Vol. 45, No. 7,
Jul. 2018
Digital Library
An Efficient SLC-buffer Management Scheme for TLC NAND Flash-based Storage
Kirock Kwon, Dong Hyun Kang, Young Ik Eom
http://doi.org/10.5626/JOK.2018.45.7.611
In recent years, almost all consumer devices have adopted NAND flash storage as their main storage, and their performance and capacity requirements are getting higher. To meet these requirements, many researchers have focused on combined SLC-TLC storage consisting of high-speed SLC and high-density TLC. In this paper, we redesign the internal structure of the combined SLC-TLC storage to efficiently manage the SLC region inside the storage and propose a scheme that improves the performance of the storage by employing the I/O characteristics of file system journaling. We implemented our scheme on the real storage platform, the OpenSSD jasmine board, and compared it with the conventional techniques. Our evaluation results show that our technique improves the storage performance by up to 65%, compared with the conventional techniques.
Experiments on Upscaling Feasibility of GPU-Based Postprocessing Effects
http://doi.org/10.5626/JOK.2018.45.7.618
The cost of the pixel-based post-processing commonly used in real-time rendering increases rapidly depending on the resolution. This paper describes the experiments and discusses effectiveness of techniques of upscaling after post-processing at low resolution to ensure the actual spatial performance of post-processing effects at high resolution. The experiment first looks at the differences in the quality of the upgraded results at lower resolutions, after classifying typical post processing effects based on GPU. We also perform user experiments to find the visible differences between these upscalings, and analyze the relationship between the results of user testing and the quality. Although these comparisons show that the effects associated with the isotropic blur can be used, but the effects associated with the anisotropic and those without blur can not be effectively used. Especially, it has been shown that the application of anti-aliasing is effective in these effects. Based on these conclusions, we discuss the use of appropriate resolution of post-processing at ultra-high resolution.
RANSAC-based Time Synchronization over Wireless Networks
Junyoup Hwang, Daehyeon Wi, Jungjin Lee
http://doi.org/10.5626/JOK.2018.45.7.626
This paper describes a method to improve the quality of time synchronization for synchronous video streaming over a wireless network using a Simple Network Time Protocol (SNTP) with the RANdom SAmple Consensus (RANSAC) Algorithm. Time synchronization is mostly accomplished by using the NTP server. Speed and accuracy are important in synchronizing the video play module; however, NTP is limited because too much time is spent to operate the module. It also requires a complex structure and algorithms for customization. And SNTP, which is easy to implement, is also inappropriate for the wireless network because there is no filtering process for network failure. Our method refined those timestamps of SNTP using RANSAC, so that it achieves a stable synchronization, with 5 milliseconds of accuracy, which takes less than 5 seconds. This method enables high- quality time synchronization with a simple implementation for any kind of system, and will be useful for IoT devices.
An Effective Comparative Framework for Cross-Project Defect Prediction Based on the Feature Selection Technique
http://doi.org/10.5626/JOK.2018.45.7.635
Software defect prediction (SDP) can help optimally allocate software testing resources on fault-prone modules. Typically, local data within a company are used to build classifiers. Unlike such Within-Project Defect Prediction (WPDP), there may exist some cases, e.g., pilot projects, without any collected data from historical projects. Cross-project defect prediction (CPDP) using data from other projects can be employed in such cases. The defect prediction performance may be degraded in the presence of irrelevant or redundant information. To address this issue, various feature selection techniques have been suggested. Until now, there has been no research on identifying effective feature selection techniques for CPDP. We present a comparative framework using feature selection to produce a high performance for CPDP. We compare eight existing feature selection techniques, for three CPDP and one WPDP model, based on feature subset evaluators and feature ranking methods. After the features are chosen that perform the best, classifiers are built, tested, and evaluated using the statistical significance and effect size tests. Hybrid Instance Selection using Nearest-Neighbor (HISNN) is better than the other CPDP models and comparable to the WPDP model. Results from the comparison show that a different distribution, class imbalance and feature selection should be considered to obtain a high performance CPDP model.
