• Journal of the Korea Society of Computer and Information
• /
• v.20 no.2
• /
• pp.21-28
• /
• 2015

In this paper, we propose a prototype-based classification learning by using the nearest-neighbor rule. The nearest-neighbor is applied to segment the class area of all the training data into spheres within which the data exist from the same class. Prototypes are the center of spheres and their radii are computed by the mid-point of the two distances to the farthest same class point and the nearest another class point. And we transform the prototype selection problem into a set covering problem in order to determine the smallest set of prototypes that include all the training data. The proposed prototype selection method is based on a greedy algorithm that is applicable to the training data per class. The complexity of the proposed method is not complicated and the possibility of its parallel implementation is high. The prototype-based classification learning takes up the set of prototypes and predicts the class of test data by the nearest neighbor rule. In experiments, the generalization performance of our prototype classifier is superior to those of the nearest neighbor, Bayes classifier, and another prototype classifier.

• The KIPS Transactions:PartA
• /
• v.9A no.4
• /
• pp.399-404
• /
• 2002

Along the evolution of information and communication technologies, manufacturing embedded systems such as PDA (personal digital assistant), HPC (hand -held PC), settop box. and information appliance became realistic. And RTOS (real-time operating system) and filesystem have been played essential re]os within the embedded systems as well. For the filesystem of embedded systems, flash memory has been used extensively instead of traditional hard disk drives because of embedded system's requirements like portability, fast access time, and low power consumption. Other than these requirements, nonvolatile storage characteristic of flash memory is another reason for wide adoption in industry. However, there are some technical challenges to cope with to use the flash memory as an indispensable component of the embedded systems. These would be relatively slow cleaning time and the limited number of times to write-and-clean. In this paper, a new cleaning policy is proposed to overcome the problems mentioned above and relevant performance comparison results will be provided. Ranking cleaning policy(RCP) decides when and where to clean within the flash memory considering the cost of cleaning and the number of times of cleaning. This method will maximize not only the lifetime of flash memory but also the performance of access time and manageability. As a result of performance comparison, RCP has showed about 10 ~ 50% of performance evolution compared to traditional policies, Greedy and Cost-benefit methods, by write throughputs.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.16 no.6
• /
• pp.683-690
• /
• 2006

Reducing power consumption to extend network lifetime is one of the most important challenges in designing wireless sensor networks. One promising approach to conserving system energy is to keep only a minimal number of sensors active and put others into low-powered sleep mode, while the active sensors can maintain a connected covet set for the target area. The problem of computing such minimum working sensor set is NP-hard. In this paper, a centralized Voronoi tessellation (CVT) based approximate algorithm is proposed to construct the near optimal cover set. When sensor's communication radius is at least twice of its sensing radius, the covet set is connected at the same time; In case of sensor's communication radius is smaller than twice of its sensing radius, a connection scheme is proposed to calculate the assistant nodes needed for constructing the connectivity of the cover set. Finally, the performance of the proposed algorithm is evaluated through theoretical analysis and extensive numerical experiments. Experimental results show that the proposed algorithm outperforms the greedy algorithm in terms of the runtime and the size of the constructed connected cover set.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.30 no.7B
• /
• pp.417-426
• /
• 2005

Routing and wavelength assignment problem is one of the most important issues in optical transport networks based on wavelength division multiplexing(WDM) technique. In this paper, we propose a novel approach using path conflict graphs and an algorithm for finding all edge disjoint paths. And then we compare the performance of the proposed algorithm with that of bounded greedy approach for EDP(BGAforEDP). The proposed one outperforms up to about 20$\%$ in the fixed traditional topology(NSFNET) and about 32$\%$ in random topologies over the BGA for EDP algorithm.

• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.7 no.3
• /
• pp.95-99
• /
• 2008

This paper proposes a new minimal resource high-level synthesis algorithm for low power design automation. The proposed algorithm executes an efficient approach to minimize the power consumption of the functional units in a circuit during the high level synthesis. In this paper, we visit all control steps one by one to reduce the switching activity in CDFG. The register sharing algorithm determines the minimum register after the life time analysis of all variable. According to property of input signal for functional unit, the proposed method visits all control step one by one and determines the resource allocation with minimal power consumption at each control step in a greedy fashion. The effect of the proposed algorithm has been proved through various filter benchmark to adopt a new scheduling and allocation algorithm considering the low rover.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.3 no.6
• /
• pp.612-627
• /
• 2009

