• Smart Structures and Systems
• /
• v.10 no.4_5
• /
• pp.459-470
• /
• 2012

This paper summarizes a study for the modal analysis and model updating conducted using the monitoring data obtained from the Canton Tower of 610 m tall, which was established as an international benchmark problem by the Hong Kong Polytechnic University. Modal properties of the tower were successfully identified using frequency domain decomposition and stochastic subspace identification methods. Finite element model updating using the measurement data was further performed to reduce the modal property differences between the measurements and those of the finite element model. Over-fitting during the model updating was avoided by using an optimization scheme with a regularization term.

• Journal of Information Processing Systems
• /
• v.15 no.6
• /
• pp.1449-1461
• /
• 2019

Face sketch synthesis plays an important role in public security and digital entertainment. In this paper, we present a novel face sketch synthesis method via local similarity and nonlocal similarity regularization terms. The local similarity can overcome the technological bottlenecks of the patch representation scheme in traditional learning-based methods. It improves the quality of synthesized sketches by penalizing the dissimilar training patches (thus have very small weights or are discarded). In addition, taking the redundancy of image patches into account, a global nonlocal similarity regularization is employed to restrain the generation of the noise and maintain primitive facial features during the synthesized process. More robust synthesized results can be obtained. Extensive experiments on the public databases validate the generality, effectiveness, and robustness of the proposed algorithm.

• Geomechanics and Engineering
• /
• v.12 no.2
• /
• pp.197-210
• /
• 2017

This paper focuses on the regularization of structural configurations by employing meta-heuristic optimization algorithms such as Particle Swarm Optimization (PSO) and Biogeography-Based Optimization (BBO). The regularization of structural configuration means obtaining a structure whose members have equal or almost equal lengths, or whose member's lengths are based on a specific pattern; which in this case, by changing the length of these elements and reducing the number of different profiles of needed members, the construction of the considered structure can be made easier. In this article, two different objective functions have been used to minimize the difference between member lengths with a specific pattern. It is found that by using a small number of iterations in these optimization methods, a structure made of equal-length members can be obtained.

• Proceedings of the Korean Vacuum Society Conference
• /
• 2010.02a
• /
• pp.472-472
• /
• 2010

The tomography has played a key role in tokamak plasma diagnostics for image reconstruction. The Phillips-Tikhonov (P-T) regularization method was attempted in this work to reconstruct cross-sectional phantom images of the plasma by minimizing the gradient between adjacent pixel data. Recent studies about the comparison of the several tomographic reconstruction methods showed that the P-T method produced more accurate results. We have studied existing Laplacian matrix used in Phillips-Tikhonov regularization method and developed modified Laplacian matrix (Modified L). The comparison of the reconstruction result by the modified L and existing L showed that modified L produced more accurate result. The difference was significantly pronounced when a portion of plasma was reconstructed. These results can be utilized in the Edge Plasma diagnostics; especially in divertor diagnostics on tokamak a large impact is expected. In addition, accurate reconstruction results from received data in only one direction were confirmed through phantom test by using P-T method with modified L. These results can be applied to the tangentially viewing pin-hole camera diagnostics on tokamak.

• Journal of Korea Multimedia Society
• /
• v.24 no.1
• /
• pp.13-20
• /
• 2021

A Wasserstein GAN(WGAN), optimum in terms of minimizing Wasserstein distance, still suffers from inconsistent convergence or unexpected output due to inherent learning instability. It is widely known some kinds of restriction on the discriminative function should be considered to solve such problems, which implies the importance of Lipschitz continuity. Unfortunately, there are few known methods to satisfactorily maintain the Lipschitz continuity of the discriminative function. In this paper we propose techniques to stably maintain the Lipschitz continuity of the discriminative function by adding effective regularization terms to the objective function, which limit the magnitude of the gradient vectors of the discriminator to one or less. Extensive experiments are conducted to evaluate the performance of the proposed techniques, which shows the single-sided penalty improves convergence compared with the gradient penalty at the early learning process, while the proposed additional penalty increases inception scores by 0.18 after 100,000 number of learning.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.18 no.10
• /
• pp.75-80
• /
• 2017

