• Nuclear Engineering and Technology
• /
• v.22 no.4
• /
• pp.359-370
• /
• 1990

This study concerns with comparing low-dimensional reactor kinetics methods with a three-dimensional kinetics method to be used for safety analysis of light water reactors in order to suggest means of preparing input parameters required for low-dimensional methods. For this purpose a one-dimensional finite difference two-group diffusion theory code ODTRAN and a third-order Hermit polynomial-based point kinetics code POTRAN are developed and used to obtain low-dimensional solutions to the LRA-BWR kinetics benchmark problem. The results are compared with a three-dimensional modified Borresen's coarse-mesh solution of the kinetics problem by CMSNACK code. Through this comparison some simple but practical means of preparing input parameters of low-dimensional kinetics analysis methods are suggested.

• Proceedings of the Korea Multimedia Society Conference
• /
• 2003.05b
• /
• pp.410-414
• /
• 2003

본 논문은 비디오 시퀀스에 카메라 패닝 보상과 2차원 시공간 엔트로피 임계법을 적용하여 추출한 객체포함영역을 대상으로 영상 분할을 수행하는 이동 객체 분할 기법에 관한 것이다. 우선, 웨이블렛 변환에 의해 구성한 피라미드 계층 구조상에서 카메라 패닝 벡터를 추정하여 전역 움직임을 보상한다. 이후, 전역 움직임이 보상된 기준영상을 대상으로 각 프레임간에서 2차원 시공간 엔트로피 임계법을 적용하여 이동 객체가 포함될 가능성이 있는 영역을 블록 단위로 추출한다. 다음으로, 2차원 시공간 엔트로피 입계법에 의해 분류된 영역을 토대로 각 블록을 움직임블록, 준 움직임 블록, 비 움직임 블록 중 어느 하나로 분류한 검색 테이블을 작성한다. 이어서, 검색 테이블을 참조하여 초기 탐색 계층 및 탐색 영역을 적응적으로 선정함으로써 피라미드 계층 구조상에서 효율적인 고속 움직임 추정을 수행하여 이동 객체에 해당하는 객체포함영역만을 추출한다. 최종적으로, 이렇게 추출된 객체포함영역에서 임계 기울기 영상을 정의한 후, 이를 기준 삼아 객체포함영역에 화소 단위의 형태학 기반 영상 분할 알고리즘을 적용함으로써 비디오 시퀀스에 포함된 이동 객체를 분할한다. 컴퓨터 시뮬레이션 결과를 통해 고찰할 때, 제안된 방법은 이동 객체에 대한 상대적으로 우수한 분할 특성을 제공할 수 있고, 특히 저대조 경계면의 분할 특성을 제고시키고 있음을 확인할 수 있다.

• Journal of IKEEE
• /
• v.18 no.1
• /
• pp.165-171
• /
• 2014

This paper presents a low complexity QR decomposition (QRD) architecture for MIMO detector. In the proposed approach, various CORDIC-based QRD algorithms are efficiently combined together to reduce the computational complexity of the QRD hardware. Based on the computational complexity analysis on various QRD algorithms, a low complexity approach is selected at each stage of QRD process. The proposed QRD architecture can be applied to any arbitrary dimension of channel matrix, and the complexity reduction grows with the increasing matrix dimension. Our QR decomposition hardware was implemented using Samsung $0.13{\mu}m$ technology. The numerical results show that the proposed architecture achieves 47% increase in the QAR (QRD Rate/Gate count) with 28.1% power savings over the conventional Householder CORDIC-based architecture for the $4{\times}4$ matrix decomposition.

• Annual Conference on Human and Language Technology
• /
• 2023.10a
• /
• pp.638-643
• /
• 2023

문서 기반 대화 시스템은 크게 질문으로부터 문서를 검색하는 과정과 응답 텍스트를 생성하는 과정으로 나뉜다. 이러한 대화 시스템의 응답 생성 과정에 디코더 기반 LLM을 사용하기 위해서 사전 학습된 LLM을 미세 조정한다면 많은 메모리, 연산 자원이 소모된다. 본 연구에서는 SVD에 기반한 LLM의 경량화를 시도한다. 사전 학습된 polyglot-ko 모델의 행렬을 SVD로 분해한 뒤, full-fine-tuning 해보고, LoRA를 붙여서 미세 조정 해본 뒤, 원본 모델을 미세 조정한 것과 점수를 비교하고, 정성평가를 수행하여 경량화된 모델의 응답 생성 성능을 평가한다. 문서 기반 대화를 위한 한국어 대화 데이터셋인 KoDoc2Dial에 대하여 평가한다.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.5 no.7
• /
• pp.1346-1353
• /
• 2001

This paper presents an algorithm for the design of 2 dimension separable denominator digital filter(SDDF). The proposed algorithm is based on the reduced dimensional decomposition not only 2 dimension SDDF's but also of given 2 dimension specification. The frequency domain design of 2 dimension separable denominator digital filters based on the reduced dimensional decomposition can be realized when the given 2 dimension frequency specification are optimally decomposed into a pair of 1 dimension digital filter specification via singular value decomposition. the algorithm is computationally efficient and numerically stable. In case of the low pass filter, the approximation error of the proposed design algorithm is $e_{m}$=5.17, $e_{r1}$ =8.78, $e_{r2}$=7.34, while in case of band pass filter, the approximation error is $e_{m}$=13.00, $e_{r1}$=62.76, $e_{r2}$=62.7676.7676

