경기침체로 경매 하루 1200건씩 나와

오흥선,O, Heung-Seon 대한설비공학회 2004 설비저널 Vol.33 No.11

儀式으로 본 通度寺 空間動線 조사 연구

오흥선,이영한,박기홍,문종만,노정선 서울産業大學校 1999 논문집 Vol.49 No.2

This study deals mainly with relationships between the spatial composition and the Buddhist rituals of monks at the Tongdosa temple. The results of study are summarized as followed. Many of rituals are performed in the Sangrohjun(上爐殿) and, in particular, handled in the Sulbujun(說法殿), but not performed in the Jungrohjun(上爐殿) and Harohjun(下爐殿) except for the time of Buddhist mass held in the every morning and night. The main activities of all rituals were based on the Kumkang Stupa in the past and such a tendency continues in the present time. The characteristics of temple space reflected by the movement of monks and believers who take advantage of the Kumkang Stupa's Structure in the significant rituals such as the entrance of priesthood, reception of Buddhist confirmation, Buddha's Birthday and establishment of Tongdosa Temple.

자동 트렌드 탐지를 위한 속성의 정의 및 트렌드 순위 결정 방법

오흥선(Heung-Seon Oh),최윤정(Yoonjung Choi),신욱현(Wookhyun Shin),정윤재(Yoonjae Jeong),맹성현(Sung-Hyon Myaeng) 한국정보과학회 2009 정보과학회논문지 : 소프트웨어 및 응용 Vol.36 No.3

특허, 뉴스, 블로그와 같이 시간 정보가 있는 문서들로부터의 자동적인 트렌드 분석(trend analysis)은 토픽탐지 및 추적 기술(TDT: Topic Detection and Tracking)과 더불어 중요한 연구 분야로 대두되고 있다. 과거 연구들은 대부분 트렌드과 관련된 단어의 출현 빈도 정보를 이용하여 주어진 개념의 중요도를 측정하고 이 개념의 시간에 따른 트렌드 라인을 보여주는 것에 초점을 맞췄다. 신출 트렌드(emerging trend)를 탐지하기 위해서는 주어진 개념의 출현 빈도수 변화와 같은 간단한 방법이나 학습 데이타와 비교하여 차이를 탐지하여 제시하는 방법이 사용되었다. 그러나 여러 트렌드 중에서 특징적인 트렌드를 찾아서 사용자에게 제공하기 위해서는 트렌드 순위 결정 함수가 필요하다. 본 논문은 트렌드의 다양한 측면을 정량화하기 위하여 출현 빈도로 구성된 트렌드 곡선으로부터 네 가지 속성 (변동성, 지속성, 안정성, 누적량) 을 정의하고 이를 활용한 트렌드 순위 결정 방법을 제안한다. 일련의 실험을 통하여 각 속성의 유용성을 검증하고 속성들의 조합이 순위 결정에 어떤 영향을 미치는지 분석하였다. 실험결과로부터 네 가지 속성을 모두 조합할 경우 특징적인 트렌드 탐지에 더욱 기여하는 것을 알 수 있다. With advances in topic detection and tracking(TDT), automatic trend analysis from a collection of time-stamped documents, like patents, news papers, and blog pages, is a challenging research problem. Past research in this area has mainly focused on showing a trend line over time of a given concept by measuring the strength of trend-associated term frequency information. For detection of emerging trends, either a simple criterion such as frequency change was used, or an overall comparison was made against a training data. We note that in order to show most salient trends detected among many possibilities, it is critical to devise a ranking function. To this end, we define four properties(change, persistency, stability and volume) of trend lines drawn from frequency information, to quantify various aspects of trends, and propose a method by which trend lines can be ranked. The properties are examined individually and in combination in a series of experiments for their validity using the ranking algorithm. The results show that a judicious combination of the four properties is a better indicator for salient trends than any single criterion used in the past for ranking or detecting emerging trends.

역인덱스 기반 상향식 군집화 기법을 이용한 대규모 학술 핵심어 분석

오흥선(Heung-Seon Oh),정유철(Yuchul Jung) 한국산학기술학회 2018 한국산학기술학회논문지 Vol.19 No.11

특허(patent), 학술 논문(scholarly paper)과 연구 보고서(research report)와 같은 디지털 문서(digital document)에는 주제(topic)를 요약하는 저자 키워드(author keyword)가 있다. 서로 다른 문서가 동일한 키워드를 공유하고 있다면 두 문서가 동일한 주제의 내용을 기술하고 있을 가능성이 매우 높다. 문서 군집화(document clustering)는 비슷한 주제를 가지는 문서들을 비지도 학습 방법(unsupervised learning)을 이용하여 같은 군집으로 그룹(group)화 하는 것이다. 문서 군집화는 다양한 분석에 이용되지만 대용량의 문서 데이터에 적용하기 위해서는 많은 계산량이 필요함으로 쉽지 않다. 이러한 경우, 문서의 내용을 이용하는 것보다 문서의 키워드를 이용하여 군집화하면 더욱 효율적으로 대용량의 데이터를 연결할 수 있다. 기존의 상향식 군집화 방법(bottom-up hierarchical clustering)은 대용량의 키워드 군집화(keyword clustering)를 수행하는데 있어서 많은 시간이 필요하다는 문제점이 있다. 본 논문에서는 정보검색(information retrieval)에서 널리 사용되는 역인덱스(inverted-index) 구조를 상향식 군집화에 적용한 효율적인 군집화 방법을 제안하고, 제안 방법을 대용량의 키워드 데이터에 적용하였으며, 그 결과를 분석하였다. Digital documents such as patents, scholarly papers and research reports have author keywords which summarize the topics of documents. Different documents are likely to describe the same topic if they share the same keywords. Document clustering aims at clustering documents to similar topics with an unsupervised learning method. However, it is difficult to apply to a large amount of documents event though the document clustering is utilized to in various data analysis due to computational complexity. In this case, we can cluster and connect massive documents using keywords efficiently. Existing bottom-up hierarchical clustering requires huge computation and time complexity for clustering a large number of keywords. This paper proposes an inverted index based bottom-up clustering for keywords and analyzes the results of clustering with massive keywords extracted from scholarly papers and research reports.

