• Journal of the Korea Society of Computer and Information
• /
• v.25 no.2
• /
• pp.1-10
• /
• 2020

In this paper, we propose an efficient disk buffer replacement policy which improves hit ratio and reduces writing operations of flash based storages. The flash based storage has many advantages, including a small form factor, non-volatility and high reliability, but there are problems caused by own limitations, like not-in-place update, short life cycle and asymmetric I/O latencies. To redeem these problems, this paper proposes the write weighted probability of reference(2WPR) policy. 2WPR policy predicts re-referencing probability and calculates localities of each page. Furthermore, by weighting write operations to every pages, 2WPR can reduce write operations to flash based storage. In addition, we can improve the performance with higher hit ratio and reduce the number of write operations and consequently shorten the latencies of each operation. The results show that our policy provides improvements of up to 10% for the hit ratio with the reduction of up to 5% for the flash writing operation compared with other policies.

• The KIPS Transactions:PartB
• /
• v.10B no.3
• /
• pp.243-248
• /
• 2003

An incremental kernel principal component analysis (IKPCA) is proposed for the nonlinear feature extraction from the data. The problem of batch kernel principal component analysis (KPCA) is that the computation becomes prohibitive when the data set is large. Another problem is that, in order to update the eigenvectors with another data, the whole eigenspace should be recomputed. IKPCA overcomes these problems by incrementally computing eigenspace model and empirical kernel map The IKPCA is more efficient in memory requirement than a batch KPCA and can be easily improved by re-learning the data. In our experiments we show that IKPCA is comparable in performance to a batch KPCA for the feature extraction and classification problem on nonlinear data set.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.12 no.3
• /
• pp.509-516
• /
• 2008

Recently, the operator modules to control external devices are concerned about automatic management system to process continuously changed signals. These signals are the stream data of which characteristics are several numbers. a short report interval and asynchronous report time. It is necessary that the system brings about high accuracy and real time process for stream data. The typical queries of these systems consist of the current query to search the latest signal value, the snapshot query at a past time, the sliding window query from a past time to current. In this paper, we propose the efficient method to manage the above signals by using a file structured database in small-size operating systems. We also propose a query model to accommodate various queries including the sliding window query. The file database in the QNX adopts a delta version and a shared memory buffering method for the resource limit of a small storage and a low computing power.

• Phonetics and Speech Sciences
• /
• v.11 no.3
• /
• pp.39-47
• /
• 2019

We propose a method to improve the performance of spontaneous speech recognizers by extending their phone set using speech data. In the proposed method, we first extract variable-length phoneme-level segments from broadcast speech signals, and convert them to fixed-length latent vectors using an long short-term memory (LSTM) classifier. We then cluster acoustically similar latent vectors and build a new phone set by choosing the number of clusters with the lowest Davies-Bouldin index. We also update the lexicon of the speech recognizer by choosing the pronunciation sequence of each word with the highest conditional probability. In order to analyze the acoustic characteristics of the new phone set, we visualize its spectral patterns and segment duration. Through speech recognition experiments using a larger training data set than our own previous work, we confirm that the new phone set yields better performance than the conventional phoneme-based and grapheme-based units in both spontaneous speech recognition and read speech recognition.

• Journal of Broadcast Engineering
• /
• v.24 no.1
• /
• pp.105-117
• /
• 2019

Current Android smartphones include about 25 to 35 essential applications(unerasable) related to function and operation of the android smartphone itself and about 30 optional applications(removable) provided by carriers, Google and manufacturers. These applications were not able to be removed until the announcement of the smartphone applications pre-install guide from the government in January 2014, so there were memory limitations in installing new applications, causing consumer complaints by consuming data during the auto-update process of the pre-installed applications. After the announcement, we were able to delete optional applications but the complaints about the data consumption still did not disappear. Therefore, in this paper, we carried out the customer survey and analyzed the behavior information such as how carriers are operating pre-installed applications and what kind of applications customer prefers and how many applications customer wants to be pre-installed. And we proposed how to configure pre-install applications on smartphone for customer needs.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.23 no.12
• /
• pp.1638-1643
• /
• 2019

In this paper, we propose an adaptive-perturbation-aided opportunistic precoding (APOP) scheme for multiple-input multiple-output (MIMO) systems. To update a precoding matrix in MIMO systems, the proposed algorithm produces a random perturbation in each time slot. Then the additional adaptive perturbation is also applied, which depends on the reports of achievable data-rates from users. If the prior random perturbation increased the data rate, the adaptive perturbation is set to be the same as the prior random perturbation, otherwise the negative value of the prior random perturbation is applied for adaptive perturbation. Furthermore, to enhance the achievable data rates, the information on the stored precoding matrices in the memory as well as the currently generated precoding matrix is used for scheduling. Simulation results show that compared to conventional opportunistic precoding schemes, higher data rates are achieved by the proposed APOP scheme, especially when there are a relatively small number of users.

