http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
German Credit Risk 데이터를 사용한 머신러닝 모델의 단계별 요소들이 AI 모델의 성능에 미치는 영향 분석
박필원(Pill-Won Park) 한국정보기술학회 2022 한국정보기술학회논문지 Vol.20 No.11
Machine learning is used in various fields, and various algorithms have been developed according to the type and purpose of data. The performance of the machine learning model is affected by the step-by-step setting even if the same algorithm is used, and research on this is needed. However, studies on the effects of specific procedures or specific parameters on the model have been conducted, but studies that comprehensively analyze them have been insufficient. In this paper, after summarizing the processing steps required to develop the machine learning model, the effect of each step on the performance of the machine learning model was analyzed. Processing steps were divided into steps of data purification, algorithm selection, hyper-parameter adjustment, and verification ratio adjustment, which were measured using Kaggles German credit risk data and machine learning automation tools.
소나 시스템을 이용한 해저 물체에 대한 AI모델의 탐지성능 분석
박필원(Pill-Won,Park),고대식(DAE-SIK,KO) 한국정보기술학회 2022 Proceedings of KIIT Conference Vol.2022 No.12
SONAR(Sound Navigation and Ranging)는 해양 정보 수집의 기초 도구이며, 해양 자산 및 환경 조사, 표적 및 물체 인식 등에 사용되고 있다. 하지만 소나로 얻은 데이터의 분석은 사용자의 능력에 의존하는 경우가 많으며, 이러한 상황을 타개하기 위하여 소나 데이터를 AI 머신러닝 기능을 통해 분석하는 것이 바람직하다고 판단하였다. 본 논문에서는 소나의 데이터를 다양한 머신러닝 알고리즘과 하이퍼 파라미터 조정을 통해 각 단계들이 머신러닝 모델의 성능에 미치는 영향을 분석하였다. 그 결과 알고리즘의 변경에 따라 1%~46%, 하이퍼 파라미터 조정을 통해 7%~29%의 성능 향상을 확인하였다. SONAR (Sound Navigation and Ranging) is a basic tool for collecting marine information, and is used for marine asset and environmental research, target and object recognition, and the like. However, the analysis of data obtained by sonar often depends on the users ability, and in order to overcome this situation, it was decided that it is desirable to analyze sonar data through AI machine learning function. In this paper, we analyzed the effect of each stage on the performance of the machine learning model through various machine learning algorithms and hyperparameter adjustments on the sonar data. As a result, it was confirmed that the performance improved by 1%~46% according to the change of the algorithm, and 7%~29% through the hyperparameter adjustment.
챗봇의 의도 예문 자동 입력을 위한 Text-CNN 기반 의도 분류 방법
박필원(Pill-Won Park) 한국정보기술학회 2020 한국정보기술학회논문지 Vol.18 No.1
In this paper, we propose how to automatically categorize and generate examples of given intents using Text-CNN in order to increase the inference rate of the existing chatbot framework. The Intent Classification System uses Text-CNN to learn data consisting of word vectors and position vectors for each prepared sentence through the preprocessing process. The proposed Text-CNN structure has a construction layer, a max pooling layer, and a fully connected soft max layer as its output. In addition, dropout is applied to perform regularization. For the experiment, a total of 9,000 sentences were collected using webscraping. An experiment showed that the accuracy obtained from Text-CNN’s learning of kitchen intents was about 94%. The rest of the sentences, not labeled with the model produced by Text-CNN, were grouped and 63 cases of cooking were extracted in total and the sentences were input in Chatbot.
나노/마이크로 디그리를 제공하기 위한 네트워크 기반 교육 플랫폼 설계
박필원 ( Pill-won Park ) 한국정보처리학회 2021 한국정보처리학회 학술대회논문집 Vol.28 No.2
최근 대학교는 코로나로 인하여 교육에 언택트를 접목시킬 것과, 학생들에게 필요한 전문지식을 단기간에 효율적으로 습득시킬 것을 요구받고 있다. 이를 위해 학교들은 온라인 교육 플랫폼을 구성함과 동시에, 고효율 교육방법으로서 모듈형 교육과정을 개발하고 있다. 하지만 모듈형 교육과정은 기존 교육과정과 다른 부분이 있어 기존에 활용하던 온라인 교육 플랫폼으로는 모듈형 교육과정을 온전히 제공하기 힘들다. 따라서 본 논문에서는 다양한 국내외 온라인 교육 플랫폼을 조사하여 교육에 필요한 기능들을 정리하였다. 또한 다수의 동영상, 퀴즈, 실습, 과제, 그리고 프로젝트로 구성된 모듈형 교육과정을 가정하고 이를 활용하기 위한 기능들을 도출하였다. 이를 기반으로 모듈형 교육과정용 플랫폼의 기능별 구성, 구성요소별 연결 방식, 활용 시나리오를 제안하고, 온라인 기반 모듈형 교육과정 플랫폼을 보다 효율적으로 활용할 수 있을 것으로 생각되며, 향후 모듈형 교육과정을 제공하기 위한 기반을 구성할 수 있을 것으로 생각된다.
이완재(Wan-Jae Lee),박필원(Pill-Won Park) 한국정보기술학회 2019 한국정보기술학회논문지 Vol.17 No.12
In this paper, we designed a system that provides information for safe driving of the vehicle, which enables the driver to operate safely, using vehicle speed data generated during the operation of the vehicle. The configuration of the system consists of a terminal for data collection, a smartphone application with a gateway function for transmitting information collected from a vehicle to a server, and a server that implements an algorithm that can use the collected data to provide information for driver safety. The implementation of the algorithm for safe driving was embodied using the drivers information, information of the road, and vehicle speed received from the terminal. By experimenting with the vehicle speed data collected according to the location of the road and age of the driver, the proposed method in this paper is able to warn the dangerous driving if the driver is outrunning the speed limit or driving over the average speed of age.
당뇨병성 척수병 ( Diabetic Myelopathy ) 에서의 척수액의 변화에 대한 연구
박완양 ( Wan Yang Park ),기춘석 ( Choon Suhk Kee ),박필원 ( Pill Won Park ),조성경 ( Seong Keong Cho ),최영길 ( Yong Kil Choi ) 대한내과학회 1971 대한내과학회지 Vol.14 No.1
It has been reported that a significant portion of patients with diabetic neuropathy may Lave changes in spinal fluid indicating myelopathy. Myelopathy due to diabetes mellitus is characterized by asymmetrical polyneuropathy and frequent)y by motor involve
장면분석법에 참고태그를 활용한 근로자 위치확인 시스템의 성능
김보연(Bo-Yeon Kim),박필원(Pill-Won Park) 한국정보기술학회 2019 한국정보기술학회논문지 Vol.17 No.12
The technology for identifying the location of objects is very important for the realization of the ubiquitous environment in various fields, including logistics, manufacturing, and disaster prevention activities. This study proposed a technique for estimating the location of workers by applying reference tags and positional correction algorithms to scene analysis methods. The positioning correction algorithm applied to this study is a method that calibrates the received radio signal strength of a measuring tag with the radio signal strength of the distance difference in order to improve the accuracy of workers positioning in the working environment. The average error was measured at ±0.75m as a result of our experiment using an active RFID tag and position reading system as well as setting the distance between the reference tag for error correction and the measurement tag within 6m. The proposed technique confirmed that the location error could be improved by 12-15% compared to the existing positioning technique that uses only radio wave signals.