• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.24 no.2
• /
• pp.93-98
• /
• 2024

With the rapid advance in AI (artificial intelligence) and high-performance computing technologies, deep learning is being used in various fields. Deep learning proceeds training by randomly reading a large amount of data and repeats this process. A large number of files are randomly repeatedly referenced during deep learning, which shows different access characteristics from traditional workloads with temporal locality. In order to cope with the difficulty in caching caused by deep learning, we propose a new sampling method that aims at reducing the randomness of dataset reading and adaptively operating on existing buffer cache algorithms. We show that the proposed policy reduces the miss rate of the buffer cache by 16% on average and up to 33% compared to the existing method, and improves the execution time by up to 24%.

• Journal of Biomedical Engineering Research
• /
• v.21 no.5
• /
• pp.441-448
• /
• 2000

방출단층영상 재구성을 위한 최대우도 기대값최대화(maximum likelihood expectation maximization, MLEM) 방법은 영상 획득과정을 통계학적으로 모델링하여 영상을 재구성한다. MLEM은 일반적으로 사용하여 여과후역투사(filtered backprojection)방법에 비해 많은 장점을 가지고 있으나 반복횟수 증가에 따른 발산과 재구성 시간이 오래 걸리는 단점을 가지고 있다. 이 논문에서는 이러한 단점을 보완하기 위해 계산시간을 현저히 단축시킨 배열된부분집합 기대값최대화(ordered subsets expectation maximization. OSEM)에 Gibbs 선행치인 membrance (MM) 또는 thin plate(TP)을 첨가한 OSEM-MAP (maximum a posteriori)을 구현함으로써 알고리즘의 안정성 및 재구성된 영상의 질을 향상시키고자 g나다. 실험에서 알고리즘의 수렴시간을 가속화하기 위해 투사 데이터를 16개의 부분집합으로 분할하여 반복연산을 수행하였으며, 알고리즘의 성능을 비교하기 위해 소프트웨어 모형(원숭이 뇌 자가방사선, 수학적심장흉부)을 사용한 영상재구성 결과를 제곱오차로 비교하였다. 또한 알고리즘의 사용 가능성을 평가하기 위해 물리모형을 사용하여 PET 기기로부터 획득한 실제 투사 데이터를 사용하였다.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2019.05a
• /
• pp.281-281
• /
• 2019

수리모형실험은 수로 내에서 장시간 파랑을 발생시킬 경우, 수로 내에 반사 파랑의 성분이 누적될 수 있어 상당한 계측 오차를 발생시킬 우려가 있어 수리모형실험 결과의 검증이 필요하다. 일반적으로 수리모형실험 결과의 검증을 위해서는 동일 실험을 무수히 반복하여 불확실성을 제거하거나 다양한 수리실험실에서 수리모형실험을 수행하고 결과를 분석하여 불확실성을 제거할 수 있다. 그러나 이는 엄청난 시간과 노력은 물론 막대한 실험비용이 소요되기 때문에 경제적으로 효용성이 매우 낮아 현실적으로 수행이 어렵다. 이에 비해 수치모형실험은 상대적으로 저렴한 비용으로 수행할 수 있으며, 다수의 실험을 수행하지 않아도 불확실성을 제거할 수 있어 수리모형실험의 검증에 효율적이다. 일반적으로 난류 거동을 동반하는 복잡한 구조물 주변의 흐름 해석에는 3차원 CFD 모형이 필요하다. 특히, 병렬연산이 가능한 CFD 모형을 활용하면 수리모형실험에서도 재현이 쉽지 않은 다양한 조건에 따른 복잡한 흐름을 해석할 수 있어 효용성이 점점 증가하고 있다. 그러나 복잡한 구조물이 존재하게 되면 구조물에 재현에 막대한 격자구조가 필요하여 현실적으로 적용이 쉽지 않다. 이에 대한 대안으로 복잡한 구조물을 비교적 큰 격자에서 재현할 수 있는 가상경계법을 활용하는 연구가 활발히 진행되고 있다. 가상경계법은 Navier-Stokes 방정식에서 유체 내에 존재하는 고체를 모멘텀 이론으로 대체하여 고려하는 기법으로 수치모델링 수행 시 매질을 유체만으로 구성할 수 있어 안정적으로 적용할 수 있는 것으로 알려져 있다. 본 과업에서는 다양한 분야에서 널리 활용되고 있는 3차원 CFD 모형인 OpenFOAM®기반으로 파랑해석에 필요한 경계조건을 계산할 수 있는 olaFlow를 활용하여 복잡한 구조물을 지나는 파랑해석을 수행하기 위해 가상경계법을 olaFlow에 도입한 수치 알고리즘을 개발하였다. 개발한 수치알고리즘을 활용하여 복잡한 구조를 수치모델에서 재현하였으며, 수치모델에 적용된 수치 알고리즘의 안정성에 대해 고찰하였다.

