• International Journal of Fuzzy Logic and Intelligent Systems
• /
• 제8권3호
• /
• pp.196-201
• /
• 2008

Statistical learning theory has three analytical tools which are support vector machine, support vector regression, and support vector clustering for classification, regression, and clustering respectively. In general, their performances are good because they are constructed by convex optimization. But, there are some problems in the methods. One of the problems is the subjective determination of the parameters for kernel function and regularization by the arts of researchers. Also, the results of the learning machines are depended on the selected parameters. In this paper, we propose an efficient method for objective determination of the parameters of support vector clustering which is the clustering method of statistical learning theory. Using evolutionary algorithm and bootstrap method, we select the parameters of kernel function and regularization constant objectively. To verify improved performances of proposed research, we compare our method with established learning algorithms using the data sets form ucr machine learning repository and synthetic data.

• 대한전자공학회논문지
• /
• 제27권6호
• /
• pp.972-981
• /
• 1990

Many researchers have been interested in whether complex computational problems can be solved by the neural net or not. Especially, problems of early vision are integrated by Tikhonov's regularization theory. Regularization technique can be realized in resistive net. In this paper, we suggest the resistive net with upper and lower thresholder to be able to compute shape from shading and to solve its discontinuous problem. We simulate three algorithms-Horn's algorithm, resistive net and up-low thrwsholding net -with respect to three cases-fully boundary, boundary losing partly and noisy image. As being able to cope with crease and discontinuous parts, we get the good 3D shape from shading.

• International Journal of Fuzzy Logic and Intelligent Systems
• /
• 제14권4호
• /
• pp.249-255
• /
• 2014

We present a simple method to watermark three-dimensional (3D) triangular meshes that have been generated from the depth data of the Kinect sensor. In contrast to previous methods, which maintain the shape of 3D triangular meshes and decide the embedding place, requiring calculations of vertices and their neighbors, our method is based on selecting one of the coordinate axes. To maintain shape, we use discrete wavelet transform and constant regularization. We know that the watermarking system needs the information to be embedded; we used a text to provide that information. We used geometry attacks such as rotation, scales, and translation, to test the performance of this watermarking system. Performance parameters in this paper include the vertices error rate (VER) and bit error rate (BER). The results from the VER and BER indicate that using a correction term before the extraction process makes our system robust to geometry attacks.

• Communications for Statistical Applications and Methods
• /
• 제29권1호
• /
• pp.41-51
• /
• 2022

We forecast the US oil consumption level taking advantage of google trends. The google trends are the search volumes of the specific search terms that people search on google. We focus on whether proper selection of google trend terms leads to an improvement in forecast performance for oil consumption. As the forecast models, we consider the least absolute shrinkage and selection operator (LASSO) regression and the structured regularization method for large vector autoregressive (VAR-L) model of Nicholson et al. (2017), which select automatically the google trend terms and the lags of the predictors. An out-of-sample forecast comparison reveals that reducing the high dimensional google trend data set to a low-dimensional data set by the LASSO and the VAR-L models produces better forecast performance for oil consumption compared to the frequently-used forecast models such as the autoregressive model, the autoregressive distributed lag model and the vector error correction model.

• 대한기계학회:학술대회논문집
• /
• 대한기계학회 2007년도 춘계학술대회B
• /
• pp.3027-3032
• /
• 2007

To understand the complex phenomena performed in steam explosion, the fast and global measurement of the steam distribution is imperative for this extremely rapid transient stimulation of the bubble breakup and coalescence due to turbulent eddies and shock waves. TROI, the experimental facility requests more robust sensor system to meet this requirement. In Europe, researchers are prefer a X-ray method but this method is very expensive and has limited measurement range. There is an alternative technology such as ECT. Because of TROI's geometry, however, we need axial tomography method. This paper reviews image reconstruction algorethms for axial tomography, including Tikhonov regularization and iterative Tikhonov regularization. Axial tomography method is examined by simulation and experiment for typical permittivity distributions. Future works in axial tomography technology is discussed.

