동적 선택률 추정 기법을 위한 오차 분석

황환규,김상욱 강원대학교 정보통신연구소 1997 정보통신논문지 Vol.1 No.-

In this paper, we analyze the errors occurred in a selectivity estimation method based on dynamic maintenance of data distribution, which employs the mulitilevel grid file(MLGF), a multidimensional file structure. We first demonstrate that the estimation errors stem from the uniformity assumption that records are uniformly distributed in their belonging region represented by an entry in a level of an MLGF directory. Based on this demonstration, we then investigate five factors affecting the accuracy of estimation: (1) the data distribution in a region, (2) the number of records stored in an MLGF, (3) the page size, (4) the query region size, and (5) the level of an MLGF directory. Next, we present the tendency of estimation errors according to the change of values for each factor through experiments. The results show that the errors decrease when (1) the distribution of records in a region becomes closer to the uniform one, (2) the number of records in an MLGF increases, (3) the page size decreases, (4) the query region size increases, and (5) the level of an MLGF directory employed as data distribution information becomes lower.

새로운 데이타베이스 응용을 위한 대형 데이타 관리 기법에 관한 고찰

황환규,김상욱 강원대학교 정보통신연구소 1997 정보통신논문지 Vol.1 No.-

Previous DBMSs have been built based on the simple assumption that a date item can reside on a disk page since traditional database applications such as bank service systems, business automation systems, and office automation systems have dealt with only small data items. However, with the advent of new advanced multimedia applications frequently dealing with large data items, this assumption is no longer acceptable. For supporting such new applications, DBMSs have to provide the additional facilities for managing large data items whose size are larger than that of a page. Recently, some researches have been performed aiming at this issue through developing advanced DBMSs and storage systems. In this paper, we discuss recent research results on large data management. We first present design considerations for large data management techniques. Next, we introduce the five techniques employed in commercial DBMSs and advanced storage systems such as WiSS, Exodus, Starburst, and EOS, and then compare those techniques focusing on (1) data structures, (2) algorithms, and (3) advantages and disadvantages. We also introduce domestic research activities, and finally suggest future research directions related to this issue.

B+트리를 위한 벌크로드

황환규,김상욱 江原大學校 産業技術硏究所 1995 産業技術硏究 Vol.15 No.-

In this paper, we propose a bulk-load algorithm for B' -trees, the most widely used index structures in database systems. The main characteristic of our algorithm is to simultaneously process all the keys to be placed on each B' -tree page when accessing the page. This avoids the overhead for accessing the same page multiple times, which results from applying the B' -tree insertion algorithm repeatedly. For performance evaluation, we analyze our algorithm in terms of the number of disk accesses. The results show that the number of disk accesses excluding those in the redistribution process in identical to the number of B' -tree pages. Considering that the redistribution process is an unavoidable preprocessing step for bulk-loading, our algorithm requires just one disk access per B' -tree page, and therefore turns out to be optimal. We also present performance tendancy according to the changes of parameter values via simulation.

CSPAM: 닫힌 빈발 시퀀스 패턴을 위한 효과적인 가지치기 전략

황환규 강원대학교 정보통신연구소 2009 정보통신논문지 Vol.13 No.-

Sequential pattern mining finds all of the frequent sequences satisfying a minimum support threshold in a large database. However, when mining long frequent sequences, or when using very low support thresholds, the performance of currently reported algorithms often degrades dramatically. In this paper we present a novel algorithm, CSPAM (Closed Sequential PAttern Mining), using only closed frequent sequences which are small subset of very large frequent sequences. When systematically exploring the search space of closed frequent sequences, we present two novel pruning strategies, which prune the unpromising parts of search space effectively. The experimental results show that the proposed method outperforms the previous algorithms.

GIS에서의 공간 조인 기법의 관한 고찰

황환규,김상욱 강원대학교 정보통신연구소 1998 정보통신논문지 Vol.2 No.-

The spatial join operation is one of the fundamental spatial database query operations. It facilitates the retrieval of information from two different spatial objects based on spatial relationships. The join is one of the most difficult operations to implement efficiently in spaital database systems, as it was with relational database ststems. In this paper, different kinds of spatial joins and various implementation techniques are surveyed. We investigate the basic idea, algorithms, and performance of the various spatial join techniques. In addiotion, we discuss the issues of the different spatial join techniques.

공간 영역 질의의 선택률 추정을 위한 향상된 면적 균등 분할 방법

황환규,문현수 강원대학교 정보통신연구소 2000 정보통신논문지 Vol.4 No.-

Selectivity estimation of queries in relational databases has been studied extensively. Although relational selectivity estimation problems are well studied in the database literature, spatial selectivity estimation problems are not well understood. In this paper we examine selectivity estimation in spatial databases. In particular, we focus on spatial range queries over two-dimensional rectangular data. We propose and implement an improved equi-area partition for selectivity estimation of spatial range queries. We present a detailed experimental study comparing the proposed techniques and the currently used equi-area and nonequi-area partition for selectivity estimation.

대용량의 데이터셋에서 깊이우선탐색을 사용한 순차 패턴 마이닝

김형근,황환규 강원대학교 정보통신연구소 2005 정보통신논문지 Vol.9 No.-

Recent sequential pattern mining algorithms mine all of the frequent sequences satisfying a minimum support threshold in a large database. However, when a frequent sequence becomes very long, such mining will generate an explosive number of frequent sequence, which is prohibitively expensive in time. In this paper, we proposed a novel sequential pattern algorithm using only closed frequent sequences which are small subset of very large frequent sequences. Our algorithm extends the sequence by depth-first search strategy with effective pruning. Using bitmap representation of underlying databases, we can obtain a closed frequent sequence considerably faster than the currently reported methods.

적외선 센서를 사용한 가상 키보드의 구현

장수호,황환규 江原大學校 産業技術硏究所 2009 産業技術硏究 Vol.29 No.A

In this paper, we implement a virtual keyboard using infrared distance sensor. A virtual keyboard allows a user to enter characters by selecting keyboard layouts with sensing areas. By projecting infrared light on any flat surface and detection devices we can sense the user's input characters. Unlike a conventional physical keyboard, the virtual keyboard provides convenience in terms of mobility, portability, and space savings.

효율적인 닫힌 빈발 시퀀스 마이닝

김형근,황환규 江原大學校 産業技術硏究所 2005 産業技術硏究 Vol.25 No.A

Recent sequential pattern mining algorithms mine all of the frequent sequences satisfying a minimum support threshold in a large database. However, when a frequent sequence becomes very long, such mining will generate an explosive number of frequent sequence, which is prohibitively expensive in time. In this paper, we proposed a novel sequential pattern algorithm using only closed frequent sequences which are small subset of very large frequent sequences. Our algorithm extends the sequence by depth-first search strategy with effective pruning. Using bitmap representation of underlying databases, we can obtain a closed frequent sequence considerably faster than the currently reported methods.

영역기반 이미지 검색 시스템 설계 및 구현

정호영,황환규 강원대학교 정보통신연구소 2001 정보통신논문지 Vol.5 No.-

In the real world, natural scenes are rich in color and texture. It is known to be difficult to identify image regions containing the same color-texture pattern in natural scenes. In this paper, we propos a new approach to designing a region-based image retrieval. First, colors in the image are quantized to serveral representing classes that can be used to differentiate edge in the image. We get an gray level edge image with Fisher's multi-class linear discriminant, make a binary image using a local threshold value, find a snapped edge, and then link to the nearest edge. Finally, we can get the last segmentation image by merging similar regions. In addition, we implement the region-based image retrieval system using that algorithm.