CNN Architecture Predicting Movie Rating from Audience’s Reviews Written in Korean

김형찬,오흥선,김덕수 한국정보처리학회 2020 정보처리학회논문지. 컴퓨터 및 통신시스템 Vol.9 No.1

In this paper, we present a movie rating prediction architecture based on a convolutional neural network (CNN). Our prediction architecture extends TextCNN, a popular CNN-based architecture for sentence classification, in three aspects. First, character embeddings are utilized to cover many variants of words since reviews are short and not well-written linguistically. Second, the attention mechanism (i.e., squeeze-and-excitation) is adopted to focus on important features. Third, a scoring function is proposed to convert the output of an activation function to a review score in a certain range (1-10). We evaluated our prediction architecture on a movie review dataset and achieved a low MSE (e.g., 3.3841) compared with an existing method. It showed the superiority of our movie rating prediction architecture.

CNN-Based Novelty Detection with Effectively Incorporating Document-Level Information

조성웅,오흥선,임상훈,김선호 한국정보처리학회 2020 정보처리학회논문지. 컴퓨터 및 통신시스템 Vol.9 No.10

With a large number of documents appearing on the web, document-level novelty detection has become important since it can reduce the efforts of finding novel documents by discarding documents sharing redundant information already seen. A recent work proposed a convolutional neural network (CNN)-based novelty detection model with significant performance improvements. We observed that it has a restriction of using document-level information in determining novelty but assumed that the document-level information is more important. As a solution, this paper proposed two methods of effectively incorporating document-level information using a CNN-based novelty detection model. Our methods focus on constructing a feature vector of a target document to be classified by extracting relative information between the target document and source documents given as evidence. A series of experiments showed the superiority of our methods on a standard benchmark collection, TAP-DLND 1.0.

실시간 스트리밍을 위한 딥러닝 기반의 얼굴마스킹 시스템

손성빈,정준욱,오흥선,정유철 제어·로봇·시스템학회 2020 제어·로봇·시스템학회 논문지 Vol.26 No.6

With a rapid increase in the use of personal broadcasting platforms (e.g., YouTube, afreecaTV, and Twitch), infringement of portrait rights has become a serious social problem as most people are unwilling to participate in the broadcasting and are exposed indiscriminately. To alleviate this problem, this paper proposes DeepMask, a real-time face masking system integrating face recognition and face masking technologies using advanced deep neural networks. Our system supports easy, selective masking of people via face registration. Thus, it can reduce the infringement of portrait rights. .

이미지 캡셔닝을 위한 스타일이 반영된 데이터 생성 기술

손성빈,김경민,오흥선 제어·로봇·시스템학회 2023 제어·로봇·시스템학회 논문지 Vol.29 No.12

. Recently, with the extensive use of chatbots and language generation models, including ChatGPT, in various applications, there has been a growing interest in multi-modal research that simultaneously processes natural language and images. Image captioning, as a prominent multi-modal field of study, has gained attention for its potential to enhance the efficiency and productivity of companies, particularly in customized image search and management. However, image captioning faces challenges due to limited downstream data and out-of-distribution (OOD) issues. This paper introduces a novel technology designed to generate data that closely resembles downstream data by leveraging pre-trained language and image generation models trained on extensive datasets. The proposed method addresses the challenges of limited downstream data and OOD issues, ultimately achieving better generalization. Furthermore, the proposed method has yielded significant results in the NICE competition, proving its exceptional performance and efficiency.

독립모델을 이용한 딥 러닝 기반의 MLCC 적층 얼라인먼트 자동 검사 시스템

손성빈,김현재,정준욱,오흥선 제어·로봇·시스템학회 2022 제어·로봇·시스템학회 논문지 Vol.28 No.10

The multi-layer ceramic capacitor (MLCC) staking inspection process is an essential step when determining whether a dielectric is well laminated However, conventional inspection processes are performed by inspectors, who measure and inspect margins manually. Therefore, there are deviations according to the subjectivity of the examiner, and accurate inspection may not be performed due to noises generated during shooting. In a recent study, an automated system using a single model based on deep learning was proposed, although it experienced difficulties in generating accurate segmentation masks. In this paper, we propose a system based on an independent model for fine-grained segmentation. The proposed system exhibited a 2% performance improvement compared to a single model and achieved a final accuracy of 98% using algorithms such as post-processing refinement and edge dielectric removal. .

딥러닝을 이용한 실시간 게임 승률 예측

이선훈 ( Seon Hoon Lee ),오흥선 ( Heung-seon Oh ) 한국정보처리학회 2019 한국정보처리학회 학술대회논문집 Vol.26 No.1

일반적으로 멀티 온라인 배틀 게임은 게임의 참가자들이 팀을 이루어 전략을 짜고 협력하여 주어진 목적을 성취하면 승리한다. 게임에서는 승리를 판가름 할 수 있는 다양한 요소(e.g. 골드, 아이템, 캐릭터의 레벨 등)들이 있다. 본 논문에서는 게임 플레이 중에 다양한 요소를 분석하여 실시간으로 승률을 예측할 수 있는 딥러닝 기반의 모델을 제안하고 이를 리그오브레전드 게임에 적용하여 그 결과를 분석하였다.