Vol. 48, No. 8,
Aug. 2021
Digital Library
Data Transfer Optimized High-level FPGA Host Programming Interface
Jongwoo Kim, Seongsoo Park, Hwansoo Han
http://doi.org/10.5626/JOK.2021.48.8.859
Along with general-purpose CPUs, hardware accelerators have been widely adopted to execute various workloads efficiently. Recently, FPGAs have emerged in the area of software-level development as high-level languages such as C/C++ support FPGA programming. OpenCL supports most heterogeneous processors in high-level programming, but different optimization techniques are required depending upon the unique architectural features in the accelerators. In particular, developing FPGA kernel programs requires more knowledge of hardware architecture than other heterogeneous processors. Due to this characteristic, optimization should be collaborated with the host program as well. In this paper, we proposed SimFL, a high-level programming interface for developing host programs to use FPGAs as accelerators. To evaluate our optimization, we used the host programs for FPGA with SimFL and verified a performance improvement of up to 44.7% by applying multi-threaded copying within SimFL.
A Digital Forensic Process for Ext4 File System in the Flash Memory of IoT Devices
Junho Jeong, Beomseok Kim, Jinsung Cho
http://doi.org/10.5626/JOK.2021.48.8.865
With the recent rapid advances in digital communication technology, the spread of IoT(Internet of Things) has accelerated and IoT devices can be utilized to investigate crimes and accidents due to the close connection between human society and IoT devices. Accordingly, with the increasing importance of digital forensics, numerous studies have been conducted. However, most digital forensics research proposed only abstract methodologies due to the various types of IoT devices. In addition, binwalk, which is actively used as a firmware analysis tool, does not adequately analyze and extract the ext4 file system. To solve these problems, this paper proposes a proper extraction and analysis method and a practical process that could extract the ext4 file system from the flash memory of IoT devices using the binwalk with the proposed method. This study also verifies the proposed process with DJI Phantom 4 Pro V2.0 drone.
Reinforcement Learning-based Traffic Signal Control under Real-World Constraints
Mingyu Pi, Hunsoon Lee, Moonyoung Chung
http://doi.org/10.5626/JOK.2021.48.8.871
Traffic signal control plays an important role in efficiently using the limited capacity of the road. Since traditional traffic signal control methods operate based on preset signals, it is difficult to cope with frequently changing traffic conditions. Recently, as reinforcement learning has attracted attention as a method for solving complex problems, studies using reinforcement learning for efficient traffic signal control are being conducted. Compared to the traditional method, it has been proved through simulation that waiting time and travel time were improved. However, since most of the studies did not reflect the limitations of the actual signal, it was designed inappropriately for practical application. In this paper, we proposed a signal control method based on reinforcement learning that could be applied to real situations by reflecting the constraints of the signal operating system that exist in reality, and that could respond to changes in traffic volume.
Query Intent Detection for Medical Advice: Training Data Construction and Intent Classification
Tae-Hoon Lee, Young-Min Kim, Eunji Jeong, Seon-Ok Na
http://doi.org/10.5626/JOK.2021.48.8.878
In most task-oriented dialogue systems, intent detection and named entity recognition need to precede. This paper deals with the query intent detection to construct a dialogue system for medical advice. We start from the appropriate intent categories for the final goal. We also describe in detail the data collection, training data construction, and the guidelines for the manual annotation. BERT-based classification model has been used for query intent detection. KorBERT, a Korean version of BERT has been also tested for detection. To verify that the DNN-based models outperform the traditional machine learning methods even for a mid-sized dataset, we also tested SVM, which produces a good result in general for such dataset. The F1 scores of SVM, BERT, and KorBERT are 69%, 78%, and 84% respectively. For future work, we will try to increase intent detection performance through dataset improvement.
Vehicle Image Data Augmentation by GAN-based Viewpoint Transformation
Hangyel Sun, Myeonghee Lee, Charmgil Hong, Injung Kim
http://doi.org/10.5626/JOK.2021.48.8.885
We introduce a novel GAN-based image synthesis method that transforms vehicle images captured from arbitrary viewpoints into images taken from a specific viewpoint. Training a vehicle image recognizer requires a large number of vehicle images taken from a specific viewpoint. However, in practice, it is difficult to collect such training data, especially for newly released vehicles. Therefore, we propose a method of augmenting vehicle image data by converting a vehicle image from an arbitrary viewpoint into an image from a specific viewpoint. The proposed method first transforms a vehicle image from an arbitrary viewpoint to an image taken from the top-front view using DRGAN, then enhances the image quality with DeblurGAN, and finally, improves the resolution using SRGAN. The experimental results demonstrated that the proposed method successfully converted an image taken within 45 degrees left and right into an image from the top-frontal view and was effective in improving the image quality and resolution.