• Journal of The Korean Association of Information Education
• /
• v.9 no.3
• /
• pp.453-462
• /
• 2005

Recently, due to technical improvements of storage devices and networks, the amount of data increase rapidly. In addition, it is required to find the knowledge embedded in a data stream as fast as possible. Huge data in a data stream are created continuously and changed fast. Various algorithms for finding frequent itemsets in a data stream are actively proposed. Current researches do not offer appropriate method to find frequent itemsets in which flow of time is reflected but provide only frequent items using total aggregation values. In this paper we proposes a novel algorithm for finding the relative frequent itemsets according to the time in a data stream. We also propose the method to save frequent items and sub-frequent items in order to take limited memory into account and the method to update time variant frequent items. The performance of the proposed method is analyzed through a series of experiments. The proposed method can search both frequent itemsets and relative frequent itemsets only using the action patterns of the students at each time slot. Thus, our method can enhance the effectiveness of learning and make the best plan for individual learning.

• The KIPS Transactions:PartC
• /
• v.17C no.6
• /
• pp.441-450
• /
• 2010

Recently, a lot of interest is increased in sensor network which gathers various data through many sensor nodes deployed in wired and wireless network environment. However, because of the limitation in memory, computation, and energy of the sensor nodes, security problem is very important issue. In sensor network, not only the security problem, but also computing power should be seriously considered. In this paper, considering these characteristics, we make the sensor network consist of normal sensor nodes and clusterheaders with enough space and computing power, and propose a group key rekeying scheme adopting PCGR(Predistribution and local Collaborationbased Group Rekeying) for secure group communication. In our proposal, we enhance the security by minimizing the risk to safety of the entire network through verifying the new key value from clusterheader by sensor nodes. That is, to update the group keys, clusterheaders confirm sensor nodes through verifying the information from sensor nodes and send the new group keys back to authentic member nodes. The group keys sent back by the clusterheaders are verified again by sensor nodes. Through this mutual authentication, we can check if clusterheaders are compromised or not. Qualnet simulation result shows that our scheme not only guarantees secure group key rekeying but also decreasesstorage and communication overhead.

• Journal of Korea Multimedia Society
• /
• v.9 no.9
• /
• pp.1118-1130
• /
• 2006

Existed pattern analysis algorithms in data streams environment have researched performance improvement and effective memory usage. But when new data streams come, existed pattern analysis algorithms have to analyze patterns again and have to generate pattern tree again. This approach needs many calculations in real situation that needs real time pattern analysis. This paper proposes a method that continuously analyzes patterns of incoming data streams in real time. This method analyzes patterns fast, and thereafter obtains real time patterns by updating previously analyzed patterns. The incoming data streams are divided into several sequences based on time based window. Informations of the sequences are inputted into a hash table. When the number of the sequences are over predefined bound, patterns are analyzed from the hash table. The patterns form a pattern tree, and later created new patterns update the pattern tree. In this way, real time patterns are always maintained in the pattern tree. During pattern analysis, suffixes of both new pattern and existed pattern in the tree can be same. Then a pointer is created from the new pattern to the existed pattern. This method reduce calculation time during duplicated pattern analysis. And old patterns in the tree are deleted easily by FIFO method. The advantage of our algorithm is proved by performance comparison with existed method, MILE, in a condition that pattern is changed continuously. And we look around performance variation by changing several variable in the algorithm.

• Journal of Korea Spatial Information System Society
• /
• v.7 no.2 s.14
• /
• pp.47-56
• /
• 2005

Recently, with the rapid development of mobile computing, wireless positioning technologies, and the generalization of wireless internet, LBS (Location Based Service) which utilizes location information of moving objects is serving in many fields. In order to serve LBS efficiently, the location data server that periodically stores location data of moving objects is required. Formerly, GIS servers have been used to store location data of moving objects. However, GIS servers are not suitable to store location data of moving objects because it was designed to store static data. Therefore, in this paper, we designed and implemented an extended SLDS(Short-term Location Data Subsystem) for real-time Location Based Services. The extended SLDS is extended from the SLDS which is a subsystem of the GALIS(Gracefully Aging Location Information System) architecture that was proposed as a cluster-based distributed computing system architecture for managing location data of moving objects. The extended SLDS guarantees real-time service capabilities using the TMO(Time-triggered Message-triggered Object) programming scheme and efficiently manages large volume of location data through distributing moving object data over multiple nodes. The extended SLDS also has a little search and update overhead because of managing location data in main memory. In addition, we proved that the extended SLDS stores location data and performs load distribution more efficiently than the original SLDS through the performance evaluation.