• 정보처리학회논문지B
• /
• 제14B권7호
• /
• pp.513-522
• /
• 2007

군집화는 주어진 데이터를 분할하여 데이터 속에 숨겨져 있는 의미를 자동으로 발견하는 방법으로, 사람이 일일이 살펴보기 어려운 데이터를 분석해서 비슷한 성향을 가진 데이터들끼리 모은 여러 개의 군집들을 만들어 낸다. 온라인 문서 군집화는 검색 엔진을 통해 검색된 문서들을 대상으로 군집화를 실행하여 유사한 특성의 문서들을 묶어서 보여줌으로써 사용자의 검색 환경의 편의성을 증진시키는 것이 목적이다. 문서군집화는 사람의 개입이 없이 자동으로 이루어져야 하고, 군집화 결과에 영향을 미치는 군집의 개수 선정도 자동으로 이루어져야 한다. 또한, 온라인 시스템에서는 빠른 응답 시간을 보장하는 것이 중요하다. 본 논문에서는 기하학적인 정보를 이용하여 군집의 수를 결정하는 방법을 제안한다. 제안하는 방법은 군집의 중심을 저차원 평면에 사상하는 것과 사상된 군집 중심의 거리 정보를 이용하여 군집들을 병합하는 두 단계로 이루어져 있다. 제안하는 방법을 실데이터에 적용하여 실험한 결과 군집화 성능이 향상되고, 처리 시간도 온라인 환경에 적합한 것을 확인 할 수 있었다.

• 한국음향학회지
• /
• 제21권3호
• /
• pp.304-309
• /
• 2002

LSP (Line Spectrum Pairs) 파라미터는 음성코덱 (codec)이나 인식기에서 음성신호를 분석하여 전송형이나 저장형 파라미터로 변환되어, 주로 저전송률 음성부호화기에 사용된다. 그러나 LPC (Linear Predictive Coding) 계수를 LSP로 변환하는 방법이 복잡하여 계산시간이 많이 소요된다는 단점이 있다. 기존의 LSP변환 방법 중 음성 부호화기에서 주로 사용하는 실근 (real root)방법은 근을 구하기 위해 주파수 영역을 순차적으로 검색하기 때문에 계산시간이 많이 소요되는 단점을 갖는다. 본 논문에서 기존의 실근 방법과 비교 평가한 알고리즘은 첫 번째 검색 대역에 멜 스케일 (met scale)을 사용하였고, 두 번째는 LSP 파라미터의 분포 특성을 조사하여 이를 토대로 검색구간의 순서와 검색간격을 달리 하였다. 실험결과, 기존의 실근 방식에 비하여 두 가지 방식 모두가 변환시간의 47% 이상이 감소되는데 반하여 동일한 근을 찾음을 알 수가 있었다.