Denoising Multivariate Time Series Modeling for Multi-step Time Series Prediction
Jungsoo Hong, Jinuk Park, Jieun Lee, Kyeonghun Kim, Seung-Kyun Hong, Sanghyun Park
http://doi.org/10.5626/JOK.2021.48.8.892
The research field of time series forecasting predicts the future time point using seasonality in time series. In the industrial environment, since decision-making through continuous perspective prediction of the future is important, multi-step time series forecasting is necessary. However, multi-step prediction is highly unstable because of its dependency on predicted value of previous time prediction result. Therefore, the traditional time series forecasting makes a statistical prediction for the single time point. To address this limitation, we propose a novel encoder-decoder based neural network named ‘DTSNet’ which predicts multi-step time points for multivariate time series. To stabilize multi-step prediction, we exploit positional encoding to enhance representation for time point and propose a novel denoising training method. Moreover, we propose dual attention to resolve long-term dependencies and modeling complex patterns in time series, and we adopt multi-head strategy at linear projection layer for variable-specific modeling. To verify the performance improvement of our approach, we compare and analyze it with baseline models, and we demonstrate the proposed methods through comparison tests, such as, component ablation study and denoising degree experiment.
Automatic Segmentation of Lung Cancer in Chest CT Images through Capsule Network-based Dual-Window Ensemble Learning
Jumin Lee, Julip Jung, Helen Hong, Bong-Seog Kim
http://doi.org/10.5626/JOK.2021.48.8.905
It is difficult to accurately segment lung cancer in chest CT images when it has an irregular shape or nearby structures have a similar intensity as lung cancer. In this study, we proposed a dual window ensemble network that uses a capsule network to learn the relationship between lung cancer and nearby structures and additionally considers the mediastinal window image with the lung window image to distinguish lung cancer from the nearby structures. First, intensity and spacing normalization was performed on the input images of the lung window setting and mediastinal window setting. Second, two types of 2D capsule network were performed with the lung and mediastinal setting images. Third, the final segmentation mask was generated by ensemble the probability maps of the lung and mediastinal window images through average voting by reflecting the weight based on the characteristics of each image. The proposed method showed a Dice similarity coefficient(DSC) of 75.98% which was 0.53% higher than the method not considering the weight of each window setting. Furthermore, segmentation accuracy was improved even when lung cancer was surrounded by nearby structures.
A Perimeter-Based IoU Loss for Efficient Bounding Box Regression in Object Detection
http://doi.org/10.5626/JOK.2021.48.8.913
In object detection, neural networks are generally trained by minimizing two types of losses simultaneously, namely classification loss and regression loss for bounding boxes. However, the regression loss often fails to achieve its ultimate goal, that is, it often obtains a predicted bounding box that maximally intersects with its target box. This is due to the fact that the regression loss is not highly correlated with the IoU, which actually measures how much the bounding box and its target box overlap with each other. Although several penalty terms have been invented and added to the IoU loss in order to address the problem of regression losses, they still show some inefficiency particularly when penalty terms become zero by enclosing another box or overlapping with the center point before the bounding box and its target box are perfectly the same. In this paper, we propose a perimeter based IoU (PIoU) loss exploiting the perimeter differences of the minimum bounding rectangle of both a predicted box and its target box from those of two boxes themselves. In our experiments using the state-of-the-art object detection models (e.g., YOLO v3, SSD, and FCOS), we show that our PIoU loss consistently achieves better accuracy than all the other existing IoU losses.
A Selection Technique of Source Project in Heterogeneous Defect Prediction based on Correlation Coefficients
Eunseob Kim, Jongmoon Baik, Duksan Ryu
http://doi.org/10.5626/JOK.2021.48.8.920
The software defect prediction techniques try to predict defect-prone modules and ensure the quality of the developing software using previous defect data. Nowadays, heterogeneous defect prediction (HDP) techniques have been applying defect prediction techniques even when the metrics between source and target projects are different. Previous HDP techniques focused on improving prediction performance when the source and target projects were given. However in a real development environment, more than one source projects exist for one target project, thus identifying a project that is suitable for source data is challenging. This paper suggests a correlation-based selection technique for source projects in HDP. After the metric matching process, correlation coefficients are calculated for each corresponding metric, and the project with the highest score is selected for source data. The experiment shows that the performance of the proposed selection method is higher than the results of random selection, and removing projects with less than 100 instances from the source candidates improves the performance. Therefore, using the proposed selection technique could improve the prediction accuracy in HDP.
