• 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.

• Proceedings of the Korean Information Science Society Conference
• /
• 2011.06c
• /
• pp.112-114
• /
• 2011

On multimedia databases, in order to realize the fast access method, indexing methods for the multidimension data space are used. However, since it is a premise to use the Euclid distance as the distance measure, this method lacks in flexibility. On the other hand, there are metric indexing methods which require only to satisfy distance axiom. Since metric indexing methods can also apply for distance measures other than the Euclid distance, these methods have high flexibility. This paper proposes an improved method of VP-tree which is one of the metric indexing methods. VP-tree follows the node which suits the search range from a route node at searching. And distances between a query and all objects linked from the leaf node which finally arrived are computed, and it investigates whether each object is contained in the search range. However, search speed will become slow if the number of distance calculations in a leaf node increases. Therefore, we paid attention to the candidates selection method using the triangular inequality in a leaf node. As the improved methods, we propose a method to use the nearest neighbor object point for the query as the datum point of the triangular inequality. It becomes possible to make the search range smaller and to cut down the number of times of distance calculation by these improved methods. From evaluation experiments using 10,000 image data, it was found that our proposed method could cut 5%~12% of search time of the traditional method.

• Journal of Information Processing Systems
• /
• v.17 no.5
• /
• pp.1020-1033
• /
• 2021

Nowadays, online or mobile social network services (SNS) are very popular and widely spread in our society and daily lives to instantly share, disseminate, and search information. In particular, SNS such as YouTube, Flickr, Facebook, and Amazon allow users to upload billions of images or videos and also provide a number of multimedia information to users. Information retrieval in multimedia-rich SNS is very useful but challenging task. Content-based media retrieval (CBMR) is the process of obtaining the relevant image or video objects for a given query from a collection of information sources. However, CBMR suffers from the dimensionality curse due to inherent high dimensionality features of media data. This paper investigates the effectiveness of the kernel trick in CBMR, specifically, the kernel principal component analysis (KPCA) for dimensionality reduction. KPCA is a nonlinear extension of linear principal component analysis (LPCA) to discovering nonlinear embeddings using the kernel trick. The fundamental idea of KPCA is mapping the input data into a highdimensional feature space through a nonlinear kernel function and then computing the principal components on that mapped space. This paper investigates the potential of KPCA in CBMR for feature extraction or dimensionality reduction. Using the Gaussian kernel in our experiments, we compute the principal components of an image dataset in the transformed space and then we use them as new feature dimensions for the image dataset. Moreover, KPCA can be applied to other many domains including CBMR, where LPCA has been used to extract features and where the nonlinear extension would be effective. Our results from extensive experiments demonstrate that the potential of KPCA is very encouraging compared with LPCA in CBMR.

• Journal of KIISE:Databases
• /
• v.34 no.6
• /
• pp.546-553
• /
• 2007

As ontologies are used widely, interest for semantic similarity search is also increasing. In this paper, we suggest a query evaluation scheme for k-nearest neighbor query, which retrieves k most similar objects to the query object. We use the best match method to calculate the semantic similarity between objects and use the signature tree to index annotation information of objects in database. The signature tree is usually used for the set similarity search. When we use the signature tree in similarity search, we are required to predict the upper-bound of similarity for a node; the highest similarity value which can be found when we traverse into the node. So we suggest a prediction function for the best match similarity function and prove the correctness of the prediction. And we modify the original signature tree structure for same signatures not to be stored redundantly. This improved structure of signature tree not only reduces the size of signature tree but also increases the efficiency of query evaluation. We use the Gene Ontology(GO) for our experiments, which provides large ontologies and large amount of annotation data. Using GO, we show that proposed method improves query efficiency and present several experimental results varying the page size and using several node-splitting methods.

• Journal of Korea Spatial Information System Society
• /
• v.11 no.3
• /
• pp.55-62
• /
• 2009

Location based Services in the ubiquitous computing environment, namely u-LBS, use very large and skewed spatial objects that are closely related to locational information. It is especially essential to achieve fast search, which is looking for POI(Point of Interest) related to the location of users. This paper examines how to search large and skewed POI efficiently in the u-LBS environment. We propose the Dynamic-level Grid based R-Tree(DGR-Tree), which is an index for point data that can reduce the cost of stationary POI search. DGR-Tree uses both R-Tree as a primary index and Dynamic-level Grid as a secondary index. DGR-Tree is optimized to be suitable for point data and solves the overlapping problem among leaf nodes. Dynamic-level Grid of DGR-Tree is created dynamically according to the density of POI. Each cell in Dynamic-level Grid has a leaf node pointer for direct access with the leaf node of the primary index. Therefore, the index access performance is improved greatly by accessing the leaf node directly through Dynamic-level Grid. We also propose a K-Nearest Neighbor(KNN) algorithm for DGR-Tree, which utilizes Dynamic-level Grid for fast access to candidate cells. The KNN algorithm for DGR-Tree provides the mechanism, which can access directly to cells enclosing given query point and adjacent cells without tree traversal. The KNN algorithm minimizes sorting cost about candidate lists with minimum distance and provides NEB(Non Extensible Boundary), which need not consider the extension of candidate nodes for KNN search.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.31 no.4
• /
• pp.259-267
• /
• 2013