• 화약ㆍ발파
• /
• 제9권1호
• /
• pp.3-13
• /
• 1991

발파에 의한 지반진동의 크기는 화약류의 종류에 따른 화약의 특성, 장약량, 기폭방법, 전새의 상태와 화약의 장전밀도, 자유면의 수, 폭원과 측간의 거리 및 지질조건 등에 따라 다르지만 지질 및 발파조건이 동일한 경우 특히 측점으로부터 발파지점 까지의 거리와 지발당 최대장약량 (W)간에 깊은 함수관계가 있음이 밝혀졌다. 즉 발파진동식은 $V=K{\cdot}(\frac{D}{W^b})^n{\;}{\cdots\cdots\cdots\cdots\cdots\cdots\cdots\cdots\cdots\cdots\cdots}$ (1) 여기서 V ; 진동속도, cm /sec D ; 폭원으로부터의 거리, m W ; 지발 장약량, kg K ; 발파진동 상수 b ; 장약지수 R ; 감쇠지수 이 발파진동식에서 b=1/2인 경우 즉 $D{\;}/{\;}\sqrt{W}$를 자승근 환산거리(Root scaled distance), $b=\frac{1}{3}$인 경우 즉 $D{\;}/{\;}\sqrt[3]{W}$를 입방근환산거리(Cube root scaled distance)라 한다. 이 장약 및 감쇠지수와 발파진동 상수를 구하기 위하여 임의거리와 장약량에 대한 진동치를 측정, 중회귀분석(Multiple regressional analysis)에 의해 일반식을 유도하고 Root scaling과 Cube root scaling에 대한 회귀선(regression line)을 구하여 회귀선에 대한 적합도가 높은 쪽을 택하여 비교, 검토하였다. 위 (1)식의 양변에 log를 취하여 linear form(직선형)으로 바꾸어 쓰면 (2)式과 같다. log V=A+BlogD+ClogW ----- (2) 여기서, A=log K B=-n C=bn (2)식은 다시 (3)식으로 표시할 수 있다. $Yi=A+BXi_{1}+CXi_{2}+{\varepsilon}i{\;}{\cdots\cdots\cdots\cdots\cdots\cdots\cdots\cdots\cdots\cdots\cdots}$(3) 여기서, $Xi_{1},{\;}Xi_{2} ;(두 독립변수 logD, logW의 i번째 측정치. Yi ; ($Xi_1,{\;}Xi_2$)에 대한 logV의 측정치 ${\varepsilon}i$ ; error term 이다. (3)식에서 n개의 자료를 (2)식의 회귀평면으로 대표시키기 위해서는 $S={\sum}^n_{i=1}\{Yi-(A+BXi_{1}+CXi_{2})\}\^2$을 최소로하는 A, B, C 값을 구하면 된다. 이 방법을 최소자승법이 라 하며 S를 최소로 하는 A, B, C의 값은 (4)식으로 표시한다. $\frac{{\partial}S}{{\partial}A}=0,{\;}\frac{{\partial}S}{{\partial}B}=0,{\;}\frac{{\partial}S}{{\partial}C}=0{\;}{\cdots\cdots\cdots\cdots\cdots\cdots\cdots\cdots\cdots\cdots\cdots}$ (4) 위식을 Matrix form으로 간단히 나타내면 식(5)와 같다. [equation omitted] (5) 자료가 많아 계산과정이 복잡해져서 본실험의 정자료들은 전산기를 사용하여 처리하였다. root scaling과 Cube root scaling의 경우 각각 $logV=A+B(logD-\frac{1}{2}W){\;}logV=A+B(logD-\frac{1}{3}W){\;}\}{\;}{\cdots\cdots\cdots\cdots\cdots\cdots\cdots\cdots\cdots}$ (6) 으로 (2)식의 특별한 형태이며 log-log 좌표에서 직선으로 표시되고 이때 A는 절편, B는 기울기를 나타낸다. $\bullet$ 측정치의 검토 본 자료의 특성을 비교, 검토하기 위하여 지금까지 발표된 국내의 몇몇 자료를 보면 다음과 같다. 물론, 장약량, 폭원으로 부터의 거리등이 상이하지만 대체적인 경향성을 추정하는데 참고할수 있을 것이다. 금반 총실측자료는 총 88개이지만 환산거리(5.D)와 진동속도의 크기와의 관계에서 차이를 보이고 있어 편선상 폭원과 측점지점간의 거리에 따라 l00m말만인 A지역과 l00m이상인B지역으로 구분하였다. 한편 A지역의 자료 56개중, 상하로 편차가 큰 19개를 제외한 37개자료와 B지역의 29개중 2개를 낙외한 27개(88개 자료중 거리표시가 안된 12월 1일의 자료3개는 원래부터 제외)의 자료를 computer로 처리하여 얻은 발파진동식은 다음과 같다. $V=41(D{\;}/{\;}\sqrt[3]{W})^{-1.41}{\;}{\cdots\cdots\cdots\cdots\cdots\cdots\cdots\cdots\cdots}$ (7) (-100m)(R=0.69) $V=124(D{\;}/{\;}\sqrt[3]{W})^{-1.66){\;}{\cdots\cdots\cdots\cdots\cdots\cdots\cdots\cdots\cdots}$ (8) (+100m)(R=0.782) 식(7) 및 (8)에서 R은 구한 직선식의 적합도를 나타내는 상관계수로 R=1인때는 모든 측정자료가 하나의 직선상에 표시됨을 의미하며 그 값이 낮을수록 자료가 분산됨을 뜻한다. 본 보고에서는 상관계수가 자승근거리때 보다는 입방근일때가 더 높기 때문에 발파진동식을 입방근($D{\;}/{\;}\sqrt[3]{W}$)으로 표시하였다. 특히 A지역에서는 R=0.69인데 비하여 폭원과 측점지점간의 거리가 l00m 이상으로 A지역보다 멀리 떨어진 B지역에서는 R=0.782로 비교적 높은 값을 보이는 것은 진동성분중 고주파성분의 상당량이 감쇠를 당하기 때문으로 생각된다.

