Vol. 47, No. 6, Jun. 2020
Digital Library


Model-Based Reinforcement Learning with Discriminative Loss

Guang Jin, Yohwan Noh, DoHoon Lee

http://doi.org/10.5626/JOK.2020.47.6.547

Reinforcement learning is a framework for training the agent to make a good sequence of decisions through interacting with a complex environment. Although reinforcement learning has shown promising results in many tasks, sample efficiency still remains a major challenge for its real world application. We propose a novel model-based reinforcement learning framework that incorporates the discriminative loss function, in which models are trained to discriminate one action from another. The encoder pre-trained in this framework shows the feature alignment property, which aligns with the policy gradient method. The proposed method showed better sample efficiency than conventional model-based reinforcement learning approaches in the Atari game environment. In the early stage of the training, the proposed method surpassed the baseline by a large margin.

Pattern Extraction from Lifelog Based on Semantic Network Structure Using Petri-Net

Tae-Young Kim, Sung-Bae Cho

http://doi.org/10.5626/JOK.2020.47.6.553

Recently, with the spread of smart devices, the user’s lifelog data is automatically stored through various types of sensors. But the lifelog collected from smart devices records heterogeneous information from different sensors. In addition, since the user"s life patterns are determined by different judgment cycles, it is difficult to express them in a simple rule-based system. Therefore, in order to extract and provide useful life patterns for users from the lifelog, it is necessary to express the relationship of numerous dynamic elements. In this paper, we propose a method to automatically extract user life patterns using Petri-nets from the lifelog represented by the semantic network. Petri-net reduces the uncertainty in smart device sensor data and increases the diversity of life patterns. The proposed life pattern extraction method is structured by the semantic network to represent the semantic relationship of heterogeneously collected user lifelog. Also, the Petri-net graph automatically determines the lifelog and then extracts individual sleep and eating patterns.

Approach for Managing Multiple Class Membership in Knowledge Graph Completion Using Bi-LSTM

Jae-Seung Roh, Batselem Jagvaral, Wan-Gon Lee, Young-Tack Park

http://doi.org/10.5626/JOK.2020.47.6.559

Knowledge graphs that represent real world information in a structured way are widely used in areas, such as Web browsing and recommendation systems. But there is a problem of missing links between entities in knowledge graphs. To resolve this issue, various studies using embedding techniques or deep learning have been proposed. Especially, the recent study combining CNN and Bidirectional-LSTM has shown high performance compared to previous studies. However, in the previous study, if multiple class types are defined for single entity, the amount of training data exponentially increases with the training time. Also, if class type information for an entity is not defined, training data for that entity cannot be generated. Thus, to enable the generation of training data for such entities and manage multiple class membership in knowledge graph completion, we propose two approaches using pre-trained embedding vectors of knowledge graph and the concept of vector addition. To evaluate the performance of the methods proposed in this paper, we conducted comparative experiments with the existing knowledge completion studies on NELL-995 and FB15K-237 datasets, and obtained MAP 1.6%p and MRR 1.5%p higher than that of the previous studies.

ChannelAug: A New Approach to Data Augmentation for Improving Image Classification Performance in Uncertain Environments

Hyeok Yoon, Soohan Kang, Ji-Hyeong Han

http://doi.org/10.5626/JOK.2020.47.6.568

We propose a new data augmentation method that works by separating the RGB channels of an image to improve image classification ability in uncertain environments. Many data augmentation methods, using technique such as flipping and cropping, have been used to improve the image classification ability of models. while these data augmentation methods have been effective in improving image classification, they have unperformed in uncertain conditions. To solve this problem, we propose the ChannelSplit that separates and reassembles the RGB channels of an image, along with the Mix ChannelSplit, that adopts the concept of MixUp[1,2] to express more diversity. In this paper, the proposed ChannelSplit and Mix ChannelSplit are called ChannelAug because they only utilize channels and do not perform any other image operations. Also, we compare ChannelAug to other image augmentation methods to prove it enhances robustness and uncertainty measures on image classification.

KorQuAD 2.0: Korean QA Dataset for Web Document Machine Comprehension

Youngmin Kim, Seungyoung Lim, Hyunjeong Lee, Soyoon Park, Myungji Kim

http://doi.org/10.5626/JOK.2020.47.6.577

KorQuAD 2.0 is a Korean question and answering dataset consisting of a total of 100,000+ pairs. There are three major differences from KorQuAD 1.0, which is the standard Korean Q & A data. The first is that a given document is a whole Wikipedia page, not just one or two paragraphs. Second, because the document also contains tables and lists, it is necessary to understand the document structured with HTML tags. Finally, the answer can be a long text covering not only word or phrase units, but paragraphs, tables, and lists. As a baseline model, BERT Multilingual is used, released by Google as an open source. It shows 46.0% F1 score, a very low score compared to 85.7% of the human F1 score. It indicates that this data is a challenging task. Additionally, we increased the performance by no-answer data augmentation. Through the distribution of this data, we intend to extend the limit of MRC that was limited to plain text to real world tasks of various lengths and formats.