A Multi-label Classification Bot for Issue Management System in GitHub
Doje Park, Yyejin Yang, Gwang Choi, Seonah Lee, Sungwon Kang
http://doi.org/10.5626/JOK.2021.48.8.928
The GitHub platform, on which many developers develop open-source software projects, provides an issue management system. Using the system, the stakeholders can report software problems or functional requests as issues. The issue management system provides issue report forms and allows developers to create and use labels to classify issues. However, since the labeling work is manually done, it requires a lot of effort from the developers and inaccurate labeling can easily occur. In addition, it takes a lot of time for a project manager to read and give feedback on each issue. To mitigate this problem, previous studies have proposed attaching a single label to an issue automatically. However, in practice, there are a number of issue reports that need multiple labels to be attached. Therefore, in this study, we proposed a multi-labeling bot that automatically attaches multiple labels to an issue report in order to reduce the effort required by a project manager to read issue reports and give feedback in GitHub. The multi-label classification of our bot showed F-score ranging from 0.54 to 0.78.
Deep Neural Networks and End-to-End Learning for Audio Compression
Daniela N. Rim, Inseon Jang, Heeyoul Choi
http://doi.org/10.5626/JOK.2021.48.8.940
Recent advances in end-to-end deep learning have encouraged the exploration of tasks dealing with highly structured data using unified deep network models. The fabrication and design of such models for compressing audio signals has been a challenge due to the need for discrete representations that are not easy to train with end-to-end backpropagation. In this paper, we present an end-to-end deep learning approach that combines recurrent neural networks (RNNs) within the training strategy of variational autoencoders (VAEs) with a binary representation of the latent space. We apply a reparametrization trick for the Bernoulli distribution for the discrete representations, which allows smooth backpropagation. In addition, our approach enables the separation of the encoder and decoder, which is necessary for compression tasks. To the best of our knowledge, this is the first end-to-end learning for a single audio compression model with RNNs, and our model achieves a Signal to Distortion Ratio (SDR) of 20.53dB.
Ontology Construction for Interoperability and Scalability of Human-Object Interaction Datasets
Aryoung Kim, Sangbaek Lee, Kyuchul Lee
http://doi.org/10.5626/JOK.2021.48.8.947
Human behavior can be expressed through actions and objects. In order to understand the human behavior, the interaction with objects must be considered, as well as human behavior. The Human-Object Interaction (HOI) can be expressed in rich and varied ways. As detailed representations were required, large-scale datasets were generated. However, as they were written in different file formats and expression methods, decreasing interoperability between each other made it difficult to add, modify, or delete new objects or relationships. In this paper, we constructed a HOI ontology which can be utilized in the field of HOI research. We designed ontology to express the relationship between humans and objects. We then developed a method of instance generation and extension for interoperability support between HOI datasets. The design of the ontology allowed clear identification and expression of extracted information. Moreover, the extension method provided semantic interoperability so that new objects or relationships could be continuously extended to the HOI ontology.
Community Detection Using Link Attribute-Based Classification
Jeongseon Kim, Soohwan Jeong, Sungsu Lim
http://doi.org/10.5626/JOK.2021.48.8.959
Attempts to discover knowledge through data are becoming gradually diversified to understand a fast and complex world. Graph data analysis, which models and analyzes correlated data as graphs, is drawing much attention as it is combined with the latest machine learning techniques. In this work, we propose a novel methodology for discovering graph community structures. We analyze similarity, curvature-based attributes to allow links existing inside and outside the community to have different attribute values, and exploit them to design and analyze algorithms that eliminate links that affect the community structure less to find better community structures on sparse graphs.
A Streaming Control System for Real-time Blocking of Obscene Videos in Mobile Devices
Jeongho Kang, Minsu Kim, Kwangsue Chung
http://doi.org/10.5626/JOK.2021.48.8.966
As users’ accessibility to video streaming services increases, technology for real-time blocking of obscene videos in mobile devices is drawing attention. However, a load is generated for a mobile device during the blocking process due to a low processing power. In this paper, we propose a streaming control system for real-time blocking of obscene videos in mobile devices. The proposed system can extract the frame of video and analyze the obscenity of the video through an obscenity analysis engine. In addition, the load is minimized by determining the frame extraction method in consideration of obscenity change and similarity comparison results between frames, and obscene video is blocked by performing video mosaic processing. Through the implementation results, it was confirmed that the proposed system could minimize the load generated from a mobile device and user exposure to the obscene part.
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