• 천문학회보
• /
• 제40권2호
• /
• pp.32.3-33
• /
• 2015

There exist scaling relations that link the mass of supermassive black holes with both the velocity dispersion and the mass of the central stellar cusp of their host galaxies. This implies that these two components grow in tandem. Feedback from actively accreting supermassive black holes (AGN), in the form of multi-phase gas outflows, has been argued to be the agent of this co-evolution. Here we employ the powerful GMOS integral field spectroscopy unit on the 8.2m Gemini-North telescope to investigate ionized gas outflows of luminous Type 2 AGN in the local Universe (z<0.1). Our sample of 6 galaxies is drawn from the Sloan Digital Sky Survey (SDSS) and was selected based on their [OIII] dust-corrected luminosity (>1042 erg/s) and signatures of outflows in the [OIII] line profile of their spatially integrated SDSS spectra. These are arguably the best candidates to explore AGN feedback in action since they are < 1% of a large local type 2 AGN SDSS sample selected based on their [OIII] kinematics. We combine a careful spectral decomposition of the [OIII] and $H{\alpha}$ line profiles with spatial information on ~0.5kpc scales to understand the outflow kinematics and energetics in these objects. We find clear evidence for strong outflows in [OIII] and occasionally $H{\alpha}$ that are clearly driven by the ionizing radiation of the AGN. We kinematically and spatially decompose outflowing and rotating ionized gas components. We find [OIII] to be a better tracer of AGN outflows, while $H{\alpha}$ appears to be strongly affected by both stellar rotation and outflows induced by ongoing star formation. The observed kinematics and spatial distribution of the ionized gas imply a large opening angle for the outflow. Finally, we find the projected outflow velocity to decrease as a function of distance, while its dispersion shows a more complex structure with a potentially initially increasing trend (out to 0.5-1kpc distances).

• 천문학회보
• /
• 제41권2호
• /
• pp.41.2-41.2
• /
• 2016

We perform three statistical studies on the physical properties and oscillations in the confined plasma such as a photospheric sunspot and confined coronal loop. From the statistical studies on the sunspot umbra and its oscillation, we find that (1) the total magnetic flux inside the umbra for the three groups increases proportionally with the powers of the umbral area and the power indices in the three groups significantly differ from each other; (2) the three groups have different characteristics in their umbral area, intensity, magnetic field strength, and Doppler velocity as well as their relationships; (3) the mean frequency of the umbral oscillations increases with the umbral mean magnetic field strength and height; (4) the time delay of the core intensity of Fe I absorption line relative to the continuum which are de-convolved with the frequency range higher than 3.5 mHz is mostly positive, implying that the photospheric umbral oscillations are likely upwardly propagating; (5) the umbral mean plasma beta ranges approximately 0.6-1.1 and does not vary significantly from pores to mature sunspots. From the comparative study on the quasi-periodic pulsations (QPPs) in the solar and stellar flares, (6) we find that the power index of the periods scaling the damping times observed in the stellar QPPs is consistent with that observed in the solar QPPs, suggesting that physical mechanisms responsible for the stellar QPPs are likely the magneto-hydrodynamic oscillation of solar coronal loops.

• 소음진동
• /
• 제6권2호
• /
• pp.163-177
• /
• 1996

An investigation on thermohydraulic stability of flow oscillations in the CANada Deuterium Uranium-600(CANDU-6) heat transport system has been conducted. Flow oscillations in reactor coolant loops, comprising two heat sources and two heat sinks in series, are possibly caused by the response of the pressure to extraction of fluid in two-phase region. This response consists of two contributions, one arising from mass and another from enthalpy change in the two-phase region. The system computer code used in the investigation os SOPHT, which is capable of simulating steady states as well as transients with varying boundary conditions. The model was derived by linearizing and solving one-dimensional, homogeneous single- and two-phase flow conservation equations. The mass, energy and momentum equations with boundary conditions are set up throughout the system in matrix form based on a node-link structure. Loop stability was studied under full power conditions with interconnecting the two compressible two phase regions in the figure-of-eight circuit. The dominant function of the interconnecting pipe is the transfer of mass between the two-phase regions. Parametric survey of loop stability characteristics, i. e., damping ratio and period, has been made as a function of geometrical parameters of the interconnection line such as diameter, length, height and orifice flow coefficient. The stability characteristics with interconnection line has been clarified to provide a simple criterion to be used as a guide in scaling of the pipe.

