• The Journal of the Acoustical Society of Korea
• /
• v.37 no.6
• /
• pp.483-488
• /
• 2018

Far field input utterance is one of the major causes of performance degradation of speaker verification systems. In this study, we used teacher student learning framework to compensate for the performance degradation caused by far field utterances. Teacher student learning refers to training the student deep neural network in possible performance degradation condition using the teacher deep neural network trained without such condition. In this study, we use the teacher network trained with near distance utterances to train the student network with far distance utterances. However, through experiments, it was found that performance of near distance utterances were deteriorated. To avoid such phenomenon, we proposed techniques that use trained teacher network as initialization of student network and training the student network using both near and far field utterances. Experiments were conducted using deep neural networks that input raw waveforms of 4-channel utterances recorded in both near and far distance. Results show the equal error rate of near and far-field utterances respectively, 2.55 % / 2.8 % without teacher student learning, 9.75 % / 1.8 % for conventional teacher student learning, and 2.5 % / 2.7 % with proposed techniques.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.51 no.11
• /
• pp.137-145
• /
• 2014

This paper suggests an LDA-based long distance face recognition algorithm for intelligent surveillance system. The existing face recognition algorithm using single distance face image as training images caused a problem that face recognition rate is decreased with increasing distance. The face recognition algorithm using face images by actual distance as training images showed good performance. However, this also causes user inconvenience as it requires the user to move one to five meters in person to acquire face images for initial user registration. In this paper, proposed method is used for training images by using single distance face image to automatically create face images by various distances. The test result showed that the proposed face recognition technique generated better performance by average 16.3% in short distance and 18.0% in long distance than the technique using the existing single distance face image as training. When it was compared with the technique that used face images by distance as training, the performance fell 4.3% on average at a close distance and remained the same at a long distance.

• Journal of KIISE
• /
• v.42 no.6
• /
• pp.817-822
• /
• 2015

Relation extraction is an important information extraction technique that can be widely used in areas such as question-answering and knowledge population. Previous studies on relation extraction have been based on supervised machine learning models that need a large amount of training data manually annotated with relation categories. Recently, to reduce the manual annotation efforts for constructing training data, distant supervision methods have been proposed. However, these methods suffer from a drawback: it is difficult to use these methods for collecting negative training data that are necessary for resolving classification problems. To overcome this drawback, we propose a one-class classification model that can be trained without using negative data. The proposed model determines whether an input data item is included in an inner category by using a similarity measure based on lexical information and syntactic patterns in a vector space. In the experiments conducted in this study, the proposed model showed higher performance (an F1-score of 0.6509 and an accuracy of 0.6833) than a representative one-class classification model, one-class SVM(Support Vector Machine).

• Journal of KIISE
• /
• v.43 no.2
• /
• pp.269-274
• /
• 2016

Named entity recognition is a process which extracts named entities in sentences and determines categories of the named entities. Previous studies on named entity recognition have primarily been used for supervised learning. For supervised learning, a large training corpus manually annotated with named entity categories is needed, and it is a time-consuming and labor-intensive job to manually construct a large training corpus. We propose a semi-supervised learning method to minimize the cost needed for training corpus construction and to rapidly enhance the performance of named entity recognition. The proposed method uses distance supervision for the construction of the initial training corpus. It can then effectively remove noise sentences in the initial training corpus through the use of an active bagging method, an ensemble method of bagging and active learning. In the experiments, the proposed method improved the F1-score of named entity recognition from 67.36% to 76.42% after active bagging for 15 times.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.52 no.12
• /
• pp.174-182
• /
• 2015

In this paper, we developed the small size RGB sensor that recognizes a long distance using a low-cost color sensor. Light receiving portion of the sensor was used as a camera lens for far distance recognition, and illuminating unit was increased the strength of the light by using a high-power white LED and a lens mounted on the reflector. RGB color recognition algorithm consists of the learning process and the realtime recognition process. We obtain a normalized RGB color reference data in the learning process using the specimens painted with target colors, and classifies the three colors using the Mahalanobis distance in recognition process. We apply the developed the RGB color recognition sensor to a prototype of the part classification system and evaluate the performance of its.

• Annual Conference on Human and Language Technology
• /
• 2015.10a
• /
• pp.202-205
• /
• 2015

개체명 인식은 문서에서 개체명을 추출하고 추출된 개체명의 범주를 결정하는 작업이다. 기존의 지도 학습 기법을 이용한 개체명 인식을 위해서는 개체명 범주가 수동으로 부착된 대용량의 학습 말뭉치가 필요하며, 대용량의 말뭉치 구축은 인력과 시간이 많이 들어가는 일이다. 본 논문에서는 학습 말뭉치 구축비용을 최소화하고 초기 학습 말뭉치의 노이즈를 제거하여 말뭉치의 품질을 향상시키는 방법을 제안한다. 제안 방법은 반자동 개체명 사전 구축 방법으로 구축한 개체명 사전과 원거리 감독법을 사용하여 초기 개체명 범주 부착 말뭉치를 구축한다. 그리고 휴리스틱을 이용하여 초기 말뭉치의 노이즈를 제거하여 학습 말뭉치의 품질을 향상시키고 개체명 인식의 성능을 향상시킨다. 실험 결과 휴리스틱 적용을 통해 개체명 인식의 F1-점수를 67.36%에서 73.17%로 향상시켰다.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.24 no.6
• /
• pp.1139-1145
• /
• 2014

