• KIPS Transactions on Software and Data Engineering
• /
• v.7 no.2
• /
• pp.51-62
• /
• 2018

High utility sequential pattern (HUSP) mining has been considered as an important research topic in data mining. Although some algorithms have been proposed for this topic, they incur the problem of producing a large search space for HUSPs. The tighter utility upper bound of a sequence can prune more unpromising patterns early in the search space. In this paper, we propose a sequence expected utility (SEU) as a new utility upper bound of each sequence, which is the maximum expected utility of a sequence and all its descendant sequences. A sequence utility list for each pattern is used as a new data structure to maintain essential information for mining HUSPs. We devise an algorithm, high sequence utility list-span (HSUL-Span), to identify HUSPs by employing SEU. Experimental results on both synthetic and real datasets from different domains show that HSUL-Span generates considerably less candidate patterns and outperforms other algorithms in terms of execution time.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.15 no.1
• /
• pp.157-162
• /
• 2011

In this paper, we propose enhanced methods of path finding based on an abstract graph with extension of search space to improve the quality of path. The proposed methods that are called simple buffering method, velocity constrained method and distance constrained method are to extract buffering-cells for using search space with valid-cells. The simple buffering method is to extract adjacent cells of valid-cells as buffering-cells. velocity constrained method and distance constrained method are based on simple buffering method, these eliminate buffering-cells through each of threshold. In experiment, proposed methods can improve the quality of path. The proposed methods are applicable to develop various kinds of telematics application, such as path finding and logistics.

• Journal of KIISE:Databases
• /
• v.34 no.1
• /
• pp.70-83
• /
• 2007

Most of works for skyline queries have focused on static data objects. With the advance in mobile applications, however, the need of continuous skyline queries for moving objects has been increasing. To process continuous skyline queries, the 4-phased decision method of skyline regions has been proposed recently. However, it is not feasible for a large number of data because of the high cost of computing skyline regions. To solve this problem, this paper first provides a theoretical analysis of the 4-phased decision method. Then we propose a progressive decision method of skyline regions for the 4-phased decision method, which consists of a distance-based pruning and an extent shrinking of region decision lines. The proposed method can efficiently reduce the cost of the decision of skyline region in the 4-phased decision method. This paper also presents the experimental results to show the effectiveness of the proposed method.

• Journal of KIISE:Databases
• /
• v.37 no.4
• /
• pp.185-196
• /
• 2010

Existing methods to compute reverse skyline queries are not correct to process the queries in dataset with region objects which have conditions like a price is 5~7 dollars and a distance to beach is 1km~2km, since they consider datasets with only point objects. To solve the problem, we propose a novel method to process reverse skyline queries for region objects in this paper. It has advantages. First, it is expected to get a good performance, because it is extended from efficient reverse skyline (ERSL) algorithm which is a best algorithm to computing reverse skyline queries in datasets with point objects. Second, it can give a right of choice unlike the others to a person requesting the query. That is because results of reverse skyline have a difference preference according to proposed pruning methods and overlap relations. This algorithm is a first for supporting region objects. Therefore there are not any other algorithms to compare their performance. For that reason, our experiment to prove the efficiency of proposed algorithm is focused what conditions give an effect to its performance and result and how much time it needs to process the query.

• Journal of KIISE:Software and Applications
• /
• v.37 no.2
• /
• pp.85-94
• /
• 2010

This paper introduces a novel way to leverage convolution parse tree kernel to extract the interaction information between two proteins in a sentence without multiple features, clues and complicated kernels. Our approach needs only the parse tree alone of a candidate sentence including pairs of protein names which is potential to have interaction information. The main contribution of this paper is two folds. First, we show that for the PPI, it is imperative to execute parse tree pruning removing unnecessary context information in deciding whether the current sentence imposes interaction information between proteins by comparing with the latest existing approaches' performance. Secondly, this paper presents that tree kernel decay factor can play an pivotal role in improving the extraction performance with the identical learning conditions. Consequently, we could witness that it is not always the case that multiple kernels with multiple parsers perform better than each kernels alone for PPI extraction, which has been argued in the previous research by presenting our out-performed experimental results compared to the two existing methods by 19.8% and 14% respectively.