The high-level contribution of this paper is to illustrate the effectiveness of using graph theory tree traversal algorithms (pre-order, in-order and post-order traversals) to generate the chain of sensor nodes in the classical Power Efficient-Gathering in Sensor Information Systems (PEGASIS) data aggregation protocol for wireless sensor networks. We first construct an undirected minimum-weight spanning tree (ud-MST) on a complete sensor network graph, wherein the weight of each edge is the Euclidean distance between the constituent nodes of the edge. A Breadth-First-Search of the ud-MST, starting with the node located closest to the center of the network, is now conducted to iteratively construct a rooted directed minimum-weight spanning tree (rd-MST). The three tree traversal algorithms are then executed on the rd-MST and the node sequence resulting from each of the traversals is used as the chain of nodes for the PEGASIS protocol. Simulation studies on PEGASIS conducted for both TDMA and CDMA systems illustrate that using the chain of nodes generated from the tree traversal algorithms, the node lifetime can improve as large as by 19%-30% and at the same time, the energy loss per node can be 19%-35% lower than that obtained with the currently used distance-based greedy heuristic.

• 제어로봇시스템학회:학술대회논문집
• /
• 2004.08a
• /
• pp.710-715
• /
• 2004

The design problem of manufacturing system is addressed, adopting the closed queueing network model with multiple loops and re-entrant flows. The entire design problem is divided into two hierarchical sub-problems of (1) determining the station configuration and (2) optimizing the lot constitution; then they are tackled by neighbor search algorithm (NSA) and greedy mean value analysis (GMVA), respectively. Unlike the conventional MVA concerning multi-class closed queueing networks, the GMVA doesn't stick to a fixed lot proportion; rather it tries to find the optimal balance. The NSA, on the other hand, improves the object function value by altering the station configuration successively with its superior neighbor. The moderate time complexity, presented in big-${o}$ notation, enables us to apply the method even to the large-size practical cases, and the CPU time of an enlarged problem can be approximated by the same equation. The validity of our analytic approach is backed up by simulation studies with a widespread simulation package.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.12 no.4
• /
• pp.1415-1435
• /
• 2018

Resource allocation in device-to-device (D2D) aided cellular systems, in which the proximity users are allowed to communicate directly with each other without relying on the intervention of base stations (BSs), is investigated in this paper. A new uplink resource allocation policy is proposed by exploiting the relationship between D2D-access probability and channel gain among variant devices, such as cellular user equipments (CUEs), D2D user equipments (DUEs) and BSs, etc., under the constraints of their minimum signal to interference-plus-noise ratio (SINR) requirements. Furthermore, the proposed resource-allocation problem can be formulated as the cost function of "maximizing the number of simultaneously activated D2D pairs subject to the SINR constraints at both CUEs and DUEs". Numerical results relying on system-level simulations show that the proposed scheme is capable of substantially improving both the D2D-access probability and the network throughput without sacrificing the performance of conventional CUEs.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.12 no.4
• /
• pp.1693-1713
• /
• 2018

An important technique used by database administrators (DBAs) is to improve performance in decision-support workloads associated with a Star schema is multi-level partitioning. Queries will then benefit from performance improvements via partition elimination, due to constraints on queries expressed on the dimension tables. As the task of multi-level partitioning can be overwhelming for a DBA we are proposing a wizard that facilitates the task by calculating a partitioning scheme for a particular workload. The system resides completely on a client and interacts with the costing estimation subsystem of the query optimizer via an API over the network, thereby eliminating any need to make changes to the optimizer. In addition, since only cost estimates are needed the wizard overhead is very low. By using a greedy algorithm for search space enumeration over the query predicates in the workload the wizard is efficient with worst-case polynomial complexity. The technology proposed can be applied to any clustering or partitioning scheme in any database management system that provides an interface to the query optimizer. Applied to the Teradata database the technology provides recommendations that outperform a human expert's solution as measured by the total execution time of the workload. We also demonstrate the scalability of our approach when the fact table (and workload) size increases.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.12 no.6
• /
• pp.2806-2825
• /
• 2018

Unsupervised learning has shown good performance on image, video and audio classification tasks, and much progress has been made so far. It studies how systems can learn to represent particular input patterns in a way that reflects the statistical structure of the overall collection of input patterns. Many promising deep learning systems are commonly trained by the greedy layerwise unsupervised learning manner. The performance of these deep learning architectures benefits from the unsupervised learning ability to disentangling the abstractions and picking out the useful features. However, the existing unsupervised learning algorithms are often difficult to train partly because of the requirement of extensive hyperparameters. The tuning of these hyperparameters is a laborious task that requires expert knowledge, rules of thumb or extensive search. In this paper, we propose a simple and effective unsupervised feature learning algorithm for image classification, which exploits an explicit optimizing way for population and lifetime sparsity. Firstly, a sparse target matrix is built by the competitive rules. Then, the sparse features are optimized by means of minimizing the Euclidean norm ($L_2$) error between the sparse target and the competitive layer outputs. Finally, a classifier is trained using the obtained sparse features. Experimental results show that the proposed method achieves good performance for image classification, and provides discriminative features that generalize well.