• 한국컴퓨터정보학회논문지
• /
• 제27권10호
• /
• pp.29-41
• /
• 2022

최근 딥 러닝 기술의 발전으로 방대한 텍스트 데이터를 사전에 학습한 우수한 성능의 거대한 모델들이 다양하게 고안되었다. 하지만 이러한 모델을 실제 서비스나 제품에 적용하기 위해서는 빠른 추론 속도와 적은 연산량이 요구되고 있으며, 이에 모델 경량화 기술에 대한 관심이 높아지고 있다. 대표적인 모델 경량화 기술인 지식증류는 교사 모델이 이미 학습한 지식을 상대적으로 작은 크기의 학생 모델에 전이시키는 방법으로 다방면에 활용 가능하여 주목받고 있지만, 당장 주어진 문제의 해결에 필요한 지식만을 배우고 동일한 관점에서만 반복적인 학습이 이루어지기 때문에 기존에 접해본 문제와 유사성이 낮은 문제에 대해서는 해결이 어렵다는 한계를 갖는다. 이에 본 연구에서는 궁극적으로 해결하고자 하는 과업에 필요한 지식이 아닌, 보다 상위 개념의 지식을 학습한 교사 모델을 통해 지식을 증류하는 이질적 지식증류 방법을 제안한다. 또한, 사이킷런 라이브러리에 내장된 20 Newsgroups의 약 18,000개 문서에 대한 분류 실험을 통해, 제안 방법론에 따른 이질적 지식증류가 기존의 일반적인 지식증류에 비해 학습 효율성과 정확도의 모든 측면에서 우수한 성능을 보임을 확인하였다.

• 한국멀티미디어학회논문지
• /
• 제24권7호
• /
• pp.868-879
• /
• 2021

PCB (Printed Circuit Board) inspection using a deep learning model requires a large amount of data and storage. When the amount of stored data increases, problems such as learning time and insufficient storage space occur. In this study, the existing object detection model is changed to a continual learning model to enable the recognition and classification of PCB components that are constantly increasing. By changing the structure of the object detection model to a knowledge distillation model, we propose a method that allows knowledge distillation of information on existing classified parts while simultaneously learning information on new components. In classification scenario, the transfer learning model result is 75.9%, and the continual learning model proposed in this study shows 90.7%.

• Communications for Statistical Applications and Methods
• /
• 제30권5호
• /
• pp.501-516
• /
• 2023

Knowledge distillation (KD) refers to extracting knowledge from a large and complex model (teacher) and transferring it to a relatively small model (student). This can be done by training the teacher model to obtain the activation function values of the hidden or the output layers and then retraining the student model using the same training data with the obtained values. Recently, relational KD (RKD) has been proposed to extract knowledge about relative differences in training data. This method improved the performance of the student model compared to conventional KDs. In this paper, we propose a new method for RKD by introducing a new loss function for RKD. The proposed loss function is defined using the area difference between the teacher model and the student model in a specific hidden layer, and it is shown that the model can be successfully compressed, and the generalization performance of the model can be improved. We demonstrate that the accuracy of the model applying the method proposed in the study of model compression of audio data is up to 1.8% higher than that of the existing method. For the study of model generalization, we demonstrate that the model has up to 0.5% better performance in accuracy when introducing the RKD method to self-KD using image data.

• 지식경영연구
• /
• 제3권1호
• /
• pp.1-18
• /
• 2002

Many organizations implement knowledge management initiates by developing knowledge management systems. This study aims at investigating knowledge sharing behaviors in a knowledge management system and identifying factors affecting such behaviors. To do this, we defined knowledge sharing behaviors in a knowledge management system as registration and view of knowledge at a system. Based on this definition, we established a research model by identifying seven factors affecting both behaviors as independent variables: Learning orientation, Pressure to share knowledge, Top management support, Reward for knowledge sharing, Level of experience in IT, System quality, and Knowledge quality. The 14 hypotheses derived from a research model were tested by a correlation analysis and a multiple regression analysis with data from 165 respondents of the 21 organizations which implemented knowledge management initiatives. As results, both of knowledge registration and knowledge review were strongly affected by the learning-orientedness of an organization. Finally, we discussed results and limitations of this study.

• 로봇학회논문지
• /
• 제2권1호
• /
• pp.80-92
• /
• 2007

For the pervasive computing in ubiquitous environment, it is very important to manage the context model to provide pertinent context knowledge to context-aware applications. The context model should be able to support efficiently the context knowledge reusing and sharing as well as reasoning. Previous works focus mainly on the context knowledge representation scheme for reasoning. This paper proposes a context knowledge modeling scheme especially for reusing and sharing. This scheme provides well-established principles and guides for 1) context knowledge modularization and hierarchization, and for 2) context knowledge identification and organization. Once the context models are built according to the scheme, the structure of the context model and the meanings of the context knowledge elements become clear and consistent, so that context-aware applications can share and reuse the context knowledge in easy and error-reduced manner. This paper also discusses the implementation of a context model and an application for Presentation Helper scenario running on a software middleware system (CAMUS) for ubiquitous service robots which is being developed by ETRI Korea.

