• ETRI Journal
• /
• 제41권1호
• /
• pp.109-116
• /
• 2019

This paper explores new techniques that are based on a hidden-layer linear transformation for fast speaker adaptation used in deep neural networks (DNNs). Conventional methods using affine transformations are ineffective because they require a relatively large number of parameters to perform. Meanwhile, methods that employ singular-value decomposition (SVD) are utilized because they are effective at reducing adaptive parameters. However, a matrix decomposition is computationally expensive when using online services. We propose the use of an extended diagonal linear transformation method to minimize adaptation parameters without SVD to increase the performance level for tasks that require smaller degrees of adaptation. In Korean large vocabulary continuous speech recognition (LVCSR) tasks, the proposed method shows significant improvements with error-reduction rates of 8.4% and 17.1% in five and 50 conversational sentence adaptations, respectively. Compared with the adaptation methods using SVD, there is an increased recognition performance with fewer parameters.

• 대한기계학회논문집A
• /
• 제29권6호
• /
• pp.842-851
• /
• 2005

In this paper, a new method for constrained optimization of noisy functions is proposed. In approximate optimization using response surface methods, if constraints have severe noise, the approximate feasible region defined by approximate constraints is apt to include some of the infeasible region defined by actual constraints. This can cause the approximate optimum to converge into the infeasible region. In the proposed method, the approximate optimization is performed with the approximate constraints shifted by their deviations, which are calculated using a diagonal quadratic response surface method. This can prevent the approximate optimum from converging into the infeasible region. To fit the objective and constraints into diagonal quadratic models, we select the center and 4 additional points along each axis of design variables as experimental points. The deviation of each function is calculated using the differences between the real and approximate function values at the experimental points. A sequential approximate optimization technique based on the trust region algorithm is adopted to manage approximate models. The proposed approach is validated by solving some design problems. The results of the problems show the effectiveness of the proposed method.

• 한국정보통신학회논문지
• /
• 제14권7호
• /
• pp.1581-1588
• /
• 2010

대부분의 디지털 촬영 장치는 비용 및 속도 상의 이점을 위해 Color Filter Array(CFA)를 포함하고 있는 단일 이미지 센서를 사용한다. 따라서 완전한 컬러 영상으로 복원하기 위하여 다양한 컬러 보간 방법이 개발되고 있다. 본 논문은 이러한 컬러 보간 방법 중, 방향성 필터를 이용하여 수직, 수평, 대각선 방향의 에지를 고려한 컬러 보간 방법에 관한 것이다. 하드웨어 구현을 위해 하드웨어 자원의 효율성을 고려하였다. Kodak 테스트 영상 24장으로 실험하여 기존 방법과 비교함으로써 제안한 방법의 성능을 확인하였다. Verilog HDL로 구현하였으며, Virtex4 FPGA 보드와 CMOS 이미지 센서를 이용하여 검증하였다.

• 대한기계학회논문집B
• /
• 제23권6호
• /
• pp.703-712
• /
• 1999

This paper suggests an automatic elliptic grid generation method that is well-suited for the numerical mapping of complex geometries which are easily obtained from general CAD programs. An LBLADI solver is used for the governing mapping equations to have the strong diagonal dominance. The full boundary control method is adopted to determine the control functions of the equations, which allows the control of the grid regarding spacing and angle control at all boundary surfaces. The solution method presented here provides the capability of mapping very complicated geometries by defining grid point locations only along the boundaries. In the automated elliptic grid generation procedure, it is showed that strong diagonal dominance is essential to achieve successful mapping irrespective of the initial grid condition provided. To demonstrate the robustness of this method, it is applied to the thermal flow like the natural convection between eccentric cylinders. The results agree well with others.

• 대한수학회지
• /
• 제45권4호
• /
• pp.1075-1087
• /
• 2008

We study a class of KMS-symmetric quantum Markovian semigroups on a quantum spin system ($\mathcal{A},{\tau},{\omega}$), where $\mathcal{A}$ is a quasi-local algebra, $\tau$ is a strongly continuous one parameter group of *-automorphisms of $\mathcal{A}$ and $\omega$ is a Gibbs state on $\mathcal{A}$. The semigroups can be considered as the extension of semi groups on the nontrivial abelian subalgebra. Let $\mathcal{H}$ be a Hilbert space corresponding to the GNS representation con structed from $\omega$. Using the general construction method of Dirichlet form developed in [8], we construct the symmetric Markovian semigroup $\{T_t\}{_t_\geq_0}$ on $\mathcal{H}$. The semigroup $\{T_t\}{_t_\geq_0}$ acts separately on two subspaces $\mathcal{H}_d$ and $\mathcal{H}_{od}$ of $\mathcal{H}$, where $\mathcal{H}_d$ is the diagonal subspace and $\mathcal{H}_{od}$ is the off-diagonal subspace, $\mathcal{H}=\mathcal{H}_d\;{\bigoplus}\;\mathcal{H}_{od}$. The restriction of the semigroup $\{T_t\}{_t_\geq_0}$ on $\mathcal{H}_d$ is Glauber dynamics, and for any ${\eta}{\in}\mathcal{H}_{od}$, $T_t{\eta}$, decays to zero exponentially fast as t approaches to the infinity.

