• Proceedings of the KAIS Fall Conference
• /
• 2009.12a
• /
• pp.445-448
• /
• 2009

DSS 응용들의 대용량 집계 데이터 집중 시스템에서는 효율적이고 즉각적인 의사결정 지원을 위한 근사 질의응답의 연구가 필요하다. 따라서 본 연구에서는 FCM 클러스터링 기법과 ANFIS을 이용한 기법을 제안한다. 제안된 기법은 다차원 데이터 큐브의 데이터 특성을 가지며 질의에 대한 근사적인 응답을 제공할 수 있는 모델을 생성한다. 제안된 기법을 통해 학습된 모델은 기존의 기법보다 근사 질의응답의 정확성이 향상되었음을 비교 실험을 통하여 확인한다. 따라서 제안된 기법은 기존의 기법보다 저장 공간과 시간을 줄일 수 있으며 또한 근사 응답의 정확도를 향상시킬 수 있다.

• 제어로봇시스템학회:학술대회논문집
• /
• 2001.10a
• /
• pp.93.4-93
• /
• 2001

Based on input and output data pairs of nonlinear systems, this paper proposes a simple and effective Just-In-Time (JIT) method, called Direct JIT Control, for nonlinear control design. It uses an inverse model of controlled plant to compute an initial control action, and then adapts the initial control action by adding a fine-tuning control action, depended on the errors between the real outputs and the expected reference signals. Meanwhile, the proposed JIT method accomplishes the adaptation of the inverse model just simply by means of the refreshment of input and output data pairs. In addition, the JIT modeling technique guarantees this method to obtain an approximate inverse model of the controlled nonlinear plant in the neighborhood of a query. Based on a ...

• Journal of the Korean Institute of Telematics and Electronics
• /
• v.27 no.4
• /
• pp.571-579
• /
• 1990

A dynamic relocation algorithm for non-deterministic process graph in distributed computer systems is proposed. A method is represented for determining the optimal policy for processing a process tree. A general database query request is modelled by a process tree which represent a set of subprocesses together with their precedence relationship. The process allocation model is based on operating cost which is a function fo selection of site for processing operation, data reduction function and file size. By using expected values of parameters for non-deterministic process tree, the process graph and optimal policy that yield minimum operating cost are determined. As process is relocated according to threshold value and new information of parameters after the execution of low level process for non-deterministic process graph, the assigned state that approximate to optiaml solution is obtained. The proposed algorihtm is heuristic By performing algorithm for sample problems, it is shown that the proposed algorithm is good in obtaining optimal solution.

• Journal of KIISE:Databases
• /
• v.30 no.3
• /
• pp.336-349
• /
• 2003

Histograms have been getting a lot of attention recently. Histograms are commonly utilized in commercial database systems to capture attribute value distributions for query optimization Recently, in the advent of researches on approximate query answering and stream data, the interests in histograms are widely being spread. The simplest approach assumes that the attributes in relational tables are independent by AVI(Attribute Value Independence) assumption. However, this assumption is not generally valid for real-life datasets. To alleviate the problem of approximation on multi-dimensional data with multiple one-dimensional histograms, several techniques such as wavelet, random sampling and multi-dimensional histograms are proposed. Among them, GENHIST is a multi-dimensional histogram that is designed to approximate the data distribution with real attributes. It uses overlapping buckets that allow more efficient approximation on the data distribution. In this paper, we propose a scheme, OPT that can determine the optimal frequencies of overlapped buckets that minimize the SSE(Sum Squared Error). A histogram with overlapping buckets is first generated by GENHIST and OPT can improve the histogram by calculating the optimal frequency for each bucket. Our experimental result confirms that our technique can improve the accuracy of histograms generated by GENHIST significantly.

• Journal of Digital Contents Society
• /
• v.18 no.6
• /
• pp.1151-1156
• /
• 2017

It is common to achieve lower performance in traversing tree data structures in GPU than one expects. In this paper, we analyze the reason of lower-than-expected performance in GPU tree traversal and present that the warp divergences is caused by the branch instructions ("if${\ldots}$ else") which appear commonly in tree traversal CUDA codes. Also, we compare the parallel shifted sort algorithm which can reduce the number of warp divergences with a kd-tree CUDA implementation to show that the shifted sort algorithm can work faster than the kd-tree CUDA implementation thanks to less warp divergences. As the analysis result, the shifted sort algorithm worked about 16-fold faster than the kd-tree CUDA implementation for $2^{23}$ query points and $2^{23}$ data points in $R^3$ space. The performance gaps tend to increase in proportion to the number of query points and data points.

