• The KIPS Transactions:PartB
• /
• v.8B no.3
• /
• pp.231-236
• /
• 2001

일반적인 다층 신경망에서 가중치의 갱신 알고리즘으로 사용하는 오류 역전과 방식은 가중치 갱신 결과를 고정된(fixed) 한 개의 값으로 결정한다. 이는 여러 갱신의 가능성을 오직 한 개의 값으로 고정하기 때문에 다양한 가능성들을 모두 수용하지 못하는 면이 있다. 하지만 모든 가능성을 확률적 분포로 표현하는 갱신 알고리즘을 도입하면 이런 문제는 해결된다. 이러한 알고리즘을 사용한 베이지안 신경망 모형(Bayesian Neural Networks Models)은 주어진 입력값(Input)에 대해 블랙 박스(Black-Box)와같은 신경망 구조의 각 층(Layer)을 거친 출력값(Out put)을 계산한다. 이 때 주어진 입력 데이터에 대한 결과의 예측값은 사후분포(posterior distribution)의 기댓값(mean)에 의해 계산할 수 있다. 주어진 사전분포(prior distribution)와 학습데이터에 의한 우도함수(likelihood functions)에 의해 계산한 사후확률의 함수는 매우 복잡한 구조를 가짐으로 기댓값의 적분계산에 대한 어려움이 발생한다. 따라서 수치해석적인 방법보다는 확률적 추정에 의한 근사 방법인 몬테 칼로 시뮬레이션을 이용할 수 있다. 이러한 방법으로서 Hybrid Monte Carlo 알고리즘은 좋은 결과를 제공하여준다(Neal 1996). 본 논문에서는 Hybrid Monte Carlo 알고리즘을 적용한 신경망이 기존의 CHAID, CART 그리고 QUEST와 같은 여러 가지 분류 알고리즘에 비해서 우수한 결과를 제공하는 것을 나타내고 있다.

• Proceedings of the Korea Air Pollution Research Association Conference
• /
• 1999.04a
• /
• pp.293-294
• /
• 1999

• Journal of Radiation Protection and Research
• /
• v.39 no.1
• /
• pp.30-37
• /
• 2014

Recently High-Definition Reference Korean-Man (HDRK-Man) and High-Definition Reference Korean-Woman (HDRK-Woman) were constructed in Korea. The HDRK phantoms were designed to represent respectively reference Korean male and female to calculate effective doses for Korean by performing Monte Carlo dose calculation. However, the Monte Carlo dose calculation requires detailed knowledge on computational human phantoms and Monte Carlo simulation technique which regular researchers in radiation protection dosimetry and practicing health physicists do not have. Recently the UFPE (Federal University of Pernambuco) research group has developed, and opened to public, an online Monte Carlo dose calculation system called CALDOSE_X(www.caldose.org). By using the CALDOSE_X, one can easily perform Monte Carlo dose calculations. However, the CALDOSE_X used caucasian phantoms to calculate organ doses or effective doses which are limited for Korean. The present study developed an online reference Korean dose calculation system which can be used to calculate effective doses for Korean.

• Progress in Medical Physics
• /
• v.10 no.1
• /
• pp.55-62
• /
• 1999

In an effort to study the characteristics of x-rays utilized in radiation therapy, we calculated the energy distribution and the mean energy of x-rays generated from a tungsten target bombarded by 6, 10, and 15 MeV electron beams, using a Monte Carlo technique. The average photon energies calculated as a function of the beam radius lied in 1.4 ∼ 1.6, 2.1 ∼ 2.5 and 2.8 ∼ 3.3 MeV ranges for 4, 10, and 15 MV electron beams, respectively, which turned out to have no strong dependence on the radius. Using the energy distributions of 6,10, and 15 MV x-rays obtained for the target distance of 100 cm, percentage depth doses were determined using Monte Carlo calculations. For the case 10 MV, a comparison was made between our calculation and measurement performed by others. The calculated percentage depth dose appeared somewhat smaller than the measured one except in the surface region. We conclude that this is due to the fact that the beam hardening effect resulting from the flattening filter was not properly allowed for in our Monte Carlo calculations.

• Proceedings of the Korean Information Science Society Conference
• /
• 2000.04b
• /
• pp.235-237
• /
• 2000

본 논문에서는 환경의 변화에 민감한 시계열 데이터인 주가의 변동과정을 보다 효과적으로 설명하기 위한 방안의 하나로 강화 학습 모형의 도입을 제안하며, 특정 시점의 주가 추세를 평가하는 기준으로 가치도 함수를 채택한다. 가치도 함수의 계산에는 강화 학습 알고리즘의 일종인 Monte Carlo 알고리즘을 적용하고, 신경망 구현에 의해 구한 근사 가치도의 적절성을 평가하였다. 실험 결과로 볼 때, 여타 강화 학습 알고리즘을 추가적으로 적용할 경우, 주가 변동의 시계열적 특성을 더욱 잘 반영할 수 있을 것으로 판단된다.