• Steel and Composite Structures
• /
• 제26권3호
• /
• pp.361-373
• /
• 2018

This paper presents the development of finite element (FE) models to simulate the behavior of diagonally stiffened steel plate shear wall systems (SPSWs) with differently shaped openings subjected to a cyclic load. This walling system has the potential to be used for shear elements that resist lateral loads in steel-framed buildings. A number of $\text\tiny{^1/_2}$-scale one-story buildings that were un-stiffened, stiffened and stiffened with opening SPSWs are modeled and simulated using the finite element method based on experimental data from previous research. After validating the finite element (FE) models, the effects of infill plate thickness on the cyclic behavior of steel shear walls are investigated. Furthermore, triple diagonal stiffeners are added to the steel infill plates of the SPSWs, and the effects are studied. Moreover, the effects of a number of differently shaped openings applied to the infill plate are studied. The results indicate that the bearing capacity and shear resistance are affected positively by increasing the infill plate thickness and by adding triple diagonal stiffeners. In addition, the cyclic behavior of SPSWs is improved, even with an opening in the SPSWs.

• Steel and Composite Structures
• /
• 제19권2호
• /
• pp.349-368
• /
• 2015

This paper presents a new algorithm to find the optimal distribution of steel diagonal braces (SDB) using artificial bee colony optimization technique. The four different objective functions are employed based on the transfer function amplitude of; the top displacement, the top absolute acceleration, the base shear and the base moment. The stiffness parameter of SDB at each floor level is taken into account as design variables and the sum of the stiffness parameter of the SDB is accepted as an active constraint. An optimization algorithm based on the Artificial Bee Colony (ABC) algorithm is proposed to minimize the objective functions. The proposed ABC algorithm is applied to determine the optimal SDB distribution for planar buildings in order to rehabilitate existing planar steel buildings or to design new steel buildings. Three planar building models are chosen as numerical examples to demonstrate the validity of the proposed method. The optimal SDB designs are compared with a uniform SDB design that uniformly distributes the total stiffness across the structure. The results of the analysis clearly show that each optimal SDB placement, which is determined based on different performance objectives, performs well for its own design aim.

• 대한건축학회논문집:구조계
• /
• 제35권9호
• /
• pp.191-198
• /
• 2019

All over the Republic of Korea, there are many masonry buildings which have been built since 1970s. When the earthquake at Po-Hang occurred, this is the main cause of huge damage because the masonry buildings have not seismic capacity. When masonry buildings are failed, two type of the failure modes can be shown, which are in-plane mode and out-plane mode. In-plane mode can have seismic capacity in masonry so diagonal shear test is performed in this study. The purpose of this study was to find the best way to reinforce the materials through the diagonal shear test. Through the test, shear stress and shear modulus of elasticity will be calculated, referred to the ASTM E 519-02. The variables in this test are ${\phi}3$ wire, three types of wire meshes, polypropylene strap and different types of brick. Each variable is applied to the same condition of the $1.2m{\times}1.2m$ masonry walls which are made by ASTM E 519-02. Compared to each variable with shear stress and shear modulus of elasticity, the best way of reinforcing method to have seismic capacity will be proved in this study.

• 소성∙가공
• /
• 제27권4호
• /
• pp.250-256
• /
• 2018

Oxygen-free copper (OFC) was prepared as a 90 mm cube and then processed with Multi-Axial Diagonal Forging - Initialization of Prior manufacturing History (MADF). The MADF process has been newly developed as a severe plastic deformation method. The MADF process consists of upset forging with a thickness reduction of 30% and diagonal forging with a diagonal angle of $135^{\circ}$. 1 cycle process consists of a 12 passes forging process. In order to analyze the characteristic changes according to the number of iterations, 1, 2, and 3 cycles of the MADF process were performed. The OFC specimens were MADF processed without surface cracks up to 3 cycles. The microstructure, hardness and tensile test of processed materials were analyzed to study the change of material properties according to the amount of MADF process. The results showed that the MADF process effectively refined the microstructure and increased the strength of OFC. In the case of specimens processed for more than 2 cycles, the grains of all measurement regions were refined to be less than $7{\mu}m$ of grain size. The 1 cycle MADF processed OFC showed the highest mechanical properties with the hardness of 132 HV and tensile strength of 395 MPa. Hardness and strength seemed to be saturated when processed over 2 cycles.

• 디지털융복합연구
• /
• 제12권12호
• /
• pp.441-447
• /
• 2014

본 연구는 대기업에 근무하는 산업체 근로자를 대상으로 비만도에 따른 척추안정화근력의 차이를 분석하는데 목적을 두었다. 편의표본 추출방법(Convenience Sampling Method)에 의하여 50명을 표본을 추출하여, 체질량지수(BMI) $25kg/m^2$ 이상을 비만군(25명), 이하를 정상군(25명)으로 분류하여 독립표본 t검정(Independent t-test)으로 분석하였다. 집단간 척추안정화근력의 측정결과 정상군의 평균이 비만군에 비해 모두 높았으며, Sagittal Plane에서는 $0^{\circ}$ Forward에서 유의한 차이를 보였고, Coronal plane에서는 $90^{\circ}$ Left, $90^{\circ}$ Right에서 유의한 차이를 보였으며, Diagonal Plane에서는 $45^{\circ}$ Left Tilt, $45^{\circ}$ Right Tilt에 유의한 차이를 보였다. BMI 분류에 따른 근육량과 골격근량의 분석결과에서는 비만군이 정상군에 비해 모두 높았으며 통계적으로 유의한 차이를 보였다. 비만군이 정상군에 비해 근육량은 많지만 척추안정화근력은 약하다는 것을 감안하여 Sagittal Plane, Coronal plane, Diagonal Plane에서의 비만인을 위한 척추안정화 강화 프로그램의 개발이 요구된다.