• The KIPS Transactions:PartB
• /
• v.18B no.6
• /
• pp.365-374
• /
• 2011

Recently, malicious video classification and filtering techniques are of practical interest as ones can easily access to malicious multimedia contents through the Internet, IPTV, online social network, and etc. Considerable research efforts have been made to developing malicious video classification and filtering systems. However, the malicious video classification and filtering is not still being from mature in terms of reliable classification/filtering performance. In particular, the most of conventional approaches have been limited to using only the spatial features (such as a ratio of skin regions and bag of visual words) for the purpose of malicious image classification. Hence, previous approaches have been restricted to achieving acceptable classification and filtering performance. In order to overcome the aforementioned limitation, we propose new malicious video classification framework that takes advantage of using both the spatial and temporal features that are readily extracted from a sequence of video frames. In particular, we develop the effective temporal features based on the motion periodicity feature and temporal correlation. In addition, to exploit the best data fusion approach aiming to combine the spatial and temporal features, the representative data fusion approaches are applied to the proposed framework. To demonstrate the effectiveness of our method, we collect 200 sexual intercourse videos and 200 non-sexual intercourse videos. Experimental results show that the proposed method increases 3.75% (from 92.25% to 96%) for classification of sexual intercourse video in terms of accuracy. Further, based on our experimental results, feature-level fusion approach (for fusing spatial and temporal features) is found to achieve the best classification accuracy.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.14 no.8
• /
• pp.4033-4038
• /
• 2013

Sedimentary rocks from construction waste are discarded through open storage and landfilling, which causes an increase in construction cost and inefficient of execution of works. Some sandstone are selected and utilized as aggregates, but shale is buried as industrial waste. Therefore, in this research, we evaluated weathering properties of shale aggregate that is widely distributed throughout Daegu-Kyeongbuk region and freeze-thaw characteristics of concrete according to the replacement ratio of shale aggregate, in an effort to stabilize aggregate supply-demand in Daegu-Kyeongbuk region and develop alternative aggregates. We used red shale and black shale in the experiment, which were exported from a construction site in Deagu. We verified the usage of shale as a concrete aggregate by comparing andesite, which is broadly used as a thick aggregate for concrete, to hornfels, which is a metamorphic sedimentary rock. As a result of the experiment, we observed no degradation phenomenon for andesite and hornfels. However, a part of country rock containing black shale was found to be exfoliated. Red shale started having cracks in the direction of stratification after 1.5 months of direct exposure, and it broke into smaller pieces after approximately 4 months. After 300 cycles of freeze-thaw process on the concrete manufactured according to the replacement ratio of shale aggregate, the modulus of elasticity was 97% for plain and 95% for hornfels. In the case of RS_100, it was 57% after 210 cycles, and for BS_100, it was 54% after 240 cycles. Therefore, we established that, as the number of repetition increases, the freeze-thaw resistance decreases dramatically.

• Tuberculosis and Respiratory Diseases
• /
• v.40 no.2
• /
• pp.104-111
• /
• 1993

Background: Gram-negative bacterial endotoxin induced septicemia is known to be a leading cause in the development of adult respiratory distress syndrome(ARDS). The mechanism of endotoxin induced lung injury is mainly due to the activated neutrophils which injure the capillary endothelial cells by releasing oxidant radical and resulted in pulmonary edema. We studied the change of antioxidant enzyme in the case of large or small, intermittant dose of endotoxin infused rat lungs. Methods: Endotoxin was given to the rat through the peritoneal cavity in the dose of 7 mg/kg body weight in the large dose group and 1 mg/kg for 10 days in the small dose group. Bronchoalveolar lavage (BAL) was done and rats were killed at 6, 12, 24 hours after single endotoxin injection in the large dose group and 3, 7, 10 days after daily endotoxin injection for 10 days in the small dose group. The lungs were perfused with normal saline through the pulmonary artery to remove the blood and were homogenized in 5 volume of 50 mM potassium phosphate buffer containing 0.1 mM EDTA. After centrifuging at 100,000 g for 60 minute, the supernatent was removed and stored at $-70^{\circ}C$ until measuring for superoxide dismutase (SOD), catalase, glutathione peroxidase (GSH-Px) and protein. Results: We observed the following results. 1) The lung wet/dry weight ratio and albumin concentration in the BAL fluids were increased to peak at 12 hours and neutrophil number in the BAL fluids were peak at 6 hours after endotoxin injection in the large dose group. 2) Cu, Zn SOD (IU/mg protein) was significantly decreased after 6, 12 hours after endotoxin injection in the large dose group. 3) There were no singnificant change in the level of Mn SOD, catalase, GSH-Px after endotoxin injection in both groups. Conclusion: Endotoxin in the large dose group produced the acute pulmonary edema and decreased the Cu, Zn SOD in the lung tissue after injecting endotoxin at 6 and 12 hours. These phenomenon may be due to the cell membrane damage by endotoxin. Further research would be necessary whther giving SOD by intratracheal route or method to increase the synthesis of SOD may lessen the acute lung injury by endotoxin.