• Korean Journal of Remote Sensing
• /
• v.16 no.1
• /
• pp.65-72
• /
• 2000

Scattering coefficients of randomly rough lossy dielectric surfaces were computed by using the FDTD(Finite-Difference Time-Domain) method and the Monte Carlo method in this paper. The FDTD method was applied to compute electromagnetic wave scattering characteristics at any incident angles, any linear polarizations by dividing the computation region into the total-field region and the scattered-field region. The radar cross sections(RCS) of conducting cylinders have been computed and compared with theoretical results, measurement data and the results from the method of moment(MoM) to verify the FDTD algorithm. Then, to apply the algorithm to compute scattering coefficients of distributed targets, a two-dimensionally rough surface was generated numerically for given roughness characteristics. The far-zone scattered fields of 50 statistically independent dielectric rough surfaces were computed and the scattering coefficient of the surface was calculated from the scattered fields by using the Monte Carlo method. It was found that these scattering coefficients agree well with the SPM(Small Pertubation Method) model in its validity region.

• Water for future
• /
• v.26 no.2
• /
• pp.89-97
• /
• 1993

For the system benefit optimization by considering risk or reliability from a multiple reservoir system using the Monte Carlo technique, many stochastically generated inflow series have to be used for the system analysis. In this study, the stochastically generated inflow series for the multiple reservoir system operation are preprocessed according to the considered system objectives and operating time periods. Through this procedure, several representative inflow series which have discrete probability levels and operation horizons are selected among the thousands of generated inflows. Then a deterministic optimization technique is applied to the power energy estimation from the Han River Reservoirs System which considers five reservoirs in the study. It took much lower computational requirements then using the original Monte Carlo Technique, even though estimated result was almost similar.

• KSCE Journal of Civil and Environmental Engineering Research
• /
• v.26 no.5B
• /
• pp.501-509
• /
• 2006

The safety of levee that depends on the river flood elevation has been regarded as very important keys to build up various flood prevention systems. However, deterministic methods for computation of water surface profile cannot reflect the effect of possible inaccuracies in the input parameters. The purpose of this study is to develop a methodology of uncertainty computation of design flood level based on steady flow analysis and Monte Carlo simulation. This study addresses the uncertainty of water surface elevation by Manning's coefficients, design discharges, river cross sections and boundary condition. Monte Carlo simulation with the variations of these parameters is performed to quantify the variations of water surface elevations in a river. The proposed model has been applied to the Kumho-river. The reliability analysis was performed within 38.5 km (95 sections) reach considered the variations of the above-mentioned parameters. Overtopping risks were evaluated by comparing the elevations of the flood condition with the those of the levees. The results show that there is a necessity which will raise the levee elevation between 1 cm and 56 cm at 7 sections. The model can be used for preparing flood risk maps, flood forecasting systems and establishing flood disaster mitigation plans as well as complement of conventional levee design.

• Journal of Radiation Protection and Research
• /
• v.26 no.2
• /
• pp.93-99
• /
• 2001

Voxel head phantom for overcoming the limitation of mathematical phantom in depleting anatomical details was constructed and example dose calculation for BNCT was performed. The repeated structure algorithm of the general purpose Monte Carlo code, MCNP4B was applied for yokel Monte Carlo calculation. Simple binary yokel phantom and combinatorial geometry phantom composed of two materials were constructed for validating the voxel Monte Carlo calculation system. The tomographic images of VHP man provided by NLM(National Library of Medicine) were segmented and indexed to construct yokel head phantom. Comparison of doses for broad parallel gamma and neutron beams in AP and PA directions showed decrease of brain dose due to the attenuation of neutron in eye balls in case of yokel head phantom. The spherical tumor volume with diameter, 5cm was defined in the center of brain for BNCT dose calculation in which accurate 3 dimensional dose calculation is essential. As a result of BNCT dose calculation for downward neutron beam of 10keV and 40keV, the tumor dose is about doubled when boron concentration ratio between the tumor to the normal tissue is $30{\mu}g/g$ to $3{\mu}g/g$. This study established the voxel Monte Carlo calculation system and suggested the feasibility of precise dose calculation in therapeutic radiology.

• Progress in Medical Physics
• /
• v.19 no.1
• /
• pp.21-34
• /
• 2008

The parallel Monte Carlo electron and photon transport (PMCEPT) code [Kum and Lee, J. Korean Phys. Soc. 47, 716 (2006)] for calculating electron and photon beam doses has been developed based on the three dimensional geometry defined by computed tomography (CT) images and implemented on the Beowulf PC cluster. Understanding the limitations of Monte Carlo codes is useful in order to avoid systematic errors in simulations and to suggest further improvement of the codes. We evaluated the PMCEPT code by comparing its normalized depth doses for electron and photon beams with those of MCNP5, EGS4, DPM, and GEANT4 codes, and with measurements. The PMCEPT results agreed well with others in homogeneous and heterogeneous media within an error of $1{\sim}3%$ of the dose maximum. The computing time benchmark has also been performed for two cases, showing that the PMCEPT code was approximately twenty times faster than the MCNP5 for 20-MeV electron beams irradiated on the water phantom. For the 18-MV photon beams irradiated on the water phantom, the PMCEPT was three times faster than the GEANT4. Thus, the results suggest that the PMCEPT code is indeed appropriate for both fast and accurate simulations.