• Journal of the Korean GEO-environmental Society
• /
• v.10 no.5
• /
• pp.49-57
• /
• 2009

In this study, one of the complicated geotechnical problem, compression index was predicted by a artificial neural network method of Levenberg-Marquardt (LM) algorithm. Predicted values were compared and evaluated by the results of the Back Propagation (BP) method, which is used extensively in geotechnical engineering. Also two different results were compared with experimental values estimated by verified experimental methods in order to evaluate the accuracy of each method. The results from experimental method generally showed higher error than the results of both artificial neural network method. The predicted compression index by LM algorithm showed better comprehensive results than BP algorithm in terms of convergence, but accuracy was similar each other.

• Journal of the Korea Society of Computer and Information
• /
• v.14 no.8
• /
• pp.11-18
• /
• 2009

This paper proposes the way of improving learning speed in Levenberg-Marquardt algorithm using the principal submatrix of Jacobian matrix. The Levenberg-Marquardt learning uses Jacobian matrix for Hessian matrix to get the second derivative of an error function. To make the Jacobian matrix an invertible matrix. the Levenberg-Marquardt learning must increase or decrease ${\mu}$ and recalculate the inverse matrix of the Jacobian matrix due to these changes of ${\mu}$. Therefore, to have the proper ${\mu}$, we create the principal submatrix of Jacobian matrix and set the ${\mu}$ as the eigenvalues sum of the principal submatrix. which can make learning speed improve without calculating an additional inverse matrix. We also showed that our method was able to improve learning speed in both a generalized XOR problem and a handwritten digit recognition problem.

• Journal of the Korea Society of Computer and Information
• /
• v.15 no.4
• /
• pp.57-63
• /
• 2010

The damping parameter of Levenberg-Marquardt algorithm switches between error backpropagation and Gauss-Newton learning and affects learning speed. Fixing the damping parameter induces some oscillation of error and decreases learning speed. Therefore, we propose the way of a variable damping parameter with referring to the alternation of error. The proposed method makes the damping parameter increase if error rate is large and makes it decrease if error rate is small. This method so plays the role of momentum that it can improve learning speed. We tested both iris recognition and wine recognition for this paper. We found out that this method improved learning speed in 67% cases on iris recognition and in 78% cases on wine recognition. It was also showed that the oscillation of error by the proposed way was less than those of other algorithms.

• The Journal of Korean Institute of Electromagnetic Engineering and Science
• /
• v.14 no.8
• /
• pp.803-808
• /
• 2003

Several dielectric pipes buried in the lossy half space are reconstructed from the scattered fields measured along the interface between the air and the lossy ground. Iterative inversion method by using the hybrid optimization algorithm combining the genetic and the Levenberg-Marquardt algorithm enables us to find the positions, the sizes, and the medium parameters such as the permittivities and the conductivities of the buried pipes as well as those of the background lossy half space even when the dielectric pipes are close together. Illposedness of the inversion caused by the errors in the measured scattered fields are regularized by filtering the evanescent modes of the scattered fields out.

• The Journal of Korean Institute of Electromagnetic Engineering and Science
• /
• v.8 no.5
• /
• pp.534-544
• /
• 1997

The permittivity distribution of a two-dimensional high-contrast object with large size, which leads to the global minimum of cost function, is reconstructed by iteratively using the hybrid algorithm of Levenberg-magquardt algorithm(LMA) plus Genetic Algorithm(GA). The scattered fields calculated in a cost function are expanded in angular spectral modes, of which only effective propagating modes are used. The definition of cost function based on the effective propagating modes enables us to formulate the minimum number of incident waves for the reconstruction of object. It is numerically shown that LMA has an advantage of fast convergence but can't reconstruct a high-contrast object with large size and GA can reconstruct a high-contrast object with large size but has an disadvantage of slow convergence, whereas an inverse scattering technique using the hybrid algorithm adopts only advantages of LMA and GA.

• Smart Media Journal
• /
• v.3 no.2
• /
• pp.9-13
• /
• 2014

The Levenberg-Marquardt (LM) algorithm is the most widely used fitting algorithm. It outperforms simple gradient descent and other conjugate gradient methods in a wide variety of problems. Based on the work of paper[1], we propose a modified Levenberg-Marquardt algorithm for better performance of mobile system. The LM parameter at the $k_{th}$ iteration is chosen ${\mu}=A{\bullet}{\parallel}f(x){\parallel}{\bullet}I$ where f is the residual function, and J is the Jacobi of f. In this paper, we show this method is more efficient than traditional method under the situation that the system iteration is limited.

• Journal of the Korean Data and Information Science Society
• /
• v.23 no.5
• /
• pp.1027-1035
• /
• 2012

This paper proposes a new Levenberg-Marquardt algorithm that is accelerated by adjusting a Jacobian matrix and a quasi-Hessian matrix. The proposed method partitions the Jacobian matrix into block matrices and employs the inverse of a partitioned matrix to find the inverse of the quasi-Hessian matrix. Our method can avoid expensive operations and save memory in calculating the inverse of the quasi-Hessian matrix. It can shorten the training time for fast convergence. In our results tested in a large application, we were able to save about 20% of the training time than other algorithms.

• Proceedings of the KIEE Conference
• /
• 2001.11c
• /
• pp.204-208
• /
• 2001

In this paper, we induce the linear transfer function of Downcomer water level of NPP(Nuclear Power Plant) Steam Generator using Genetic Algorithm and Levenberg-Marquardt Algorithm. The characteristic of NPP S/G mechanism is so high-non-linear that it is hard to achieve mathematical expression. So we use non-mathematical Algorithms to get the model function of NPP S/G water level. S/G level controller would be designed with this transfer function as the plant.

• Geophysics and Geophysical Exploration
• /
• v.22 no.4
• /
• pp.195-201
• /
• 2019

The logarithmic objective function used in waveform inversion minimizes the logarithmic differences between the observed and modeled data. Laplace-domain waveform inversions usually adopt the logarithmic objective function and the diagonal elements of the pseudo-Hessian for optimization. In this case, we apply the Levenberg-Marquardt algorithm to prevent the diagonal elements of the pseudo-Hessian from being zero or near-zero values. In this study, we analyzed the diagonal elements of the pseudo-Hessian of the logarithmic objective function and showed that there is no zero or near-zero value in the diagonal elements of the pseudo-Hessian for acoustic waveform inversion in the Laplace domain. Accordingly, we do not need to apply the Levenberg-Marquardt algorithm when we regularize the gradient direction using the pseudo-Hessian of the logarithmic objective function. Numerical examples using synthetic and field datasets demonstrate that we can obtain inversion results without applying the Levenberg-Marquardt method.

• The Transactions of The Korean Institute of Electrical Engineers
• /
• v.58 no.9
• /
• pp.1769-1774
• /
• 2009

An algorithm for extracting the BJT DC model parameter values for SPICE model is proposed. The nonlinear optimization method for analyzing the device I-V data using the Levenberg-Marquardt algorithm is proposed and the method for calculating initial conditions of model parameters to improve the convergence characteristics is proposed. The base current and collector current obtained from the proposed method shows the root mean square error of 6.04% compared with the measured data of the PNP BJT named 2SA1980.