• The Journal of Engineering Geology
• /
• v.14 no.4 s.41
• /
• pp.401-416
• /
• 2004

We have collected shale specimens from the Pungam Basin in Kangwon province and investigated change of physical properties by freezing and thawing in water as well as in acidic fluid. The temperature range was $-20{\pm}2^{\circ}C\~15{\pm}2^{\circ}C$. Specimens were frozen for 12 hours and thawed in water for 8 hours. Then, they were saturated in the vacuum chamber for 4 hours to make specimens fully saturated. This procedure was 1 cycle. We have measured absorption, ultrasonic velocity, shore hardness, slake durability and uniaxial compressive strength at every 5th cycles. The physical properties increased or decreased as freezing and thawing cycles increased. Uniaxial compressive strength decreased by 0.40MPa per cycle in water and by 0.48MPa in acidic fluid. Elastic constant also decreased by 0.21GPa per cycle in water and by 0.30GPa in acidic fluid. Absorption increased by $0.29\%$ and $0.37\%$ per cycle in water and acidic fluid, respectively. These results indicate that decrease in uniaxial compressive strength, elastic constant and absorption by freezing and thawing in acidic fluid is more rapid than in water. Ultrasonic velocities, shore hardness and slake durability show no differences in water and acidic fluid. When we compared our results with the temperatures in the Hongchon during the winter season, $6\~12$ cycles may be equivalent to 1 year.

• Journal of KIISE:Computer Systems and Theory
• /
• v.27 no.1
• /
• pp.23-33
• /
• 2000

We propose a model which overlaps communication with computation for efficient communication in the data-parallel programming paradigm. The overlapping model divides a given loop partition into several sub-partitions to obtain computation which can be overlapped with communication. A loop partition sometimes refers to other data partitions, but not all iterations in the loop partition require non-local data. So, a loop partition may be divided into a set of loop iterations which require non-local data, and a set of loop iterations which do not. Each loop sub-partition is dynamically scheduled depending on associated message arrival, The experimental results for a few benchmarks in IBM SP2 show enhanced performance in our overlapping model.

• Journal of KIISE:Software and Applications
• /
• v.30 no.11
• /
• pp.1054-1061
• /
• 2003

Q-Learning, a representative algorithm of reinforcement learning, experiences repeatedly until estimation values about all state-action pairs of state space converge and achieve optimal policies. When the state space is high dimensional or continuous, complex reinforcement learning tasks involve very large state space and suffer from storing all individual state values in a single table. We introduce Q-Map that is new function approximation method to get classified policies. As an agent learns on-line, Q-Map groups states of similar situations and adapts to new experiences repeatedly. State-action pairs necessary for fine control are treated in the form of rule. As a result of experiment in maze environment and mountain car problem, we can achieve classified knowledge and extract easily rules from Q-Map

• Proceedings of the Korea Information Processing Society Conference
• /
• 2010.11a
• /
• pp.478-481
• /
• 2010

본 논문에서는 경제학, 사회학, 수학 분야에서 수십년 전부터 연구해오던 죄수의 딜레마 게임의 협동진화에 대해 고찰해보고자 한다. 반복적 죄수의 딜레마 게임은 게임이론의 가장 기본적인 이론으로써, 사회적 상호작용, 경제활동, 국제관계 등 다양한 현상들을 모델링 하기 위한 하나의 방법이다. 그 중에 N명이 참가하는 반복적 죄수 딜레마 게임의 전략은 유전 알고리즘(Genetic Algorithms, GAs)을 통해 진화적으로 만들어 낼 수 있으며, 이 경우에 그 결과를 일반적인 내쉬 균형 이 아닌, 모든 개체들이 유전알고리즘을 통해 협동으로 수렴하도록 유도할 수 있다는 사실은 상당히 시사하는 바가 크다. 기존에 주로 연구되어오던 죄수의 딜레마 게임은 협동으로의 수렴과정에서 일반적으로 순위기반선택(Rank-based selection)과 1점 교배기법(1point crossover)을 사용한다. 그러나 순위기반선택은 모든 개체에 순위을 매겨야 하기 때문에, 개체수가 커질수록 성능이 저하되며, 1점 교배기법은 개체 값이 분산되어있을 경우, 최적해(Optimal solution)을 찾기 힘들다는 단점이 있어, 개체수가 많은 경우에 적용하기에는 비효율적이다. 본 논문에서는 토너먼트 선택기법(Tournament selection)과 자기 적응형 교배기법(Self-adaptive crossover)을 적용한 새로운 기법을 제안한다. 또한 기존 기법과 비교 실험을 통해 제안기법이 기존기법에 비해 평균 수렴시간과 수렴 횟수에서 뛰어난 성능을 보이고 있음을 확인하였다.