• Annual Conference on Human and Language Technology
• /
• 2022.10a
• /
• pp.343-347
• /
• 2022

최근 자연어처리 분야에서는 BERT, RoBERTa, 그리고 BART와 같은 사전 학습된 언어 모델 (Pre-trained Language Models, PLM) 기반 미세 조정 학습을 통하여 여러 하위 과업에서 좋은 성능을 거두고 있다. 이는 사전 학습된 언어 모델 및 데이터 집합의 크기, 그리고 모델 구성의 중요성을 보여주며 대규모 사전 학습된 언어 모델이 각광받는 계기가 되었다. 하지만, 거대한 모델의 크기로 인하여 실제 산업에서 쉽게 쓰이기 힘들다는 단점이 명백히 존재함에 따라 최근 매개변수 효율적인 미세 조정 및 Few-Shot 학습 연구가 많은 주목을 받고 있다. 본 논문은 Prompt tuning, Prefix tuning와 프롬프트 기반 미세 조정 (Prompt-based fine-tuning)을 결합한 Few-Shot 학습 연구를 제안한다. 제안한 방법은 미세 조정 ←→ 사전 학습 간의 지식 격차를 줄일 뿐만 아니라 기존의 일반적인 미세 조정 기반 Few-Shot 학습 성능보다 크게 향상됨을 보인다.

• The Journal of the Korea Contents Association
• /
• v.9 no.12
• /
• pp.467-476
• /
• 2009

As web technology and educational environments are in rapid progress, web-based cooperative learning systems have gained a lot of interests. Recently some studies have attempted to combine a learning system and simple games that enable learners to actively participate and have high interests in learning. These studies are based on TGT model, a cooperative learning model using games, and mostly remain system design levels. A few implemented systems have many problems because they focus only on pure TGT model. To solve these problems, this paper builds a extended TGT model and a new web-based cooperative learning system using this new model. The extended part contains ideas such as expert learning from Jigsaw II model, improvement scores from STAD model and making game problems by learners. A system using pure TGT model and a suggested system are implemented and used by two classes of middle school students to evaluate our system. The experimental results show that our system outperforms the other system.

• Journal of Internet Computing and Services
• /
• v.22 no.4
• /
• pp.13-28
• /
• 2021

Reinforcement Learning (RL) is a machine learning algorithm that repeat the closed-loop process that agents perform actions specified by the policy, the action is evaluated with a reward function, and the policy gets updated accordingly. The key benefit of RL is the ability to optimze the policy with action evaluation. Hence, it can effectively be applied to developing advanced intelligent systems and autonomous systems. Conventional RL incoporates a single policy, a reward function, and relatively simple policy update, and hence its utilization was limited. In this paper, we propose an extended RL model that considers multiple instances of RL elements. We define a formal model of the key elements and their computing model of the extended RL. Then, we propose design methods for applying to system development. As a case stud of applying the proposed formal model and the design methods, we present the design and implementation of an advanced car navigator system that guides multiple cars to reaching their destinations efficiently.

• Annual Conference on Human and Language Technology
• /
• 2020.10a
• /
• pp.257-260
• /
• 2020

사전 학습을 기반으로 하는 BERT계열의 모델들이 다양한 언어 및 자연어 처리 태스크들에서 뛰어난 성능을 보이고 있지만, masked language model의 경우 입력 문장의 15%만 마스킹을 함으로써 학습 효율이 떨어지고 미세 조정 시 마스킹 토큰이 등장하지 않는 불일치 문제도 존재한다. 이러한 문제를 효과적으로 해결한 ELECTRA는 영어 벤치마크에서 기존의 언어모델들 보다 뛰어난 성능을 보여주었지만 한국어에 대한 관련 연구는 부족한 실정이다. 본 연구에서는 ELECTRA를 한국어 코퍼스에 대해 학습시키고, 다양한 한국어 자연어 이해 태스크들에 대해 실험을 진행한다. 실험을 통해 ELECTRA의 모델 크기별 성능 평가를 진행하였고, 여러 한국어 태스크들에 대해서 평가함으로써 ELECTRA 모델이 기존의 언어 모델들보다 좋은 성능을 보인다는 것을 입증하였다.

• The Journal of the Korea Contents Association
• /
• v.22 no.10
• /
• pp.765-777
• /
• 2022

The purpose of this study is to present a digital transformation-based learning model that can be used in universities based on learning digital transformation in order f to be competitive in a rapidly changing environment. Literature review, case study, and focus group interview were conducted and the implications for the learning model from these are as follows. Universities that stand out in related fields are actively using learning analysis to implement dashboards, develop predictive models, and support adaptive learning based on big data, They also have actively introduced advanced edutech to classes. In addition, problems and difficulties faced by other universities and K University when implementing digital transformation were also confirmed. Based on these findings, a digital transformation-based learning model of K University was developed. This model consists of four dimensions: diagnosis, recommendation, learning, and success. It allows students to proceed with learning by diagnosing and recommending various learning processes necessary for individual success, and systematically managing learning outcomes. Finally, academic and practical implications about the research results were discussed.