• The KIPS Transactions:PartD
• /
• v.10D no.3
• /
• pp.427-438
• /
• 2003

In a data warehouse, users typically search for trends, patterns, or unusual data behaviors by issuing queries interactively. The OLAP range-sum query is widely used in finding trends and in discovering relationships among attributes in the data warehouse. In a recent environment of enterprises, data elements in a data cube are frequently changed. The problem is that the cost of updating a prefix sum cube is very high. In this paper, we propose a novel algorithm which reduces the update cost significantly by an index structure called the Δ-tree. Also, we propose a hybrid method to provide either approximate or precise results to reduce the overall cost of queries. It is highly beneficial for various applications that need quick approximate answers rather than time consuming accurate ones, such as decision support systems. An extensive experiment shows that our method performs very efficiently on diverse dimensionalities, compared to other methods.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.32 no.6
• /
• pp.1081-1090
• /
• 2022

The face recognition model is used for identity recognition of smartphones, providing convenience to many users. As a result, the security review of the DNN model is becoming important, with adversarial attacks present as a well-known vulnerability of the DNN model. Adversarial attacks have evolved to decision-based attack techniques that use only the recognition results of deep learning models to perform attacks. However, existing decision-based attack technique[14] have a problem that requires a large number of queries when generating adversarial examples. In particular, it takes a large number of queries to approximate the gradient. Therefore, in this paper, we propose a method of generating adversarial examples using orthogonal space sampling and dimensionality reduction sampling to avoid wasting queries that are consumed to approximate the gradient of existing decision-based attack technique[14]. Experiments show that our method can reduce the perturbation size of adversarial examples by about 2.4 compared to existing attack technique[14] and increase the attack success rate by 14% compared to existing attack technique[14]. Experimental results demonstrate that the adversarial example generation method proposed in this paper has superior attack performance.

• Journal of KIISE:Databases
• /
• v.34 no.4
• /
• pp.369-377
• /
• 2007

Histogram is one of tools that efficiently summarize data, and it is widely used for selectivity estimation and approximate query answering. Existing histogram construction algorithms are applicable to point data represented by a set of values. As often as point data, we can meet interval data such as daily temperature and daily stock prices. In this paper, we thus propose the histogram construction algorithms for interval data by extending several methods used in existing histogram construction algorithms. Our experiment results, using synthetic data, show our algorithms outperform naive extension of existing algorithms.

• Journal of the Korean Society for Precision Engineering
• /
• v.21 no.9
• /
• pp.85-94
• /
• 2004

As over 70% of the total life cycle cost (LCC) of a product is committed at the early design stage, designers are in an important position to substantially reduce the LCC of the products they design by giving due to life cycle implications of their design decisions. During early design stages, there may be competing concepts with dramatic differences. In addition, the detailed information is scarce and decisions must be made quickly. Thus, both the overhead in developing parametric LCC models fur a wide range of concepts, and the lack of detailed information make the application of traditional LCC models impractical. A different approach is needed, because a traditional LCC method is to be incorporated in the very early design stages. This paper explores an approximate method for providing the preliminary LCC, Learning algorithms trained to use the known characteristics of existing products might allow the LCC of new products to be approximated quickly during the conceptual design phase without the overhead of defining new LCC models. Artificial neural networks are trained to generalize product attributes and LCC data from pre-existing LCC studies. Then the product designers query the trained artificial model with new high-level product attribute data to quickly obtain an LCC for a new product concept. Foundations fur the learning LCC approach are established, and then an application is provided.

• The Transactions of the Korea Information Processing Society
• /
• v.4 no.6
• /
• pp.1418-1432
• /
• 1997

Most general content_based image retrieval techniques use color and texture as retrieval indices. In color techniques, color histogram and color pair based color retrieval techniques suffer from a lack of spatial information and text. And This paper describes the design and implementation of content_based image retrieval system using color space and keywords. The preprocessor for image retrieval has used the coordinate system of the existing HSI(Hue, Saturation, Intensity) and preformed to split One image into chromatic region and achromatic region respectively, It is necessary to normalize the size of image for 200*N or N*200 and to convert true colors into 256 color. Two color histograms for background and object are used in order to decide on color selection in the color space. Spatial information is obtained using a maximum entropy discretization. It is possible to choose the class, color, shape, location and size of image by using keyword. An input color is limited by 15 kinds keyword of chromatic and achromatic colors of the Korea Industrial Standards. Image retrieval method is used as the key of retrieval properties in the similarity. The weight values of color space ${\alpha}(%)and\;keyword\;{\beta}(%)$ can be chosen by the user in inputting the query words, controlling the values according to the properties of image_contents. The result of retrieval in the test using extracted feature such as color space and keyword to the query image are lower that those of weight value. In the case of weight value, the average of te measuring parameters shows approximate Precision(0.858), Recall(0.936), RT(1), MT(0). The above results have proved higher retrieval effects than the content_based image retrieval by using color space of keywords.