• Proceedings of the Korean Information Science Society Conference
• /
• 2006.10c
• /
• pp.183-188
• /
• 2006

허밍을 통한 유사 검색 질의가 주어질 때 효과적으로 음악 데이터베이스를 검색하는 시스템에 대한 연구는 다양한 방향으로 진행되어 왔다. 최근에는 음악 데이터와 허밍 질의를 시계열 데이터로 보고 시계열 데이터 유사 검색과 관련하여 제안되어 왔던 여러 가지 거리 척도(distance measure)나 인덱싱 기법등을 적용하여 효과적으로 질의를 처리하려는 시도가 계속 되고 있다. 허밍 질의의 특성을 고려한 균일 스케일링(Uniform Scaling)을 사용하여 효과적인 유사 검색을 하는 방법은 가장 최근 제시된 방법 중 하나이다. 본 논문에서는 허밍을 통한 유사 검색 시스템인 Humming BIRD(Humming Based similaR miDi music retrieval system)를 제안하고 구현하였다. 슬라이딩 윈도우를 사용하여 음악의 임의의 부분에 대한 허밍 질의를 처리할 수 있도록 하였으며 효율적인 검색을 위해 중심을 일치시킨(center-aligned) 균일 스케일링을 제안하고 이 거리의 하한을 계산하는 하계 함수를 사용하여 탐색 공간(search space)을 효과적으로 줄여 더 빠르고 효과적인 유사 검색을 가능하도록 하였으며 실험을 통해 중심을 일치시킨된 균일 스케일링이 이전과 같은 검색 결과를 얻으면서도 효과적으로 검색함을 탐색 공간을 줄이는 가지치기 성능을 비교함으로써 보였다.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.24 no.1
• /
• pp.141-144
• /
• 2020

Convolutional neural networks (CNNs) show high performance in computer vision tasks including object detection, but a lot of weight storage and computation is required. In this paper, a pruning scheme is applied to CNNs for object detection, which can remove much amount of weights with a negligible performance degradation. Contrary to the previous ones, the pruning scheme applied in this paper considers the base accelerator architecture. With the consideration, the pruned CNNs can be efficiently performed on an ASIC or FPGA accelerator. Even with the constrained pruning, the resulting CNN shows a negligible degradation of detection performance, less-than-1% point degradation of mAP on VOD0712 test set. With the proposed scheme, CNNs can be applied to objection dtection efficiently.

• The KIPS Transactions:PartB
• /
• v.18B no.2
• /
• pp.87-92
• /
• 2011

In Korean documents, there are diverse spellings of transliterated foreign loanwords. This fact diminishes the performance of information retrieval systems in that a foreign word can be recognized differently, which is to say, as two or several different words. Thus, information retrieval systems require preprocessing to correct nonstandard loanword spellings prior to searching and recognizing corresponding equivalent words. This paper proposes a method that improves precision and processing efficiency using the Korean alphabet's viable prefix, which prunes a virtual tree from which candidate loanwords are created.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.38A no.9
• /
• pp.782-791
• /
• 2013

This paper proposes the location estimation techniques of distributed targets with the multi-sensor data perceived through IR sensors of the military robots. In order to match up targets with measured azimuths, we apply the maximum likelihood (ML), depth-first, and breadth-first tree search algorithms, in which we use the measured azimuths and the number of pixels on IR screen for pruning branches and selecting candidates. After matching up targets with azimuths, we estimate the coordinate of each target by obtaining the intersection point of the azimuths with the least square error (LSE) algorithm. The experimental results show the probability of missing target, mean of the number of calculating nodes, and mean error of the estimated coordinates of the proposed algorithms.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2019.11a
• /
• pp.44-45
• /
• 2019

최근 CNN 등 인공신경망은 최근 이미지 분류, 객체 인식, 자연어 처리 등 다양한 분야에서 뛰어난 성능을 보이고 있다. 그러나, 대부분의 분야에서 보다 더 높은 성능을 얻기 위해 사용한 인공신경망 모델들은 파라미터 수 및 연산량 등이 방대하여, 모바일 및 IoT 디바이스 같은 연산량이나 메모리가 제한된 환경에서 추론하기에는 제한적이다. 따라서 연산량 및 모델 파라미터 수를 압축하기 위한 딥러닝 경량화 알고리즘이 연구되고 있다. 본 논문에서는 임베디트 보드에서의 압축된 CNN 모델의 성능을 검증한다. 인공지능 지원 맞춤형 칩인 QCS605 를 내장한 임베디드 보드에서 카메라로 입력한 영상에 대해서 원 CNN 모델과 압축된 CNN 모델의 분류 성능과 동작속도 비교 분석한다. 본 논문의 실험에서는 CNN 모델로 MobileNetV2, VGG16 을 사용했으며, 주어진 모델에서 가지치기(pruning) 기법, 양자화, 행렬 분해 등의 인공신경망 압축 기술을 적용하였을 때 원래의 모델 대비 추론 시간 및 분류의 정확도 성능을 분석하고 인공신경망 압축 기술의 유용성을 확인하였다.