modal characteristics as a residual function has been successful in detecting a single damage location, this method seems to have limited capabilities in the detection of multiple damage locations and typically requires cumbersome weighting values. The method is presented herein and considers cases in which damage detection must be achieved using incomplete measurements of the structural responses. Model expansion is incorporated to deal with this challenge. The unique advantage of employing the new regularization method is that it can reliably identify multiple damage locations. Through an illustrative example, the proposed damage detection method is demonstrated to be a reliable tool for identifying multiple damage locations for a planar truss structure.

• Proceedings of the KSEEG Conference
• /
• 2002.04a
• /
• pp.167-169
• /
• 2002

The extended Born, or localized nonlinear approximation of integral equation (IE) solution has been applied to inverting single-hole electromagnetic (EM) data using a cylindrically symmetric model. The extended Born approximation is less accurate than a full solution but much superior to the simple Born approximation. When applied to the cylindrically symmetric model with a vertical magnetic dipole source, however, the accuracy of the extended Born approximation is greatly improved because the electric field is scalar and continuous everywhere. One of the most important steps in the inversion is the selection of a proper regularization parameter for stability. Occam's inversion (Constable et al., 1987) is an excellent method for obtaining a stable inverse solution. It is extremely slow when combined with a differential equation method because many forward simulations are needed but suitable for the extended Born solution because the Green's functions, the most time consuming part in IE methods, are repeatedly re-usable throughout the inversion. In addition, the If formulation also readily contains a sensitivity matrix, which can be revised at each iteration at little expense. The inversion algorithm developed in this study is quite stable and fast even if the optimum regularization parameter Is sought at each iteration step. Tn this paper we show inversion results using synthetic data obtained from a finite-element method and field data as well.

• Smart Structures and Systems
• /
• v.31 no.3
• /
• pp.229-245
• /
• 2023

The absence of excitation measurements may pose a big challenge in the application of structural damage identification owing to the fact that substantial effort is needed to reconstruct or identify unknown input force. To address this issue, in this paper, an iterative strategy, a synergy of Tikhonov regularization method for force identification and modified Jaya algorithm (M-Jaya) for stiffness parameter identification, is developed for damage identification with partial output-only responses. On the one hand, the probabilistic clustering learning technique and nonlinear updating equation are introduced to improve the performance of standard Jaya algorithm. On the other hand, to deal with the difficulty of selection the appropriate regularization parameters in traditional Tikhonov regularization, an improved L-curve method based on B-spline interpolation function is presented. The applicability and effectiveness of the iterative strategy for simultaneous identification of structural damages and unknown input excitation is validated by numerical simulation on a 21-bar truss structure subjected to ambient excitation under noise free and contaminated measurements cases, as well as a series of experimental tests on a five-floor steel frame structure excited by sinusoidal force. The results from these numerical and experimental studies demonstrate that the proposed identification strategy can accurately and effectively identify damage locations and extents without the requirement of force measurements. The proposed M-Jaya algorithm provides more satisfactory performance than genetic algorithm, Gaussian bare-bones artificial bee colony and Jaya algorithm.