Defining Chunks and Chunking using Its Corpus and Bi-LSTM/CRFs in Korean

Young Namgoong, Chang-Hyun Kim, Min-ah Cheon, Ho-min Park, Ho Yoon, Min-seok Choi, Jae-kyun Kim, Jae-Hoon Kim

http://doi.org/10.5626/JOK.2020.47.6.587

There are several notorious problems in Korean dependency parsing: the head position problem and the constituent unit problem. Such problems can be somewhat resolved by chunking. Chunking seeks to locate and classify constituents referred to as chunks into predefined categories. So far, several studies in Korean have been conducted without a clear definition of chunks partially. Thus, we define chunks in Korean thoroughly and build a chunk-tagged corpus based on the definition as well as propose a Bi-LSTM/CRF chunking model using the corpus. Through experiments, we have shown that the proposed model achieved a F1-score of 98.54% and can be used for practical applications. We analyzed performance variations according to word embedding and so fastText showed the best performance. Error analysis was performed so that it could be used to improve the proposed model in the near future.

Unified Methodology of Multiple POS Taggers for Large-scale Korean Linguistic GS Set Construction

Tae-Young Kim, Pum-Mo Ryu, Hansaem Kim, Hyo-Jung Oh

http://doi.org/10.5626/JOK.2020.47.6.596

In recent years, there has been national support for constructing, sharing, and spreading a large-scale Korean linguistic GS set for Korean information processing. As part of the corpus construction project, this study proposes the methodology for constructing the Korean linguistic GS set using various Korean language analysis modules developed in Korea. To build a large-scale training set, we referred to automatic tagged candidate answers from the N-modules. We then minimized manual effort by classifying the error types from the candidate responses and semi- automatically correcting the major error types. In this study, we normalized results of the morphological analysis and constructed a large-scale Korean linguistic GS set based on the unified format U-POS. As a result of this study, 348,229 sentences, a total of 9,455,930 words, were constructed as the Korean linguistic GS set. This can be practically applied later as a basic training resource for Korean information processing.

Disassortative Network Distribution Techniques Using Hub Grouping Based On Local Differential Privacy

Yongjun Kim, Seog Park

http://doi.org/10.5626/JOK.2020.47.6.603

With the development of the wireless Internet and popularization of smartphones, many people are using social network services that connect with others in online. Personal data generated by social network services have high value, but comprise sensitive personal information that could potentially result in serious privacy breaches. The existing studies have presented techniques for generating synthetic data similar to the original network data, or anonymous user information. However, the existing techniques have inherent weaknesses in privacy and data utility because such techniques have not considered the characteristics of network graphs formed by relationships with users. In this paper, we propose the privacy-protected social network data distribution techniques by applying local differential privacy techniques that reflect the characteristics on the social network graph. Through experiments with real data, we have shown that the proposed techniques perform better than the existing differentially private social network data distribution techniques.

QoE-aware Quality Adaptation Scheme Based on Bandwidth Utilization for VBR Video Streaming

Jeongho Kang, Minsu Kim, Kwangsue Chung

http://doi.org/10.5626/JOK.2020.47.6.612

In recent years, as the demand for video streaming has increased due to the development of the network, HTTP adaptive streaming has attracted increased attention. HTTP adaptive streaming can guarantee QoE (Quality of Experience) because it adaptively determines the quality to provide based on network conditions. However, existing quality adaptation techniques do not consider VBR (Variable Bit Rate) characteristics when streaming video, which causes problems such as interruption of playback and degradation of average quality. In this paper, we propose a QoE-aware quality adaptation method based on bandwidth utilization for VBR video streaming. The proposed scheme determines the quality to provide using the actual size of the video segment, the quality of information requested in the past, and the buffer occupancy. As a result of the experiment, we confirmed that the proposed scheme improves QoE thanks to fewer quality changes, higher average quality and without generating playback interruptions when compared with existing schemes.

The Design of a Multi-Function Radar Simulator for the Identification of Friend or Foe(IFF) in the Mode-5 Product Improvement Program

Younghwan Jeong, Chansu Kim, Jungin Oh, Wonsik Lee, Sounghyouk Wi

http://doi.org/10.5626/JOK.2020.47.6.622

Identification of Friend or Foe(IFF) is a function of military surveillance system used to identify whether the monitored object is an ally or an enemy, these system are installed in fighters, ships, and interceptors among others. The United States plans to suspend the Mode-4 identification system and apply Mode-5 from July 2020 forward. The transition to Mode-5 is inevitable as it ensures the interoperability of peer identification systems used in the Republic of Korea’s military operations with the US as well as other NATO member states. If the IFF function found in control centers and multi-function radars is changed in regional air defense weapons systems, revalidation of these weapons systems is required to ensure stability and correct function. Therefore, a multi-function radar simulator is vital use in interface verification, unit tests, and integrated tests before evaluation of the new systems can be completed. This paper presents the design of a simulator for mode-5 performance testing and improvement.


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