Digital Library[ Search Result ]
Semi-Supervised Object Detection for Small Imbalanced Drama Dataset
http://doi.org/10.5626/JOK.2024.51.11.978
Images of the theme of a drama are typically zoomed-in mainly to people. As a result, people-oriented images are predominant in drama data, and class imbalance naturally occurs. This paper addresses the issue of class imbalance in drama data for object detection tasks and proposes various sampling methods to tackle this challenge within the framework of semi-supervised learning. Experimental evaluations demonstrated that the suggested semi-supervised learning approach with specialized sampling methods outperformed traditional supervised and semi-supervised methods. This study underscores the significance of selecting appropriate training data and sampling methods to optimize object detection performance in specialized datasets with unique characteristics.
Improved Open-Domain Conversation Generative Model via Denoising Training of Guide Responses
Bitna Keum, Hongjin Kim, Jinxia Huang, Ohwoog Kwon, Harksoo Kim
http://doi.org/10.5626/JOK.2023.50.10.851
In recent open-domain conversation research, research is actively conducted to combine the strengths of retrieval models and generative models while overcoming their respective weaknesses. However, there is a problem where the generative model either disregards the retrieved response or copies the retrieved response as it is to generate a response. In this paper, we propose a method of mitigating the aforementioned problems. To alleviate the former problem, we filter the retrieved responses and use the gold response together. To address the latter problem, we perform noising on the gold response and the retrieved responses. The generative model enhances the ability to generate responses via denoising training. The effectiveness of our proposed method is verified through human and automatic evaluation.
Algorithm for Detecting Double-Spending in Blockchain
Minho Kim, Sujin Kim, Hoon Choi
http://doi.org/10.5626/JOK.2018.45.8.848
The blockchain is a key technology of the Bitcoin, which is widely used as an electronic cash system. In the Bitcoin, one digital currency is valid for only one transaction. It is called double-spending, a type of illegal transaction, if two or more transactions are made by using the same digital currency. When the blockchain is forked, the blockchain specification assumes that the longer blockchain may be valid, but the blockchain containing double-spending may become longer than the blockchain containing normal transactions, so comparing lengths of the chain cannot completely prevent illegal transactions. In this paper, we propose an algorithm to detect double-spending and a mechanism to notify other nodes after detection. This algorithm is implemented and verified by using the bitcoin core.
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.
A Splitting Technique of Hardware Transactional Memory in Multicore In-Memory Databases
Munhwan Kang, Hyeongjin Kim, Hyeonkuk Ma, Jaewoo Chang
http://doi.org/10.5626/JOK.2018.45.6.582
Transactional Memory has greatly changed concurrency control paradigm by replacing locks, the conventional parallel programming mechanism. Especially, HTM(Hardware Transactional Memory) is the most promising scheme that is supported by hardware. However, the existing HTM techniques have a problem that they cannot overcome the resource limitations of HTM. To solve the problem, we propose a HTM-based transaction splitting technique to support large-sized transaction processing in multicore in-memory databases. First, the proposed technique can split a transaction into nested partition blocks when the transaction fails by resource limitation. Second, the proposed technique makes use of our adaptive split algorithm that computes the optimal size of partition blocks, according to the characteristic of a workload. Finally, through our experimental performance analysis using STAMP benchmark, the proposed technique shows about 70% better performance than the existing transaction splitting technique, i.e., Part-HTM.
Activity Prediction from Sensor Data using Convolutional Neural Networks and an Efficient Compression Method
Woojeong Jin, Dongjin Choi, Youngjin Kim, U Kang
http://doi.org/10.5626/JOK.2018.45.6.564
The identification of the number of occupants and their activities using the IoT system in a building is an important task to improve the power efficiency and reduce the cost of using smart cooling/heating systems. In the actual building management system, it is possible to use equipment such as a camera to understand the current situation in the room, and to directly determine the number of occupants and their types of behavior. However, identifying the number of people and behavior types in this way is inefficient and requires a large amount of storage space for data. In this study, indoor sensor data were collected using an infrared Grid-Eye sensor and noise sensor. Based on this data, we also propose a deep learning model that captures the number of participants and behavior patterns and a deep learning model that considers the temporal characteristics of data. The proposed model identifies the number of people with an accuracy of about 95.3% and human activities with an accuracy of 90.9%. We also propose a method to reduce the storage space while minimizing the loss of accuracy using truncated SVD.