In this paper, we describe the effect of the regularization method and the network with identity mapping on the performance of the activation functions in deep convolutional neural networks. The activation functions act as nonlinear transformation. In early convolutional neural networks, a sigmoid function was used. To overcome the problem of the existing activation functions such as gradient vanishing, various activation functions were developed such as ReLU, Leaky ReLU, parametric ReLU, and ELU. To solve the overfitting problem, regularization methods such as dropout and batch normalization were developed on the sidelines of the activation functions. Additionally, data augmentation is usually applied to deep learning to avoid overfitting. The activation functions mentioned above have different characteristics, but the new regularization method and the network with identity mapping were validated only using ReLU. Therefore, we have experimentally shown the effect of the regularization method and the network with identity mapping on the performance of the activation functions. Through this analysis, we have presented the tendency of the performance of activation functions according to regularization and identity mapping. These results will reduce the number of training trials to find the best activation function.

• Journal of the Korean Society for Industrial and Applied Mathematics
• /
• v.18 no.2
• /
• pp.143-156
• /
• 2014

This paper aims to give a quick view on denoising without comprehensive details. Denoising can be understood as removing unwanted parts in signals and images. Noise incorporates intrinsic random fluctuations in the data. Since noise is ubiquitous, denoising methods and models are diverse. Starting from what noise means, we briefly discuss a denoising model as maximum a posteriori estimation and relate it with a variational form or energy model. After that we present a few major branches in image and signal processing; filtering, shrinkage or thresholding, regularization and data adapted methods, although it may not be a general way of classifying denoising methods.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.52 no.12
• /
• pp.158-164
• /
• 2015

Lane detection is one of the key parts among autonomous vehicle technologies because lane keeping and path planning are based on lane detection. Camera is used for lane detection but there are severe limitations such as narrow field of view and effect of illumination. On the other hands, Lidar sensor has the merits of having large field of view and being little influenced by illumination because it uses intensity information. Existing researches that use methods such as Hough transform, histogram hardly handle multiple lanes in the co-occuring situation of lanes and road marking. In this paper, we propose a method based on RANSAC and regularization which provides a stable and precise detection result in the co-occuring situation of lanes and road marking in highway scenarios. This is performed by precise lane point extraction using circular model RANSAC and regularization aided least square fitting. Through quantitative evaluation, we verify that the proposed algorithm is capable of multi lane detection with high accuracy in real-time on our own acquired road data.

• Smart Structures and Systems
• /
• v.18 no.5
• /
• pp.839-864
• /
• 2016

This paper presents an efficient method for updating the structural finite element model. Model updating is performed through minimizing the difference between the recorded acceleration of a real damaged structure and a hypothetical damaged one. This is performed by updating physical parameters (module of elasticity in this study) in each step using iterative process of modified nonlinear conjugate gradient (M-NCG) and modified Broyden-Fletcher-Goldfarb-Shanno algorithm (M-BFGS) separately. These algorithms are based on sensitivity analysis and provide a solution for nonlinear damage detection problem. Three illustrative test examples are considered to assess the performance of the proposed method. Finally, it is demonstrated that the proposed method is satisfactory for detecting the location and ratio of structural damage in presence of noise.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.13 no.3
• /
• pp.1403-1417
• /
• 2019

With the rapid growth of multimedia data, research on cross-media feature learning has significance in many applications, such as multimedia search and recommendation. Existing methods are sensitive to noise and edge information in multimedia data. In this paper, we propose a semi-supervised method for cross-media feature learning by means of $L_{2,q}$ norm to improve the performance of cross-media retrieval, which is more robust and efficient than the previous ones. In our method, noise and edge information have less effect on the results of cross-media retrieval and the dynamic patch information of multimedia data is employed to increase the accuracy of cross-media retrieval. Our method can reduce the interference of noise and edge information and achieve fast convergence. Extensive experiments on the XMedia dataset illustrate that our method has better performance than the state-of-the-art methods.