• 한국지능시스템학회:학술대회논문집
• /
• 한국퍼지및지능시스템학회 1998년도 The Third Asian Fuzzy Systems Symposium
• /
• pp.562-567
• /
• 1998

This paper examined the relations among the multidimensional linear interpolation(MDI) and fuzzy reasoning , and neural networks, and showed that an showed that an MDI is a special form of Tsukamoto's fuzzy reasoning and regularization networks in the perspective of fuzzy reasoning and neural networks, respectively. For this purposes, we proposed a special Tsukamoto's membership (STM) systemand triangular basis function (TBF) networks, Also we verified the condition when our proposed TBF becomes a well-known radial basis function (RBF).

• Communications for Statistical Applications and Methods
• /
• 제9권1호
• /
• pp.87-100
• /
• 2002

Poorly-posed problems in the balanced discriminant analysis was considered. We restrict consideration to the case of observations and the number of variables are the same and small. When these problems exist, we do not use a maximum likelihood estimates(MLE) to estimate covariance matrices. Instead of MLE, an alternative estimate for the covariance matrices are proposed. This alternative method make good use of two regularization parameters, $\lambda$} and $\gamma$. A new test rule for the discriminant function is suggested and examined via limited hut informative simulation study. From the simulation study, it is shown that the suggested test rule gives better test result than other previously suggested method in terms of error rate criterion.

• 방송공학회논문지
• /
• 제26권7호
• /
• pp.862-867
• /
• 2021

Although Deep Neural Networks (DNNs) have shown remarkable performance in various artificial intelligence fields, it is well known that DNNs are vulnerable to adversarial attacks. Since adversarial attacks are implemented by adding perturbations onto benign examples, increasing the sparsity of DNNs minimizes the propagation of errors to high-level layers. In this paper, unlike the traditional pruning scheme removing low magnitude weights, we eliminate high magnitude weights that are usually considered high absolute values, named 'reverse pruning' to ensure robustness. By conducting both theoretical and experimental analyses, we observe that reverse pruning ensures the robustness of DNNs. Experimental results show that our reverse pruning outperforms previous work with 29.01% in Top-1 accuracy on perturbed CIFAR-10. However, reverse pruning does not guarantee benign samples. To relax this problem, we further conducted experiments by adding a regularization term for the high magnitude weights. With adding the regularization term, we also applied conventional pruning to ensure the robustness of DNNs.

• Journal of information and communication convergence engineering
• /
• 제21권2호
• /
• pp.139-144
• /
• 2023

Engineers prefer deep neural networks (DNNs) for solving computer vision problems. However, DNNs pose two major problems. First, neural networks require large amounts of well-labeled data for training. Second, the covariate shift problem is common in computer vision problems. Domain adaptation has been proposed to mitigate this problem. Recent work on adversarial-learning-based unsupervised domain adaptation (UDA) has explained transferability and enabled the model to learn robust features. Despite this advantage, current methods do not guarantee the distinguishability of the latent space unless they consider class-aware information of the target domain. Furthermore, source and target examples alone cannot efficiently extract domain-invariant features from the encoded spaces. To alleviate the problems of existing UDA methods, we propose the mixup regularization in adversarial discriminative domain adaptation (ADDA) method. We validated the effectiveness and generality of the proposed method by performing experiments under three adaptation scenarios: MNIST to USPS, SVHN to MNIST, and MNIST to MNIST-M.

• 대한기계학회:학술대회논문집
• /
• 대한기계학회 2004년도 추계학술대회
• /
• pp.1288-1293
• /
• 2004

Inverse radiation problems are solved for estimating the boundary conditions such as temperature distribution and wall emissivity in axisymmetric absorbing, emitting and scattering medium, given the measured incident radiative heat fluxes. Various regularization methods, such as hybrid genetic algorithm, conjugate-gradient method and Newton method, were adopted to solve the inverse problem, while discussing their features in terms of estimation accuracy and computational efficiency. Additionally, we propose a new combined approach of adopting the genetic algorithm as an initial value selector, whereas using the conjugate-gradient method and Newton method to reduce their dependence on the initial value.