• The KIPS Transactions:PartB
• /
• v.14B no.4
• /
• pp.287-294
• /
• 2007

Many classification algorithms for real world data suffer from a data class imbalance problem. To solve this problem, various methods have been proposed such as altering the training balance and designing better sampling strategies. The previous methods are not satisfy in the distribution of the input data and the constraint. In this paper, we propose a focused sampling method which is more superior than previous methods. To solve the problem, we must select some useful data set from all training sets. To get useful data set, the proposed method devide the region according to scores which are computed based on the distribution of SOM over the input data. The scores are sorted in ascending order. They represent the distribution or the input data, which may in turn represent the characteristics or the whole data. A new training dataset is obtained by eliminating unuseful data which are located in the region between an upper bound and a lower bound. The proposed method gives a better or at least similar performance compare to classification accuracy of previous approaches. Besides, it also gives several benefits : ratio reduction of class imbalance; size reduction of training sets; prevention of over-fitting. The proposed method has been tested with kNN classifier. An experimental result in ecoli data set shows that this method achieves the precision up to 2.27 times than the other methods.

• The Journal of Korean Association of Computer Education
• /
• v.18 no.5
• /
• pp.63-70
• /
• 2015

In the environment that the complex conditions need to be satisfied, skyline query have been applied to various field. To processing a skyline query in centralized scheme, several techniques have been suggested and recently map/reduce platform based approaches has been proposed which divides data space into multiple partitions for the vast volume of multidimensional data. However, the performances of these approaches are fluctuated due to the uneven data loading between servers and redundant tasks. Motivated by these issues, this paper suggests a novel technique called MR-DEAP which solves the uneven data loading using the random sampling. The experimental result gains the proposed MR-DEAP outperforms MR-Angular and MR-BNL scheme.

• Smart Media Journal
• /
• v.4 no.4
• /
• pp.24-32
• /
• 2015

Recently, SDN is actively used as datacenter networks and gradually increase its applied areas. Along with this change of networking environment, research of deploying network security systems on SDN becomes highlighted. Especially, systems for detecting network flooding attacks by monitoring every packets through ports of OpenFlow switches have been proposed. However, because of the centralized management of a SDN controller which manage multiple switches, it may be substantial overhead that the attack detection system continuously monitors all the flows. In this paper, a sampling based network flooding attack detection and prevention system is proposed to reduce the overhead of monitoring packets and to achieve reasonable functionality of attack detection and prevention. The proposed system periodically takes sample packets of network flows with the given sampling conditions, analyzes the sampled packets to detect network flooding attacks, and block the attack flows actively by managing the flow entries in OpenFlow switches. As network traffic sampler, sFlow agent is used, and snort, an opensource IDS, is used to detect network flooding attack from the sampled packets. For active prevention of the detected attacks, an OpenDaylight application is developed and applied. The proposed system is evaluated on the local testbed composed with multiple OVSes (Open Virtual Switch), and the performance and overhead of the proposed system under various sampling condition is analyzed.

• Korean Journal of Agricultural and Forest Meteorology
• /
• v.12 no.3
• /
• pp.197-206
• /
• 2010

To provide baseline information essential for assessing environmental impacts of monsoon rainfalls in a mountainous watershed under mixed land use, we investigated spatiotemporal variations in water quality using a combined approach of seasonal water quality survey and intensive storm samplings. Biannual water sampling at nine locations encompassing major land use types showed generally lower electrical conductivity and Cl- concentrations during the typical wet period compared to the dry period, indicating rainfall-induced dilution of dissolved ions. Total metal concentrations, however, were significantly higher during the monsoon period, probably associated with rainfall-induced increases in suspended sediments. Intensive storm sampling during a small monsoon rainfall event (18 mm) and an extreme event (452 mm) showed rapid changes in both suspended sediments and dissolved solutes in an agricultural stream draining the Haean Basin where arable lands have expanded rapidly over the recent decades. By contrast, a nearby forest stream derived from North Korea showed little responses to the small event compared to larges changes during the extreme event. In the agricultural stream total Pb concentrations showed significant positive relationships with suspended sediments. Although limited sampling frequency and locations require a cautious interpretation, the overall results suggest that expansion of agricultural fields in steep mountainous watersheds can increase the susceptibility of soil erosion and its off-site environmental impacts under increasing rainfall variability and extremes.

• Proceedings of the Korean Information Science Society Conference
• /
• 2012.06b
• /
• pp.324-326
• /
• 2012

기계 학습에서 입력 데이터의 차원을 줄이는 문제(dimension reduction)는 매우 중요한 문제 중의 하나이다. 입력 변수의 차원이 늘어남에 따라 처리해야하는 연산의 수와 계산 복잡도가 급격히 늘어나기 때문이다. 이를 해결하기 위하여 다수의 기계 학습 알고리즘은 명시적으로 차원을 줄이거나(feature selection), 데이터에 약간의 연산을 가하여 차원이 작은 새로운 입력 데이터를 만든다(feature extraction). 반면 사람이 여러 종류의 고차원 센서 데이터를 입력받아 빠른 시간 안에 정확하게 정보를 처리할 수 있는 가장 큰 이유 중 하나는 실시간으로 판단하여 가장 필요한 정보에 집중하기 때문이다. 본 연구는 사람의 정보 처리 과정을 기계 학습 알고리즘에 반영하여, 집중도를 이용하여 효율적으로 데이터를 처리하는 방법을 제시한다. 이 성질을 시각 하이퍼네트워크 모델에 반영하여, 효율적으로 고차원 입력 데이터를 다루는 방법을 제안한다. 실험에서는 시각 하이퍼네트워크를 이용하여 고차원의 이미지 데이터에서 행동을 분류하였다.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.8 no.6
• /
• pp.899-905
• /
• 2013