GIS-based web map service is all the more accessible to the public. Among others, location query services are most frequently utilized, which are currently restricted to only one keyword search. Although there increases the demand for the service for querying multiple keywords corresponding to sequential activities(banking, having lunch, watching movie, and other activities) in various locations POI, such service is yet to be provided. The objective of the paper is to develop the k-IPS algorithm for quickly and accurately querying multiple POIs that internet users input and locating the search outcomes on a web map. The algorithm is developed by utilizing hierarchical tree structure of $R^*$-tree indexing technique to produce overlapped geometric regions. By using recursive $R^*$-tree index based spatial join process, the performance of the current spatial join operation was improved. The performance of the algorithm is tested by applying 2, 3, and 4 multiple POIs for spatial query selected from 159 keyword set. About 90% of the test outcomes are produced within 0.1 second. The algorithm proposed in this paper is expected to be utilized for providing a variety of location-based query services, of which demand increases to conveniently support for citizens' daily activities.

• The KIPS Transactions:PartD
• /
• v.8D no.4
• /
• pp.305-312
• /
• 2001

Recently, high-dimensional index structures have been required for similarity search in such database applications s multimedia database and data warehousing. In this paper, we propose a new cell-based signature tree, called CS-tree, which supports efficient storage and retrieval on high-dimensional feature vectors. The proposed CS-tree partitions a high-dimensional feature space into a group of cells and represents a feature vector as its corresponding cell signature. By using cell signatures rather than real feature vectors, it is possible to reduce the height of our CS-tree, leading to efficient retrieval performance. In addition, we present a similarity search algorithm for efficiently pruning the search space based on cells. Finally, we compare the performance of our CS-tree with that of the X-tree being considered as an efficient high-dimensional index structure, in terms of insertion time, retrieval time for a k-nearest neighbor query, and storage overhead. It is shown from experimental results that our CS-tree is better on retrieval performance than the X-tree.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.9 no.7
• /
• pp.2599-2613
• /
• 2015

Multi-index hashing (MIH) is the state-of-the-art method for indexing binary codes, as it di-vides long codes into substrings and builds multiple hash tables. However, MIH is based on the dataset codes uniform distribution assumption, and will lose efficiency in dealing with non-uniformly distributed codes. Besides, there are lots of results sharing the same Hamming distance to a query, which makes the distance measure ambiguous. In this paper, we propose a data-oriented multi-index hashing method (DOMIH). We first compute the covariance ma-trix of bits and learn adaptive projection vector for each binary substring. Instead of using substrings as direct indices into hash tables, we project them with corresponding projection vectors to generate new indices. With adaptive projection, the indices in each hash table are near uniformly distributed. Then with covariance matrix, we propose a ranking method for the binary codes. By assigning different bit-level weights to different bits, the returned bina-ry codes are ranked at a finer-grained binary code level. Experiments conducted on reference large scale datasets show that compared to MIH the time performance of DOMIH can be improved by 36.9%-87.4%, and the search accuracy can be improved by 22.2%. To pinpoint the potential of DOMIH, we further use near-duplicate image retrieval as examples to show the applications and the good performance of our method.

• Journal of KIISE:Software and Applications
• /
• v.29 no.3
• /
• pp.192-201
• /
• 2002

Automated text classification is considered as an important method to manage and process a huge amount of documents in digital forms that are widespread and continuously increasing. Recently, text classification has been addressed with machine learning technologies such as k-nearest neighbor, decision tree, support vector machine and neural networks. However, only few investigations in text classification are studied on real problems but on well-organized text corpus, and do not show their usefulness. This paper proposes and analyzes text classification methods for a real application, email document classification task. First, we propose a combining method of multiple neural networks that improves the performance through the combinations with maximum and neural networks. Second, we present another strategy of combining multiple machine learning classifiers. Voting, Borda count and neural networks improve the overall classification performance. Experimental results show the usefulness of the proposed methods for a real application domain, yielding more than 90% precision rates.

• The KIPS Transactions:PartD
• /
• v.15D no.3
• /
• pp.399-410
• /
• 2008

This paper suggests a new retrieval method using weighted vector sum to resolve a problem of traditional location-based retrieval method, nearest neighbor (NN) query, and NN query using direction. The proposed method filters out data with the radius, and then the remained retrieval area is filtered by a direction information compounded of a user's moving direction, a pre-fixed interesting direction, and a pre-fixed retrieval angle. The moving direction is computed from a vector or a weighted vector sum of several vectors using a weight to adopt several cases. The retrieval angle can be set from traditional $360^{\circ}$ to any degree you want. The retrieval data for this method can be a still and moving image recorded shooting location, and also several type of media like text, web, picture offering to customer with location of company or resort. The suggested method guarantees more accurate retrieval than traditional location-based retrieval methods because that the method selects data within the radius and then removes data of useless areas like passed areas or an area of different direction. Moreover, this method is more flexible and includes the direction based NN.