• 지식경영연구
• /
• 제6권1호
• /
• pp.1-17
• /
• 2005

Both knowledge management and BSC practices are in its inception in the tourism field, and the study explores the relationships between knowledge and BSC perspectives using correlations and path analysis. The purpose of this study was to explore the casual relationships among tourist destination knowledge and BSC performance perspectives. The study model added knowledge perspectives to the usual BSC model (with customer, growth and learning, internal process, and financial perspectives), in addition to the modification of the financial perspectives to economic, socio-cultual and physical impact. The study found out that knowledge supports learning and growth perspectives greatly, and less extent, internal processes and customer perspectives. Learning and growth affects internal processes and customer perspectives. Internal process supports customer perspectives. Both customer and, less extent, internal process have impact on the final results. The final analysis results were different by destination type. The study concludes with recommendations for further studies including rational BSC model for tourist destinations and relationships between BSC performance indicators.

• Journal of Information Technology Applications and Management
• /
• 제17권3호
• /
• pp.191-211
• /
• 2010

Existing models of knowledge processing do not feature a systemic meaning of knowledge management and ignore the role of leadership and social energy in the knowledge processing system (KPS). This conceptual paper introduces the Leadership Invigorating Flows of Energies, (LIFE) Model as an attempt to remedy that situation and provide a more useful description of the KPS. The LIFE Model highlights the role of emergent leadership and flows of social energies as forces encouraging knowledge creation and dynamic diffusion within an organization through the Knowledge Processing Cycle in eight activities interacting with its social knowledge base in a self-organizing system.

• 한국IT서비스학회지
• /
• 제16권3호
• /
• pp.47-65
• /
• 2017

The reference guide for the cost of establishing the digital documents has been used as a basis for establishing the budget for the construction of the knowledge information resource. However, due to the abolition of the nominal unit price notification in terms of IT projects, it is necessary to conduct research to convert the standard of the current labor force grade standard to the national incompetency standard (NCS). In this study, we investigate and revise the system and contents of the current knowledge information cost estimation model. In specific, i) we conducted gap analysis of cost estimation model and existing NCS model. As the contents conforming for the construction of the knowledge information resource were not adoptable, we define the description of the construction of the knowledge information resource and to identify the core elements of NCS prior to the improvement of the cost model. ⅱ) then we proposed improve the cost model considering integration with newly proposed NCS model for knowledge information construction job. In order to ensure the validity of the application of NCS development and cost estimation model, the experts reviewed relevant contents and made plans for improvement by using experts from supply and demand groups of various fields of national knowledge informatization projects.

• 한국건설관리학회논문집
• /
• 제3권1호
• /
• pp.65-73
• /
• 2002

건설산업의 특성상 엔지니어링 등 소프트웨어 부문은 지식이 곧 기업의 경쟁력을 좌우하기 때문에 철저한 보안 강화로 지식의 소재조차 불분명한 상황이며, 건설현장의 지식은 프로젝트의 종료와 함께 사장되어 가는 현실에 직면해 있다. 이 연구의 목적은 지식경영을 추진하고 있는 기존 업체의 지식관리 학습모델을 고찰하여 이를 근간으로 새로운 모델을 개발하고 건설업체에 적용하는 시나리오를 설정하는 것이다 이를 토대로 사장되어 가는 건설현장 지식을 유통시켜, 건설산업의 전반적인 수준 향상을 도모하고자 한다. 이에 지식관리와 관련한 선행 연구를 분석하여, EIP, EDMS, 지식 및 실패사례관리, CoP, e-Learning으로 구성되는 건설분야의 지식관리 학습모델을 개발하였다.

• 지식경영연구
• /
• 제11권2호
• /
• pp.1-16
• /
• 2010

Despite the well accepted necessity and importance of knowledge management, cases of successful knowledge management are not easy to find. With a case study of a Korean organization, this research introduces an integrated knowledge management model, in which knowledge and work processes are combined to improved knowledge performance. Knowledge-work process integration facilitates knowledge asset creation, transfer, and utilization, through which inefficiencies in knowledge utilization may be removed and knowledge contribution to business performance may be enhanced. Knowledge-work process integration may further be aligned with innovation processes, enabling systematic and continuous knowledge based process innovation.