In this paper, we focus on how to calibrate multi-cameras easily and how to efficiently detect quad-copters with small-numbered particles. Each particle is a six dimensional vector that is composed of 3D position and 3D orientation of a quad-copter in the space. Due to curse of dimensionality, that leads to explosive computational costs with a large amount of high-dimensioned particles. To detect efficiently, we need to put more particles in very promising spaces and few particles in other spaces. Though computational cost is lowered by minimizing particles, in order to track a quad-copter with multiple cameras in real-time, multiple images from the cameras should be synchronized and analyzed. Therefore, lots of the computations still need to be done. Because of this, GPGPU(General-Purpose computing on Graphics Processing Units) is implemented for parallel computing. This method has been successfully tested and gives accurate results in practical situations.

• Proceedings of the Korean Information Science Society Conference
• /
• 2001.04b
• /
• pp.304-306
• /
• 2001

많은 실세계의 문제에서 일반적인 패턴 분류 알고리즘들은 데이터의 불균형 문제에 어려움을 겪는다. 각각의 학습 예제에 균등한 중요도를 부여하는 기존의 기법들은 문제의 특징을 제대로 파악하지 못하는 경우가 많다. 본 논문에서는 불균형 데이터 문제를 해결하기 위해 퍼셉트론에 기반한 부스팅 기법을 제안한다. 부스팅 기법은 학습을 어렵게 하는 데이터에 집중하여 앙상블 머신을 구축하는 기법이다. 부스팅 기법에서는 약학습기를 필요로 하는데 기존 퍼셉트론의 경우 문제에 따라 약학습기(weak learner)의 조건을 만족시키지 못하는 경우가 있을 수 있다. 이에 커널을 도입한 커널 퍼셉트론을 사용하여 학습기의 표현 능력을 높였다. Reuters-21578 문서 집합을 대상으로 한 문서 여과 문제에서 부스팅 기법은 다층신경망이나 나이브 베이스 분류기보다 우수한 성능을 보였으며, 인공 데이터 실험을 통하여 부스팅의 샘플링 경향을 분석하였다.

• Proceeding of EDISON Challenge
• /
• 2017.03a
• /
• pp.172-174
• /
• 2017

메탄올 용매 내에서 1,2-dichloroethane ($C_2H_4Cl_2$)의 photo-induced halogen elimination 과정을 계산화학적 방법으로 분석하였다. 특히 실험적 방법으로 분석이 까다로운 중간체 및 반응 메커니즘 분석에 집중하였다. DFT(${\omega}B97XD$ / aug-cc-pVTZ) 계산을 기반으로 진행하였으며, 추가적인 분석을 위해 중간체 샘플링 프로그램이 사용되었다. 그 결과 $C_2H_4Cl_2$ 반응계와 유사하게 bridged 형태의 중간체가 생성되는 것을 확인하였다. 또한 반응물, 생성물, 중간체 2개 및 transition states 2개로 구성된 반응 메커니즘을 밝혀내었다.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.7 no.1
• /
• pp.69-74
• /
• 2003

In this paper, we propose the synchronization recovery algorithm which is suitable to wireless Multimedia of wireless channel situation which is being used OFDM signaling method. The basic of the suggested clock synchronization. restoration Algorithm is to getting the shock response of channel or getting the multipath strength profile through IFTT after the getting the frequency, response of deducted channel from channel deducted of receiver and to trace the location in the channel energy concentrated area of timing area. And it also analysis the start point of 64-QAM and 16-QAM if the sampling clock offset has the sample of ${\pm}$ 1-3, and we identified the occurance of performance deterioration when occures more than 2 samples of offset to compare with star point and BER performance in optimum sampling point result of BER performance checking, and we know that the recovery algorithm proposed algorithm also provide excellent synchronization characteries under frequency, selecting fading channel as result of simulation.

• Journal of Broadcast Engineering
• /
• v.25 no.2
• /
• pp.242-251
• /
• 2020

Recently, many deep convolutional neural networks for image super-resolution have been studied. Existing deep learning-based super-resolution algorithms are architecture that up-samples the resolution at the end of the network. The post-upsampling architecture has an inefficient structure at large scaling factor result of predicting a lot of information for mapping from low-resolution to high-resolution at once. In this paper, we propose a single image super-resolution using Channel Attention Residual Dense Block based on an iterative up-down sampling architecture. The proposed algorithm efficiently predicts the mapping relationship between low-resolution and high-resolution, and shows up to 0.14dB performance improvement and enhanced subjective image quality compared to the existing algorithm at large scaling factor result.