Improving Recurrent Neural Network based Recommendations by Utilizing Embedding Matrix
Myung Ha Kwon, Sung Eon Kong, Yong Suk Choi
http://doi.org/10.5626/JOK.2018.45.7.659
Recurrent neural networks(RNNs) have recently been successfully applied to recommendation tasks. RNNs were adopted by session-based recommendation, which recommends items by the records only within a session, and a movie recommendation that recommends movies to the users by analyzing the consumption records collected through multiple accesses to the websites. The new approaches showed improvements over traditional approaches for both tasks where only implicit feedback such as clicks or purchase records are available. In this work, we propose the application of weight-tying to improve the existing movie recommendation model based on RNNs. We also perform experiments with an incremental recommendation method to more precisely evaluate the performance of recommendation models.
Using SysML for the Hazard Analysis Process at Concept Phase of Safety Critical System
http://doi.org/10.5626/JOK.2018.45.7.667
Today, the accident of safety critical system may result in catastrophic harm to people and environments. Therefore, activities designed to ensure safety, such as identifying the hazard and mitigating risks to prevent those accidents, need to be enforced in system development, especially at the concept phase. PHL and PHA can be hired to identify preliminary hazards and to do Failure Modes and Effects Analysis (FMEA), Fault Tree Analysis (FTA) to assess the causes and impacts of those hazards. For developers and safety experts to work together in the lifecycle, those safety activities need to be integrated into the modelling of system development. The common modelling language, SysML, has been recognized as a solution for this purpose and several types of research on this topic have been published. However, not much effort has been made to apply requirements diagram to acquire the visibility and traceability of safety requirements derived from those hazards identified by analyzing SysML diagrams at concept phase, with related artifacts such as implementation and verification. In this paper, we propose a hazard-identification process in which behavior and structure diagrams of SysML could be systematically analyzed. Then the safety requirements were derived to remove or mitigate the identified hazard. To demonstrate the capability of the proposed process, an example of applying it to an elevator system is presented.
Activity Detection in Untrimmed Videos with Semantic Features and Temporal Region Proposals
http://doi.org/10.5626/JOK.2018.45.7.678
In this paper, we propose a deep neural network model that effectively detects human activities in untrimmed videos. While temporal visual features extracted over several successive image frames in a video, it helps to recognize a dynamic activity itself; spatial visual features extracted from each frame help to find objects associated with the activity. To detect activities precisely in a video, therefore, both temporal and spatial visual features should be considered together. In addition to these visual features, semantic features describing video contents in high-level concepts may also help to improve video-activity detection. To localize the activity region accurately, as well as to classify an activity correctly in an untrimmed video, it is required to design a mechanism for temporal region proposal. The activity-detection model proposed in this work learns both visual and semantic features of the given video, with deep convolutional neural networks. Moreover, by using recurrent neural networks, the model effectively proposes temporal activity regions and classifies activities in the video. Experiments with large-scale benchmark datasets such as ActivityNet and THUMOS, showed the high performance of our activity-detection model.
Combinations of Text Preprocessing and Word Embedding Suitable for Neural Network Models for Document Classification
http://doi.org/10.5626/JOK.2018.45.7.690
Neural networks with word embedding have recently used for document classification. Researchers concentrate on designing new architecture or optimizing model parameters to increase performance. However, most recent studies have overlooked text preprocessing and word embedding, in that the description of text preprocessing used is insufficient, and a certain pretrained word embedding model is mostly used without any plausible reasons. Our paper shows that finding a suitable combination of text preprocessing and word embedding can be one of the important factors required to enhance the performance. We conducted experiments on AG’s News dataset to compare those possible combinations, and zero/random padding, and presence or absence of fine-tuning. We used pretrained word embedding models such as skip-gram, GloVe, and fastText. For diversity, we also use an average of multiple pretrained embeddings (Average), randomly initialized embedding (Random), task data-trained skip-gram (AGNews-Skip). In addition, we used three advanced neural networks for the sake of generality. Experimental results based on OOV (Out Of Vocabulary) word statistics suggest the necessity of those comparisons and a suitable combination of text preprocessing and word embedding.