• Proceedings of the Korean Information Science Society Conference
• /
• 2006.06b
• /
• pp.358-360
• /
• 2006

LDA는 그룹간 간격을 최대화하고 그룹내 분산을 최소화하는 선형변환을 구함으로써 차원 감소된 공간에서 분별력(classification performance)을 높이는 선형 차원 감소 방법이다. 본 논문에서는 저샘플 문제(undersampled problem)에서 LDA를 적용할 수 있도록 QR-분해를 이용한 효율적인 차원 감소 방법을 제안한다. 특히 제안되는 방법은 문서 분류 문제에서처럼 한 문서가 몇 개의 카테고리에 중복적으로 속하는 경우 등 데이터의 독립성이 보장되지 않는 경우에도 효과적으로 적용될 수 있다는 장점이 있다.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.17 no.4
• /
• pp.387-397
• /
• 1992

The spatial domain design of 2-dimensional separable denominator digital filters(SDDF) based on the reduced dimensional decomposition can be realized when the given 2-D impulse response specifications are decomposed into a pair of 1-D specifications via singular value decompositions(SVD). Because of use of the balaned approximation and equivalent transform as 1-D design algorithm, 2-D design algorithm retains the advantage that is numerically stable and can minimize quantization errors. In this paper in order to analyze and reduce these errors, minimum comfficient quantization realization is directly derived from impulse response specification. And using the equivalent trans form relation between mininum coefficient quantization error and minimum roundoff error realizations, we optimally realize a SDDF. This algorithm is analyzed by the simulation, which shows that it is superior to direct or balanced realization in quantization errors.

• Journal of KIISE:Software and Applications
• /
• v.36 no.10
• /
• pp.777-785
• /
• 2009

Collaborative filtering is a technique used to predict whether a particular user will like a particular item. User-based or item-based collaborative techniques have been used extensively in many commercial recommender systems. In this paper, a hybrid collaborative filtering method that combines user-based and item-based methods using a low-dimensional linear model is proposed. The proposed method solves the problems of sparsity and a large database by using NMF among the low-dimensional linear models. In collaborative filtering systems the methods using the NMF are useful in expressing users as semantic relations. However, they are model-based methods and the process of computation is complex, so they can not recommend items dynamically. In order to complement the shortcomings, the proposed method clusters users into groups by using NMF and selects features of groups by using TF-IDF. Mutual information is then used to compute similarities between items. The proposed method clusters users into groups and extracts features of groups on offline and determines the most suitable group for an active user using the features of groups on online. Finally, the proposed method reduces the time required to classify an active user into a group and outperforms previous methods by combining user-based and item-based collaborative filtering methods.

• Journal of KIISE:Databases
• /
• v.31 no.2
• /
• pp.122-131
• /
• 2004

Traditional works on indexing have been suggested for low dimensional data under dynamic environments. But recent database applications require efficient processing of huge sire of high dimensional data under static environments. Thus many indexing strategies suggested especially in partitioning ones do not adapt to these new environments. In our study, we point out these facts and propose a new partitioning strategy, which complies with new applications' requirements and is derived from analysis. As a preliminary step to propose our method, we apply a packing technique on the one hand and exploit observations on the Minkowski-sum cost model on the other, under uniform data distribution. Observations predict that unbalanced partitioning strategy may be more query-efficient than balanced partitioning strategy for high dimensional data. Thus we propose our method, called CSP (Cyclic Spliced Partitioning method). Analysis on this method explicitly suggests metrics on how to partition high dimensional data. By the cost model, simulations, and experiments, we show excellent performance of our method over balanced strategy. By experimental studies on other indices and packing methods, we also show the superiority of our method.

• 한국노년학
• /
• v.32 no.3
• /
• pp.759-776
• /
• 2012

This study analysed the income-related health inequality of the aged in Korea applied to EQ-5D. Two decompositions were used in analysis. One was the decomposition of income-related health inequality into six different dimensions of EQ-5D, and the other was to decompose it by sub-group such as sex, region, existence of spouse. The results are summarized as follows. First, the health concentration index(CI) of the aged was .0254, which meant that there were pro-rich inequality in elderly people's health levels. The same patterns were also seen in the analysis of different dimensions of EQ-5D such as mobility, self-care, usual activities, pain/discomfort, and anxiety/depression. Second, mobility accounted for 35.8% of total EQ-5D score, most contributing to CI of the dimensions of EQ-5D. The CIs by the dimensions ranged from .0091 for mobility to .0013 for self-care. Third, The decompositions by sub-groups showed that the contributions of sex, region, and existence of spouse to health inequality was similar to each other, all of three sub-groups accounted for 10 % of inequality respectively. Fourth, the inequality within group was higher in female group, rural area, and the aged without spouse. The average health level of these groups was lower than that of the other ones, too. These facts indicated that old women, the aged without spouse, and the elderly in rural area were the most vulnerable groups in health problems. Therefore, it is necessary to pay more attention to health problems of these groups in the policy making of health security and social welfare services.