• The Journal of the Acoustical Society of Korea
• /
• v.23 no.1
• /
• pp.75-81
• /
• 2004

Gaussian selection (GS) is a popular approach in the continuous density hidden Markov model for fast decoding. It enables fast likelihood computation by reducing the number of Gaussian components calculated. In this paper, we propose a new GS method for the phonetic tied-mixture (PTM) hidden Markov models. The PTM model can represent each state of the same topological location with a shared set of Gaussian mixture components and contort dependent weights. Thus the proposed method imposes constraint on the weights as well as the number of Gaussian components to reduce the computational load. Experimental results show that the proposed method reduces the percentage of Gaussian computation to 16.41%, compared with 20-30% for the conventional GS methods, with little degradation in recognition.

• Proceedings of the Korean Radioactive Waste Society Conference
• /
• 2004.06a
• /
• pp.368-376
• /
• 2004

The estimation of the diffusion coefficients of the Gaussian plume model and the release rate by assimilation of tracer-gas measurements on Younggwang site was tested. Diffusion coefficients were modified by linear programming of both the measurements and the simulated using the Gaussian plume model. The application of the modified diffusion coefficients improved the prediction ability of the Gaussian plume model on both 3 km and 8 km arc lines. And, the release rate of tracer gas was estimated using least squares method. The optimal source rate was estimated by minimizing the errors between the measured concentrations and the computed ones by the Gaussian plume model. The obtained release rate showed a good agreement with the real release rate of the Younggwang experiment in 24%.

• Journal of Korean Society of Industrial and Systems Engineering
• /
• v.39 no.1
• /
• pp.64-72
• /
• 2016

In this paper, we utilize a Gaussian process to predict the power consumption in the air-conditioning system. As the power consumption in the air-conditioning system takes a form of a time-series and the prediction of the power consumption becomes very important from the perspective of the efficient energy management, it is worth to investigate the time-series model for the prediction of the power consumption. To this end, we apply the Gaussian process to predict the power consumption, in which the Gaussian process provides a prior probability to every possible function and higher probabilities are given to functions that are more likely consistent with the empirical data. We also discuss how to estimate the hyper-parameters, which are parameters in the covariance function of the Gaussian process model. We estimated the hyper-parameters with two different methods (marginal likelihood and leave-one-out cross validation) and obtained a model that pertinently describes the data and the results are more or less independent of the estimation method of hyper-parameters. We validated the prediction results by the error analysis of the mean relative error and the mean absolute error. The mean relative error analysis showed that about 3.4% of the predicted value came from the error, and the mean absolute error analysis confirmed that the error in within the standard deviation of the predicted value. We also adopt the non-parametric Wilcoxon's sign-rank test to assess the fitness of the proposed model and found that the null hypothesis of uniformity was accepted under the significance level of 5%. These results can be applied to a more elaborate control of the power consumption in the air-conditioning system.

• Speech Sciences
• /
• v.5 no.1
• /
• pp.7-21
• /
• 1999

This paper proposes an improved training procedure in speech recognition based on the continuous density of the Hidden Markov Model (CDHMM). Of the three parameters (initial state distribution probability, state transition probability, output probability density function (p.d.f.) of state) governing the CDHMM model, we focus on the third parameter and propose an efficient algorithm that determines the p.d.f. of each state. It is known that the resulting CDHMM model converges to a local maximum point of parameter estimation via the iterative Expectation Maximization procedure. Specifically, we propose two independent algorithms that can be embedded in the segmental K -means training procedure by replacing relevant key steps; the adaptation of the number of mixture Gaussian p.d.f. and the initialization using the CDHMM parameters previously estimated. The proposed adaptation algorithm searches for the optimal number of mixture Gaussian humps to ensure that the p.d.f. is consistently re-estimated, enabling the model to converge toward the global maximum point. By applying an appropriate threshold value, which measures the amount of collective changes of weighted variances, the optimized number of mixture Gaussian branch is determined. The initialization algorithm essentially exploits the CDHMM parameters previously estimated and uses them as the basis for the current initial segmentation subroutine. It captures the trend of previous training history whereas the uniform segmentation decimates it. The recognition performance of the proposed adaptation procedures along with the suggested initialization is verified to be always better than that of existing training procedure using fixed number of mixture Gaussian p.d.f.