User recognition technology, which identifies or verifies a certain individual is absolutely essential under robotic environments for intelligent services. The conventional face recognition algorithm using single distance face image as training images has a problem that face recognition rate decreases as distance increases. The face recognition algorithm using face images by actual distance as training images shows good performance but this has a problem that it requires user cooperation. This paper proposes the LDA-based long distance face recognition method which uses multiple distance face images from a zoom camera for training face images. The proposed face recognition technique generated better performance by average 7.8% than the technique using the existing single distance face image as training. Compared with the technique that used face images by distance as training, the performance fell average 8.0%. However, the proposed method has a strength that it spends less time and requires less cooperation to users when taking face images.

• Proceedings of the Korean Information Science Society Conference
• /
• 2003.10a
• /
• pp.580-582
• /
• 2003

정보화 시대의 요구에 대한 교육적 대응은 학습자 중심의 교육이며, 정보통신기술을 기반으로 한 원격교육이다. 특히 인터넷에서 가장 활발하게 사용되고 있는 웹은 그 특성상 구성주의 학습 원리를 실현하는데 적절한 환경을 제공하고 있으며, 교육 시스템의 변화를 가져오고 있다. WBI(Web Based Instruction)는 웹을 매체로 활용하여 원거리에 있는 학습자를 교육시키는 형태로서 교수자와 학습자간 상호작용을 가능하게 하고 다양한 형태의 학습자료를 제공하며, 공간적 제약을 극복할 수 있다는 장점이 있다. 본 논문에서는 웹에 기반한 교육적 활용의 한 모델로서 학습자의 자기 주도적이고 적극적인 학습을 촉진하기 위한 자기 주도적 학습이론과 학습자의 개인차를 고려한 수준별 학습이론을 적용하여 효율적인 웹 기반 학습이 가능하도록 하였으며 이러한 학습이론을 웹에 기반한 정보처리 기능사 필기시험에 적용함으로써 실업계 고등학교 교육현장의 비효율성과 기존의 문제은행 시스템의 문제점을 보완하여 시험준비에 보다 효율적으로 대비하도록 설계 및 구현하였다.

• Journal of Digital Contents Society
• /
• v.16 no.5
• /
• pp.681-689
• /
• 2015

This study aims to survey the effectiveness of the international distance learning between Korean and Japanese university students. This research is based on NWCCDL (Namseoul-Waseda Cross-Cultural Distance Learning) project in the spring semester, 2015. This project is the cross-cultural distance learning project between N University in Korea and W University in Japan, and the most important thing of this project is that this project is manipulated through the utilization of ICT. This research consists with two parts: the first is to introduce the NWCCDL project; the participants' information, and the contents and procedure of the on-line chatting program, and BBS(Bulletin Board System) activities. The second is to review on the students' satisfaction for the project and the utilization of ICT in English language education context. The analytic results of the questionnaire includes the students' satisfaction on this project and their reflection on the effectiveness of using ICT in English language classroom of Korea. The results prove that the most of the students are satisfied with the NWCCDL Project in the spring semester, 2015 and the most of the students agree with the fact that the utilization of ICT is very effective in English language education of Korea.

• Proceedings of the Korean Information Science Society Conference
• /
• 2000.10b
• /
• pp.568-570
• /
• 2000

컴퓨터를 이용한 인터넷의 급속한 발전에 따라 이를 교육에 응용하려는 노력은 최근 몇 년 동안 지속적으로 진행되어 왔다. 이는 인터넷 이용인구의 급속한 증가로 인하여 웹 상에서의 가상교육 및 원격교육에 관한 필요성이 증대되었고, 이를 실용화하려는 노력은 여러 가지 방법을 통하여 논의되어 왔다. 따라서, 웹 코스웨어는 인터넷을 통한 원거리 학습자들에 대하여 학습욕구를 충족시키기 위한 목적으로 구현 및 개발되었다. 그러나, 이를 이용하려면 사용자의 시스템이 정보를 제공하는 서버와의 정보교환을 위하여 온라인 상태를 유지하였으며, 학습자가 서버에서 갱신되는 학습정보를 받아 보려면 온라인 상에서 학습자의 시스템이 서버로 접속이 이루어져야만 확인이 가능하였다. 본 논문은 이에 관하여 XML(eXtensible Markup Language)을 기반으로 한 웹 캐스팅 기법으로 웹 코스웨어를 구현하였으며 이를 위하여 CDF(Channel Definition Format) 형식을 이용하였다. 즉, 채널을 용하여 학습자가 자율적으로 채널을 선택하면, 학습자가 원하는 정보를 자동적으로 갱신되도록 하였다. 또한, 기존의 웹 코스웨어는 정보를 제공하는 서버의 학습내용이 변화될 때다 일일이 확인을 하여야 하는 반면 본 논문에서는 갱신되어진 내용을 학습자에게 자동적으로 업데이트되도록 하였다.