• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.17 no.12
• /
• pp.2793-2799
• /
• 2013

This paper proposes an effective feature compensation method to improve speech recognition performance in time-varying background noise condition. The proposed method employs principal component analysis to improve the variational model composition method. The proposed method is employed to generate multiple environmental models for the PCGMM-based feature compensation scheme. Experimental results prove that the proposed scheme is more effective at improving speech recognition accuracy in various SNR conditions of background music, compared to the conventional front-end methods. It shows 12.14% of average relative improvement in WER compared to the previous variational model composition method.

• Communications for Statistical Applications and Methods
• /
• v.29 no.1
• /
• pp.27-40
• /
• 2022

In this paper, the focus on the removal noise in the binary image based on the variational Bayesian method with the Ising model. The observation and the latent variable are the degraded image and the original image, respectively. The posterior distribution is built using the Markov random field and the Ising model. Estimating the posterior distribution is the same as reconstructing a degraded image. MCMC and variational Bayesian inference are two methods for estimating the posterior distribution. However, for the sake of computing efficiency, we adapt the variational technique. When the image is restored, the iterative method is used to solve the recursive problem. Since there are three model parameters in this paper, restoration is implemented using the VECM algorithm to find appropriate parameters in the current state. Finally, the restoration results are shown which have maximum peak signal-to-noise ratio (PSNR) and evidence lower bound (ELBO).

• Journal of the Korean Society for Industrial and Applied Mathematics
• /
• v.18 no.2
• /
• pp.129-142
• /
• 2014

We propose a variational segmentation model based on statistical information of intensities in an image. The model consists of both a local region-based energy and a global region-based energy in order to handle misclassification which happens in a typical statistical variational model with an assumption that an image is a mixture of two Gaussian distributions. We find local ambiguous regions where misclassification might happen due to a small difference between two Gaussian distributions. Based on statistical information restricted to the local ambiguous regions, we design a local region-based energy in order to reduce the misclassification. We suggest an algorithm to avoid the difficulty of the Euler-Lagrange equations of the proposed variational model.

• Journal of Korean Society for Atmospheric Environment
• /
• v.19 no.E2
• /
• pp.75-81
• /
• 2003

Variational data assimilation, which is recently introduced to the air quality modeling, is a promising tool for obtaining optimal estimates of initial conditions and other important parameters such as emission and deposition rates. In this paper. two advanced techniques for variational data assimilation, based on the adjoint and quasi-inverse methods, are tested for a simple air quality problem. The four-dimensional variational assimilation (4D-Var) requires to run an adjoint model to provide the gradient information in an iterative minimization process, whereas the inverse 3D-Var (I3D-Var) seeks for optimal initial conditions directly by running a quasi -inverse model. For a process with small dissipation, I3D-Vu outperforms 4D-Var in both computing time and accuracy. Hybrid application which combines I3D-Var and standard 4D-Var is also suggested for efficient data assimilation in air quality problems.

• ETRI Journal
• /
• v.33 no.6
• /
• pp.914-923
• /
• 2011

This paper concerns robust and reliable speaker model training for text-independent speaker verification. The baseline speaker modeling approach is the Gaussian mixture model (GMM). In text-independent speaker verification, the amount of speech data may be different for speakers. However, we still wish the modeling approach to perform equally well for all speakers. Besides, the modeling technique must be least vulnerable against unseen data. A traditional approach for GMM training is expectation maximization (EM) method, which is known for its overfitting problem and its weakness in handling insufficient training data. To tackle these problems, variational approximation is proposed. Variational approaches are known to be robust against overtraining and data insufficiency. We evaluated the proposed approach on two different databases, namely KING and TFarsdat. The experiments show that the proposed approach improves the performance on TFarsdat and KING databases by 0.56% and 4.81%, respectively. Also, the experiments show that the variationally optimized GMM is more robust against noise and the verification error rate in noisy environments for TFarsdat dataset decreases by 1.52%.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.9 no.5
• /
• pp.1249-1254
• /
• 2008

The crucial part of graphical model is to compute the posterior distribution of parameters plus with the hidden variables given the observed data. In this paper, implementation of variational Bayes method for Gaussian mixture model and derivation of factorial variational approximation have been proposed. This result can be used for data analysis tasks like information retrieval or data visualization.

• Advances in aircraft and spacecraft science
• /
• v.1 no.1
• /
• pp.93-123
• /
• 2014

A variational asymptotic composite beam model has been developed for thermoelastic analysis. Composite beams, including sandwich structure and laminates, under different boundary conditions are examined. Previously developed beam model, which is based on variational-asymptotic method, is extended to incorporate temperature-dependent materials experiencing large temperature changes. The recovery relations have been derived so that the temperatures, heat fluxes, stresses, and strains can be recovered over the cross-section. The present theory is implemented into the computer program VABS (Variational Asymptotic Beam Sectional analysis). Numerical results are compared with the 3D analysis for the purpose of demonstrating advantages of the present theory and use of VABS.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.10 no.3
• /
• pp.1182-1194
• /
• 2016

The single image dehazing algorithm in existence can satisfy the demand only for improving either the effectiveness or efficiency. In order to solve the problem, a novel variational framework for single image dehazing based on restoration is proposed. Firstly, the initial atmospheric scattering model is transformed to meet the kimmel's Retinex variational model. Then, the green light component of image is considered as an input of the variational framework, which is generated by the sensitivity of green wavelength. Finally, the atmospheric transmission map is achieved by multi-resolution pyramid reduction to improve the visual effect of the results. Experimental results demonstrate that the proposed method can remove haze effectively with less memory consumption.

• Journal of Internet Computing and Services
• /
• v.22 no.2
• /
• pp.1-9
• /
• 2021

Traditionally, most malicious codes have been analyzed using feature information extracted by domain experts. However, this feature-based analysis method depends on the analyst's capabilities and has limitations in detecting variant malicious codes that have modified existing malicious codes. In this study, we propose a ResNet-Variational AutoEncder-based variant malware classification method that can classify a family of variant malware without domain expert intervention. The Variational AutoEncoder network has the characteristics of creating new data within a normal distribution and understanding the characteristics of the data well in the learning process of training data provided as input values. In this study, important features of malicious code could be extracted by extracting latent variables in the learning process of Variational AutoEncoder. In addition, transfer learning was performed to better learn the characteristics of the training data and increase the efficiency of learning. The learning parameters of the ResNet-152 model pre-trained with the ImageNet Dataset were transferred to the learning parameters of the Encoder Network. The ResNet-Variational AutoEncoder that performed transfer learning showed higher performance than the existing Variational AutoEncoder and provided learning efficiency. Meanwhile, an ensemble model, Stacking Classifier, was used as a method for classifying variant malicious codes. As a result of learning the Stacking Classifier based on the characteristic data of the variant malware extracted by the Encoder Network of the ResNet-VAE model, an accuracy of 98.66% and an F1-Score of 98.68 were obtained.

• The Korean Journal of Applied Statistics
• /
• v.36 no.2
• /
• pp.115-127
• /
• 2023

Multinomial probit model is a popular model for multiclass classification and choice model. Markov chain Monte Carlo (MCMC) method is widely used for estimating multinomial probit model, but its computational cost is high. However, it is well known that variational Bayesian approximation is more computationally efficient than MCMC, because it uses subsets of samples. In this study, we describe multinomial probit model with Gaussian process classification and how to employ variational Bayesian approximation on the model. This study also compares the results of variational Bayesian multinomial probit model to the results of naive Bayes, K-nearest neighbors and support vector machine for the UCI mice protein expression level data.