• Electronics and Telecommunications Trends
• /
• v.34 no.4
• /
• pp.32-42
• /
• 2019

The performance of machine learning algorithms significantly depends on how a configuration of hyperparameters is identified and how a neural network architecture is designed. However, this requires expert knowledge of relevant task domains and a prohibitive computation time. To optimize these two processes using minimal effort, many studies have investigated automated machine learning in recent years. This paper reviews the conventional random, grid, and Bayesian methods for hyperparameter optimization (HPO) and addresses its recent approaches, which speeds up the identification of the best set of hyperparameters. We further investigate existing neural architecture search (NAS) techniques based on evolutionary algorithms, reinforcement learning, and gradient derivatives and analyze their theoretical characteristics and performance results. Moreover, future research directions and challenges in HPO and NAS are described.

• Journal of the Korean Operations Research and Management Science Society
• /
• v.22 no.4
• /
• pp.183-219
• /
• 1997

In this research, the intensity of solar energy, which was injected to the different angle plane every hour day by day, was technically documented and quantitatively analyzed through actual observations. In order to group every days into days with similar intensity, graph was drawn with respect to time for every dary and each area value under the curve was calculated. Then, the search for grouped days having similar intensity curve patterns was carried out. In order to maximize the efficiency of solar energy systems, the optimum incident angle of absorber plate was derived.

• Proceedings of the KSRS Conference
• /
• v.1
• /
• pp.163-165
• /
• 2006

We aim to archive all the satellite images that had been scattered into Satellite Imagery Information System with setting naming rules and metadata. More than one million of scenes were collected, rectified into error-free status with metadata . Converting various formats into HDF format after considering GEOTIFF and HDF. Intranet and Internet System had been development to allow all the images to be searched and downloaded with less effort. These system will expand the usage of meteorological satellite images for expert groups and the public outside of KMA.

• 제어로봇시스템학회:학술대회논문집
• /
• 1989.10a
• /
• pp.290-295
• /
• 1989

This paper proposes a Expert CAPP System for integrating CAD/CAM of rotational work-part by rule based approach. The CAD/CAPP integration is performed by the recognition of machined features from the 2-D CAD data (IGES) file. Selecting functions of the process planning are performed in modularized rule base by forward chaining inference, and operation sequences are determined by means of heuristic search algorithm. For CAPP/CAM integration, post-processor generates NC code from route sheet file. This system coded in OPS5 and C language on PC/AT, and EMCO CNC lathe interfaced with PC through DNC and RS-232C.

• Proceedings of the Korean Information Science Society Conference
• /
• 2000.04b
• /
• pp.298-300
• /
• 2000

전문가 검색 엔진은 전문가 시스템과 같은 목적에서 특정 전문 분야에 대한 특별한 정보를 모야 특정 정보를 검색하기 위한 엔진이다. 인터넷의 홈페이지는 서로를 연결하는데 하이퍼링크(hyperlink)를 사용하는데 이런 하이퍼링크(hyperlink)에 있는 정보를 이용하여 홈페이지와 홈페이지 사이의 연결관계를 모은 결과를 전문가 검색 엔진에서 모은 키워드와 웹 사이트를 이용하여 각 키워드 간의 관련성을 데이터 마이닝 기법을 사용해서 각 키워드나 웹 페이지간의 상관관계에 대한 개념을 추출한다. 본 논문에서는 이런 홈페이지 간의 추출을 이용한 시스템 설계와 구현 결과를 보여준다.

• Proceedings of the Korean Information Science Society Conference
• /
• 2000.04b
• /
• pp.295-297
• /
• 2000

전문가 검색 엔진은 전문가 시스템과 같은 목적에서 특정 전문 분야에 대한 특수한 정보를 수집 검색하기 위한 검색 엔진을 내용이다. 본 논문은 전문가 검색 엔진을 만드는 과정에서 초기 작업에 해당하는 웹 정보 수집에 대한 논문이다. 본 논문은 웹 페이지에서 하이퍼링크와 참조되는 웹 페이지에 대한 표면 지식을 이용하여, 홈페이지 그래프를 작성한다. 그리고 나서 홈페이지 그래프와 사전에 미리 준비된 개념 그래프를 이용하여, 웹 페이지 수집 중 특정 전문 분야에 해당하는 웹 페이지인지를 판별하여 사용자가 수집하고자 하는 분야에 대한 웹 페이지만을 수집한다. 본 논문은 이에 대한 개념, 설계 및 구현과 앞으로의 개선 상황을 제안한다.

• Proceedings of the KIEE Conference
• /
• 1998.07c
• /
• pp.1097-1099
• /
• 1998

This paper describes methods to quickly reconfigure the branch in a distribution system by using tie-line switches without branch overloading and feeder overloading, voltage in load points, and distribution system is represented by object-based structure to search efficiently.

• Journal of the Korean Society for Library and Information Science
• /
• v.52 no.3
• /
• pp.5-33
• /
• 2018

In this paper, we analyze a user's usage behavior, identify and track search intention and interest field based on the National Science and Technology Standard Classification, and use it to personalize science and technology information. In other words, we sought to satisfy both efficiency and satisfaction in searching for information that users want by improving scientific information search performance. We developed the personalization service of science and technology information and evaluated the suitability and usefulness of personalized information by comparing the search performance between expert experimental group and control group. As a result, the personalization service proposed in this study showed better search performance than comparative service and proved to provide higher usability.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.22 no.3
• /
• pp.341-346
• /
• 2012

In rehabilitation where ultrasonographic diagnosis is not popular, it could be subjective by medical expert's experience. Thus, it is necessary to develop an objective automative procedure in ultrasonic image analysis. A disadvantage of existing automative analytic procedure in musculoskeletal system is to designate an incorrect muscle area when the figure of fascia is vague. In this study, we propose a new procedure to extract more accurate muscle area in abdomen ultrasonic image for that purpose. After removing unnecessary noise from input image, we apply End-in Search algorithm to enhance the contrast between fascia and muscle area. Then after extracting initial muscle area by Up-Down search, we trace the fascia area with a mask based on morphological and directional information. By this tracing of mask movements, we can emphasize the fascia area to extract more accurate muscle area in result. This new procedure is proven to be more effective than existing methods in experiment using convex ultrasound images that are used in real world rehabilitation diagnosis.

• Journal of the Korean Society for information Management
• /
• v.23 no.2
• /
• pp.5-19
• /
• 2006

The purpose of this study is to observe and analyze information searching behaviors of various user groups in different access modes for designing user-centered query interface of content-based Music Information Retrieval System(MIRS). Two expert groups and two non-expert groups were recruited for this research. The data gathering techniques employed in this study were in-depth interviewing, participant observation, searching task experiments, think-aloud protocols, and post-search surveys. Expert users, especially majoring in music theory, preferred to input exact notes one by one using the devices such as keyboard and musical score. On the other hand, non-expert users preferred to input melodic contours by humming.