• Smart Media Journal
• /
• v.4 no.3
• /
• pp.50-55
• /
• 2015

In this paper, we present background modeling method based on Gaussian mixture model to subtract background for night-time video surveillance. In night-time video, it is hard work to distinguish the object from the background because a background pixel is similar to a object pixel. To solve this problem, we change the pixel of input frame to more advantageous value to make the Gaussian mixture model using scaled histogram stretching in preprocessing step. Using scaled pixel value of input frame, we then exploit GMM to find the ideal background pixelwisely. In case that the pixel of next frame is not included in any Gaussian, the matching test in old GMM method ignores the information of stored background by eliminating the Gaussian distribution with low weight. Therefore we consider the stacked data by applying the difference between the old mean and new pixel intensity to new mean instead of removing the Gaussian with low weight. Some experiments demonstrate that the proposed background modeling method shows the superiority of our algorithm effectively.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.17 no.8
• /
• pp.1976-1995
• /
• 2023

The anchor-free object detector CenterNet regards the object as a center point and predicts it based on the Gaussian circle region. For each object's center point, CenterNet directly regresses the width and height of the objects and finally gets the boundary range of the objects. However, the critical range of the object's center point can not be accurately limited by using the Gaussian circle region to constrain the prediction region, resulting in many low-quality centers' predicted values. In addition, because of the large difference between the width and height of different objects, directly regressing the width and height will make the model difficult to converge and lose the intrinsic relationship between them, thereby reducing the stability and consistency of accuracy. For these problems, we proposed a center point prediction method based on the Gaussian elliptic region and a size component regression method based on the small solution space. First, we constructed a Gaussian ellipse region that can accurately predict the object's center point. Second, we recode the width and height of the objects, which significantly reduces the regression solution space and improves the convergence speed of the model. Finally, we jointly decode the predicted components, enhancing the internal relationship between the size components and improving the accuracy consistency. Experiments show that when using CenterNet as the improved baseline and Hourglass-104 as the backbone, on the MS COCO dataset, our improved model achieved 44.7%, which is 2.6% higher than the baseline.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.13 no.8
• /
• pp.4285-4299
• /
• 2019

Adversarial attack is a technique that causes a malfunction of classification models by adding noise that cannot be distinguished by humans, which poses a threat to a deep learning model. In this paper, we propose an efficient method to detect adversarial images using Gaussian process regression. Existing deep learning-based adversarial detection methods require numerous adversarial images for their training. The proposed method overcomes this problem by performing classification based on the statistical features of adversarial images and clean images that are extracted by Gaussian process regression with a small number of images. This technique can determine whether the input image is an adversarial image by applying Gaussian process regression based on the intermediate output value of the classification model. Experimental results show that the proposed method achieves higher detection performance than the other deep learning-based adversarial detection methods for powerful attacks. In particular, the Gaussian process regression-based detector shows better detection performance than the baseline models for most attacks in the case with fewer adversarial examples.

• The Bulletin of The Korean Astronomical Society
• /
• v.46 no.2
• /
• pp.68.3-69
• /
• 2021

We present a novel profile stacking technique based on optimal profile decomposition of a 3D spectral line data cube, and its performance test using the HI data cubes of sample galaxies from HI galaxy surveys, THINGS and LITTLE THINGS. Compared to the previous approach which aligns all the spectra of a cube using their central velocities derived from either moment analysis, single Gaussian or hermite h3 polynomial fitting, the new method makes a profile decomposition of the profiles from which an optimal number of single Gaussian components is derived for each profile. The so-called superprofile which is derived by co-adding all the aligned profiles from which the other Gaussian models are subtracted is found to have weaker wings compared to the ones constructed in a typical manner. This could be due to the reduced number of asymmetric profiles in the new method. A practical test made on the HI data cubes of the THINGS and LITTLE THINGS galaxies shows that our new method can extract more mass of kinematically cold HI components in the galaxies than the previous results. Additionally, we fit a double Gaussian model to the superprofiles whose S/N is boosted, and quantify not only their profile shapes but derive the ratio of the Gaussian model parameters, such as the intensity ratio and velocity dispersion ratio of the narrower and broader Gaussian components. We discuss how the superprofile properties of the sample galaxies are correlated with their other physical properties, including star formation rate, stellar mass, metallicity, and gas mass.

• Wind and Structures
• /
• v.31 no.6
• /
• pp.549-560
• /
• 2020

Methods for stochastic simulation of non-Gaussian wind pressure have increasingly addressed the efficiency and accuracy contents to offer an accurate description of the extreme value estimation of the long-span and high-rise structures. This paper presents a linear prediction and z-transform (LPZ) based Cumulative distribution function (CDF) mapping algorithm for the simulation of multivariate non-Gaussian fluctuating wind pressure. The new algorithm generates realizations of non-Gaussian with prescribed marginal probability distribution function (PDF) and prescribed spectral density function (PSD). The inverse linear prediction and z-transform function (ILPZ) is deduced. LPZ is improved and applied to non-Gaussian wind pressure simulation for the first time. The new algorithm is demonstrated to be efficient, flexible, and more accurate in comparison with the FFT-based method and Hermite polynomial model method in two examples for transverse softening and longitudinal hardening non-Gaussian wind pressures.

• Proceedings of the Korean Institute of Communication Sciences Conference
• /
• 2006.07a
• /
• pp.434-434
• /
• 2006