Image Quality Assessment Considering both Computing Speed and Robustness to Distortions
Suk-Won Kim, Seongwoo Hong, Jeong-Chan Jin, Young-Jin Kim
http://doi.org/10.5626/JOK.2017.44.9.992
To assess image quality accurately, an image quality assessment (IQA) metric is required to reflect the human visual system (HVS) properly. In other words, the structure, color, and contrast ratio of the image should be evaluated in consideration of various factors. In addition, as mobile embedded devices such as smartphone become popular, a fast computing speed is important. In this paper, the proposed IQA metric combines color similarity, gradient similarity, and phase similarity synergistically to satisfy the HVS and is designed by using optimized pooling and quantization for fast computation. The proposed IQA metric is compared against existing 13 methods using 4 kinds of evaluation methods. The experimental results show that the proposed IQA metric ranks the first on 3 evaluation methods and the first on the remaining method, next to VSI which is the most remarkable IQA metric. Its computing speed is on average about 20% faster than VSI’s. In addition, we find that the proposed IQA metric has a bigger amount of correlation with the HVS than existing IQA metrics.
kNN Query Processing Algorithm based on the Encrypted Index for Hiding Data Access Patterns
Hyeong-Il Kim, Hyeong-Jin Kim, Youngsung Shin, Jae-woo Chang
In outsourced databases, the cloud provides an authorized user with querying services on the outsourced database. However, sensitive data, such as financial or medical records, should be encrypted before being outsourced to the cloud. Meanwhile, k-Nearest Neighbor (kNN) query is the typical query type which is widely used in many fields and the result of the kNN query is closely related to the interest and preference of the user. Therefore, studies on secure kNN query processing algorithms that preserve both the data privacy and the query privacy have been proposed. However, existing algorithms either suffer from high computation cost or leak data access patterns because retrieved index nodes and query results are disclosed. To solve these problems, in this paper we propose a new kNN query processing algorithm on the encrypted database. Our algorithm preserves both data privacy and query privacy. It also hides data access patterns while supporting efficient query processing. To achieve this, we devise an encrypted index search scheme which can perform data filtering without revealing data access patterns. Through the performance analysis, we verify that our proposed algorithm shows better performance than the existing algorithms in terms of query processing times.
Human Visual System-Aware and Low-Power Histogram Specification and Its Automation for TFT-LCDs
Backlight has a major factor in power consumption of TFT-LCDs which are most popular in portable devices. There have been a lot of attempts to achieve power savings by backlight dimming. At the same time, the researches have shown image compensation due to decreased brightness of a displayed image. However, existing image compensation methods such as histogram equalization have some limits in completely satisfying the human visual system (HVS)-awareness. This paper proposes an enhanced dimming technique to obtain both power saving and HVS-awareness by combining pixel compensation and histogram specification for TFT-LCDs. This method executes a search algorithm and an automation algorithm employing simplified calculations for fast image processing. Experimental results showed that the proposed method achieved significant improvement in visual satisfaction per power saving over existing backlight dimming.
A Group Modeling Strategy Considering Deviation of the User’s Preference in Group Recommendation
HyungJin Kim, Young-Duk Seo, Doo-Kwon Baik
Group recommendation analyzes the characteristics and tendency of a group rather than an individual and provides relevant information for the members of the group. Existing group recommendation methods merely consider the average and frequency of a preference. However, if the users’ preferences have large deviations, it is difficult to provide satisfactory results for all users in the group, although the average and frequency values are high. To solve these problems, we propose a method that considers not only the average of a preference but also the deviation. The proposed method provides recommendations with high average values and low deviations for the preference, so it reflects the tendency of all group members better than existing group recommendation methods. Through a comparative experiment, we prove that the proposed method has better performance than existing methods, and verify that it has high performance in groups with a large number of members as well as in small groups.
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