• Journal of the Korea Society for Simulation
• /
• v.31 no.1
• /
• pp.11-18
• /
• 2022

Recently, research to classify human activity using imagery has been actively conducted for the purpose of crime prevention and facility safety in public places and facilities. In order to improve the performance of human activity classification, most studies have applied deep learning based-transfer learning. However, despite the increase in the number of backbone network models that are the basis of deep learning as well as the diversification of architectures, research on finding a backbone network model suitable for the purpose of operation is insufficient due to the atmosphere of using a certain model. Thus, this study applies the transfer learning into recently developed deep learning backborn network models to build an intelligent system that classifies human activity using imagery. For this, 12 types of active and high-contact human activities based on sports, not basic human behaviors, were determined and 7,200 images were collected. After 20 epochs of transfer learning were equally applied to five backbone network models, we quantitatively analyzed them to find the best backbone network model for human activity classification in terms of learning process and resultant performance. As a result, XceptionNet model demonstrated 0.99 and 0.91 in training and validation accuracy, 0.96 and 0.91 in Top 2 accuracy and average precision, 1,566 sec in train process time and 260.4MB in model memory size. It was confirmed that the performance of XceptionNet was higher than that of other models.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.48 no.3
• /
• pp.36-45
• /
• 2011

An agent in collaborative e-learning has independent function for learners in any circumstance, status and task by the reasonable and general means for social learning. In order to perform it well, communication among agents requires standardized and regular information technology method. This study suggests data model as a communication tool for various agents. Therefore this study shows various agents types for collaborative learning, designation of rule for data model that enable to communicate among agents and data element of agent communication data model. A multi-agent e-learning system using like this standardized data model should able to exchange the message that is needed for communication among agents who can take charge of their independent tasks. This study should contribute to perform collaborative e-learning successfully by the application of communication data model among agents for social learning.

• Proceedings of the Korea Multimedia Society Conference
• /
• 2000.04a
• /
• pp.219-222
• /
• 2000

일반적으로 WBI가 제공하는 시간과 장소를 초월하는 가상의 공간속에서 웹 환경에서 하이퍼미디어 기술을 도입하여 학습자가 학습에 대한 선택의 자유를 가지고 자기 나름대로 학습의 상황을 수시로 점검하고, 평가할 수 방법을 가지고 학습자가 학습에 대해 책임을 가지고 학습자 자신의 인지 구조를 최대한 반영하는 개별화 교수-학습 환경을 제공하고자한다. 개발된 Cyber Class는 LCPG 모델을 기반으로 하는 학습진행 및 학습평가 알고리즘을 제공하고 학습자 개개인의 특성에 맞는 재학습 메커니즘을 제공한다.

• Annual Conference on Human and Language Technology
• /
• 2022.10a
• /
• pp.16-21
• /
• 2022

사전학습 말뭉치는 위키백과 문서 뿐만 아니라 인터넷 커뮤니티의 텍스트 데이터를 포함한다. 이는 언어적 관념 및 사회적 편향된 정보를 포함하므로 사전학습된 언어 모델과 파인튜닝한 언어 모델은 편향성을 내포한다. 이에 따라 언어 모델의 중립성을 평가할 수 있는 지표의 필요성이 대두되었으나, 아직까지 언어 인공지능 모델의 정치적 중립성에 대해 정량적으로 평가할 수 있는 척도는 존재하지 않는다. 본 연구에서는 언어 모델의 정치적 편향도를 정량적으로 평가할 수 있는 지표를 제시하고 한국어 언어 모델에 대해 평가를 수행한다. 실험 결과, 위키피디아로 학습된 언어 모델이 가장 정치 중립적인 경향성을 나타내었고, 뉴스 댓글과 소셜 리뷰 데이터로 학습된 언어 모델의 경우 정치 보수적, 그리고 뉴스 기사를 기반으로 학습된 언어 모델에서 정치 진보적인 경향성을 나타냈다. 또한, 본 논문에서 제안하는 평가 방법의 안정성 검증은 각 언어 모델의 정치적 편향 평가 결과가 일관됨을 입증한다.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.16 no.12
• /
• pp.8753-8759
• /
• 2015

Establishment of analysis procedures and validated performance measurements for generator output is required to maintain stable management of generator output in turbine power generation cycle. We developed turbine expansion model and measurement validation model for the performance calculation of generator using turbine output based on ASME (American Society of Mechanical Engineers) PTC (Performance Test Code). We also developed verification model for uncertain measurement data related to the turbine and generator output. Although the model in previous researches was developed using artificial neural network and kernel regression, the verification model in this paper was based on algorithms through Support Vector Machine (SVM) model to overcome the problems of unmeasured data. The selection procedures of related variables and data window for verification learning was also developed. The model reveals suitability in the estimation procss as the learning error was in the range of about 1%. The learning model can provide validated estimations for corrective performance analysis of turbine cycle output using the predictions of measurement data loss.