Infinite Latent Topic Models for Document Analysis
http://doi.org/10.5626/JOK.2018.45.7.701
Since the concept of the topic is highly abstract, the characterization of the topics of a text is not clearly defined. Depending on the problem’s context or needs, various levels of detail may be provided, which could make it difficult to automatically analyze documents. This paper presents infinite topic extensions to the well-known model of Latent Dirichlet Allocation (LDA) i.e., the infinite Latent Dirichlet Topic model and the infinite Latent Markov Topic model. The first model simply relaxes the constraint of fixed known number of topics in LDA using the method of the Dirichlet process. The second model further extends it by including Markov dynamics that captures the sequential evolution of topics in a text. Both models are theoretically rigorous and structurally flexible, as well as being capable of capturing document organizations at a desired level of topics. A set of experiments show interesting results and a more intuitive topic characterization and local stationarity properties than related models with Gibbs sampling and variational inferences.
A Differentially Private Query Processing Mechanism using a Batch Strategy within a Limited Privacy Budget
Minsuc Kang, Kangsoo Jung, Seog Park
http://doi.org/10.5626/JOK.2018.45.7.708
A differential privacy has the advantage of being able to protect information regardless of the attacker’s prior knowledge. However, it has a disadvantage in that each query consumes privacy budget. The larger the privacy budget applied to the query, the more accurate are the query results. However it increases the privacy budget consumption and creates a limitation in the query processing limitation. On the other hand, if the privacy budget allocated to each query is too small, the noise becomes too much. This causes the query result to become inaccurate, and this, in turn causes the data utility to deteriorate. In this paper, we propose a batch strategy that reorders differentially private query processing in interactive environment. The proposed technique uses less privacy budget while it guarantees the data utility.
CEP Rule Distribution Algorithm for In-network Processing in an IoT Network Environment
http://doi.org/10.5626/JOK.2018.45.7.722
As the number of IoT devices increases, data coming from devices are also increasing exponentially. The data generated from devices are stored and managed through a system structure using the database. However, to manage the surging data, the existing database is limited in terms of maintenance costs and in real time. Too overcome these limitations, Complex Event Processing (CEP), which processes data as much as possible within the network, has emerged, and data processing is being carried out using this strategy. In this paper, we propose a CEP Rule distribution algorithm which can reduce server burden and guarantee network performance through distribution of the CEP Rule in an IoT environment. To prove this, we perform a small experiment using open source, such as the OpenWSN and TelosB node, and verify the mitigation of server load and the performance of data processing according to the algorithm.
SDN based Mobility Management in IoT Networks
BoYeong Mun, SungChol Cho, Shimin SUN, Jin Xinan Shu, Cheongbin Kim, JunHyuk Kim, Sunyoung Han
http://doi.org/10.5626/JOK.2018.45.7.733
IoT environments are constrained by various factors, including object address, battery, network bandwidth, and computing power. We propose a system to manage mobility by applying an SDN (Software Defined Networking) centralized management method to effectively solve the problems caused by moving IoT terminals. The SDN used in this paper is programmable, which facilitates its scalability as well as its ability to manage various IoT terminals. in this paper, we propose a mobility architecture using SDN for IoT, and design as well as implement a system based on it.
Web Application Attack Detection Scheme Using Convolutional Neural Networks
Yeongung Seo, Myungjin Kim, Seungyoung Park, Seokwoo Lee
http://doi.org/10.5626/JOK.2018.45.7.744
Because rates of web application attacks are rapidly increasing, web application attack detection schemes using machine learning have recently become of interest. Existing schemes, however, require the selection of a suitable set of features representing the characteristics of expected attacks, and this set of features needs to be adjusted every time a new type of attack is discovered. In this paper, we propose a web application attack detection scheme employing a convolutional neural network (CNN) without the need to select any features in advance. Specifically, the CNN is trained in a supervised manner with images transformed from hexadecimally converted characters in HTTP traffic, without any restriction in the input characters used. Our experimental results show that the proposed scheme improves detection error rate performance by up to 84.4% over existing schemes.
Search
Journal of KIISE
- ISSN : 2383-630X(Print)
- ISSN : 2383-6296(Electronic)
- KCI Accredited Journal
Editorial Office
- Tel. +82-2-588-9240
- Fax. +82-2-521-1352
- E-mail. chwoo@kiise.or.kr