• International Journal of Fuzzy Logic and Intelligent Systems
• /
• 제5권1호
• /
• pp.13-20
• /
• 2005

This paper addresses a new blind channel equalization method using fourth-order cumulants of channel inputs and a three-layer neural network equalizer. The proposed algorithm is robust with respect to the existence of heavy Gaussian noise in a channel and does not require the minimum-phase characteristic of the channel. The transmitted signals at the receiver are over-sampled to ensure the channel described by a full-column rank matrix. It changes a single-input/single-output (SISO) finite-impulse response (FIR) channel to a single-input/multi-output (SIMO) channel. Based on the properties of the fourth-order cumulants of the over-sampled channel inputs, the iterative algorithm is derived to estimate the deconvolution matrix which makes the overall transfer matrix transparent, i.e., it can be reduced to the identity matrix by simple recordering and scaling. By using this estimated deconvolution matrix, which is the inverse of the over-sampled unknown channel, a three-layer neural network equalizer is implemented at the receiver. In simulation studies, the stochastic version of the proposed algorithm is tested with three-ray multi-path channels for on-line operation, and its performance is compared with a method based on conventional second-order statistics. Relatively good results, withe fast convergence speed, are achieved, even when the transmitted symbols are significantly corrupted with Gaussian noise.

• Structural Engineering and Mechanics
• /
• 제17권2호
• /
• pp.215-244
• /
• 2004

There is an ever-increasing demand for assessment of earthquake effects on transportation structures, emphasised by the crippling consequences of recent earthquakes hitting developed countries reliant on road transportation. In this work, vulnerability functions for RC bridges are derived analytically using advanced material characterisation, high quality earthquake records and adaptive inelastic dynamic analysis techniques. Four limit states are employed, all based on deformational quantities, in line with recent development of deformation-based seismic assessment. The analytically-derived vulnerability functions are then compared to a data set comprising observational damage data from the Northridge (California 1994) and Hyogo-ken Nanbu (Kobe 1995) earthquakes. The good agreement gives some confidence in the derived formulation that is recommended for use in seismic risk assessment. Furthermore, by varying the dimensions of the prototype bridge used in the study, and the span lengths supported by piers, three more bridges are obtained with different overstrength ratios (ratio of design-to-available base shear). The process of derivation of vulnerability functions is repeated and the ensuing relationships compared. The results point towards the feasibility of deriving scaling factors that may be used to obtain the set of vulnerability functions for a bridge with the knowledge of a 'generic' function and the overstrength ratio. It is demonstrated that this simple procedure gives satisfactory results for the case considered and may be used in the future to facilitate the process of deriving analytical vulnerability functions for classes of bridges once a generic relationship is established.

• Journal of Information Processing Systems
• /
• 제18권4호
• /
• pp.575-586
• /
• 2022

Uneven light in real-world causes visual degradation for underexposed regions. For these regions, insufficient consideration during enhancement procedure will result in over-/under-exposure, loss of details and color distortion. Confronting such challenges, an unsupervised low-light image enhancement network is proposed in this paper based on the guidance of the unpaired low-/normal-light images. The key components in our network include super-resolution module (SRM), a GAN-based low-light image enhancement network (LLIEN), and denoising-scaling module (DSM). The SRM improves the resolution of the low-light input images before illumination enhancement. Such design philosophy improves the effectiveness of texture details preservation by operating in high-resolution space. Subsequently, local lightness attention module in LLIEN effectively distinguishes unevenly illuminated areas and puts emphasis on low-light areas, ensuring the spatial consistency of illumination for locally underexposed images. Then, multiple discriminators, i.e., global discriminator, local region discriminator, and color discriminator performs assessment from different perspectives to avoid over-/under-exposure and color distortion, which guides the network to generate images that in line with human aesthetic perception. Finally, the DSM performs noise removal and obtains high-quality enhanced images. Both qualitative and quantitative experiments demonstrate that our approach achieves favorable results, which indicates its superior capacity on illumination and texture details restoration.

• Biotechnology and Bioprocess Engineering:BBE
• /
• 제5권2호
• /
• pp.118-122
• /
• 2000

A comlementary call line CR2 is curretly used to propagte the Disabled Infectious Single Cycle Herpes Simplex Virus Typee2 (DISC HSV-2) on a small Iaboratory scale upto 15 L. These cultures are initiated by passaging the cells from roller bottle cultures. Whilst this is suitable for the laboratory scale it is totally impractical for use in seeding an industrial manufacturing scaled version of the culture. It is paramount to have a robust system for passaging cells from a small microcarrierier culture system to a larger one by a serial subculturing regime. Here we report on the successes we have had in our laboratory in scaling up out production system for the DISC HSV-2 from small 1-L cultures to a 50-L vessel with the maintenance of the viral productivity. Ease of use, reproducibility and the need to minimise overall production time were factors which were taken into consideration whils developing our procedures. We were aware of the need to keep a production train simple and as short as possible as this was the amall scale study for an envisaged manufacturing process.