• Proceedings of the Korea Information Processing Society Conference
• /
• 2022.05a
• /
• pp.528-530
• /
• 2022

심층강화학습에 완전 연결 신경망과 합성곱 신경망은 잘 활용되는 것에 반해 순환 신경망은 잘 활용되지 않는다. 이는 강화학습이 마르코프 속성을 전제로 하기 때문이다. 지금까지의 강화학습은 환경이 마르코프 속성을 만족하도록 사전 작업이 필요했다, 본 논문에서는 마르코프 속성을 따르지 않는 환경에서 이러한 사전 작업 없이도 순환 신경망의 은닉 상태를 통해 마르코프 속성을 학습함으로써 학습 성능을 개선할 수 있다는 것을 소개한다.

• Journal of IKEEE
• /
• v.28 no.2
• /
• pp.145-149
• /
• 2024

Recently, research on temporal action localization has been actively conducted. In this paper, unlike existing methods, we propose two approaches for learning bidirectional information by creating reverse playback videos for fully supervised temporal action localization. One approach involves creating training data by combining reverse playback videos and forward playback videos, while the other approach involves training separate models on videos with different playback directions. Experiments were conducted on the THUMOS-14 dataset using TALLFormer. When using both reverse and forward playback videos as training data, the performance was 5.1% lower than that of the existing method. On the other hand, using a model ensemble shows a 1.9% improvement in performance.

• The KIPS Transactions:PartA
• /
• v.9A no.1
• /
• pp.105-112
• /
• 2002

The purpose of this paper is to develop a courseware based on GIS (Geographic Information System) for improving visual and spatial learning efficiency of geography learning. The existing coursewares are not easy to encourage the learners in learning motivation, because these provide only the visual information using simple texts or imamges to the learners. To overcome these constraints, our courseware using GIS that can support spatial information can control the attribute information of map. In this paper, we define the courseware as the geography learning system. This courseware system enables the learners to take the perfect learning and the repetitive learning through the feedback after evaluating the learning degree. Also using geography learning application modules we implemented, the learners can participate directly in learning as well as search information in WWW.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2005.05a
• /
• pp.431-434
• /
• 2005

근원데이터나, 이원데이터를 이용한 문제를 해결하기 위해서는 많은 경우에 완전 해를 갖는 문제로 변형시키기 위해 정규화할 필요성이 있다. 본 논문에서는 이러한 정규화 인수를 찾는 문제를 기존의 GCV, L-Curve, 그리고 이원데이터를 RBF 신경회로망에 적용시킨 커널 학습법에 대한 각각의 성능을 비교실험을 통해 고찰한다. 이때 커널을 이용한 학습법의 성능을 향상하기 위해, 전체학습과 성능의 제한적 비례관계라는 설정아래, 각각의 학습에 따라 능동적으로 변화하는 동적모멘텀의 도입을 제안한다. 끝으로 제안된 동적모멘텀이 분류문제의 표준인 Iris 데이터, Singular 시스템의 대표적 모델인 가우시안 데이터, 그리고 마지막으로 1차원 이미지 복구문제인 Shaw데이터를 이용한 각각의 실험에서 분류문제와 회계문제 양쪽 모두에 있어 기존의 GCV, L-Curve와 동등하거나 우수한 성능이 있음을 보인다.

• Journal of The Korean Association For Science Education
• /
• v.31 no.2
• /
• pp.256-270
• /
• 2011

This study was designed to explore how learners' participation develop if provided with opportunities for various scientific practices and experiences in writing science magazine articles as an extra-curricular club activity and what factors facilitate these participation development. Data from participant observations, in-depth interviews with students, and documents were used to extract the common characteristics of the practices. The learners' development was categorized into 3 stages in terms of participation in the community of practice: peripheral participation, transitional participation, and full participation. As participation develops, situational interest developed to individual interests and value attachment. Students sought to get ideas from every day life, and finally, in the stage of full participation, advances in writing showed the characteristics of knowledge transformation. Best of all, the participants enjoyed and valued their participation showing identities as journalists. The nature of science magazine article, external scaffolding, and internalization through enjoyment and value attachment appeared to be decisive factors that facilitate the development of participation. Student's enculturation of writing for learning offers a possibility that continue to do so, even after they have left formal schooling and make a basis for lifelong learning.

• The Journal of Korean Association of Computer Education
• /
• v.7 no.1
• /
• pp.97-106
• /
• 2004

Recently, the demand for the customized courseware which is required from the learners is increased. Therefore the needs of the efficient and automated education agents in the web-based instruction are recognized. In this paper we propose a multi-agent system for course scheduling of learner-oriented using weakness analysis algorithm. At first proposed system analyze learner's result of evaluation and calculates learning accomplishment. From this accomplishment the multi-agent schedules the suitable course for the learner. The learner achieves an active and complete learning from the repeated and suitable course.

• Journal of the Korea Computer Industry Society
• /
• v.5 no.4
• /
• pp.481-490
• /
• 2004

Today, most education activities in the junior or senior schools are connected with evaluation. In fact, the evaluation to be accomplished is not the practice of college or university entrance examination but the evaluation analysis of studies accomplishment. In this program, students process their teaming contents according to their level, and they evaluate the accomplishment of learning by themselves. Through the various analysis of evaluation, students who are not in the appropriate level get into the learning plan of low level course. This study makes them notice the lack of teaming ability in time and proposes proper evaluation system which offers right feedbacks and various analysis information for themselves.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 2003.09b
• /
• pp.87-90
• /
• 2003

과거의 이동로봇 시스템은 완전한 자율주행이 주된 목표였으며 그때의 영상정보는 단지 모니터링을 하는 보조적인 수단으로 사용되었다. 그러나 지금은 이동 물체의 추적, 대상 물체의 인식과 판별, 특징 추출과 같은 다양한 응용분야에서 영상정보를 이용하는 연구가 활발히 진행되고 있다 또한 제어 측면에서는 전통적인 제어기법으로는 해결하기 힘들었던 여러 가지 비선형적인 제어를 지능제어 방법을 통하여 많이 해결하곤 하였다. 그러한 지능제어에서 신경망을 많이 사용하기도 한다. 최근에는 신경망의 학습에 많이 사용하는 방법 중 강화학습이 많이 사용되고 있다. 강화학습이란 동적인 제어평면에서 시행착오를 통해, 목적을 이루기 위해 각 상황에서 행동을 학습하는 방법이다. 그러므로 이러한 강화학습은 수많은 시행착오를 거쳐 그 대응 관계를 학습하게 된다. 제어에 사용되는 제어 파라메타는 어떠한 상태에 처할 수 있는 상태와 행동들, 그리고 상태의 변화, 또한 최적의 해를 구할 수 있는 포상알고리즘에 대해 다양하게 연구되고 있다. 본 논문에서 연구한 시스템은 비젼시스템과 Strong Arm 보드를 이용하여 대상물체의 색상과 형태를 파악한 후 실시간으로 물체를 추적할 수 있게 구성하였으며, 또한 물체 이동의 비선형적인 경향성을 강화학습을 통하여 물체이동의 비선형성을 보다 유연하게 대처하여 보다 안정하고 빠르며 정확하게 물체를 추적하는 방법을 실험을 통하여 제안하였다.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.22 no.2
• /
• pp.651-658
• /
• 2021

Because bolts with anti-loosening coatings are used mainly for joining safety-related components in automobiles, accurate automatic screening of these coatings is essential to detect defects efficiently. The performance of the convolutional neural network (CNN) used in a previous study [Identification of bolt coating defects using CNN and Grad-CAM] increased with increasing number of data for the analysis of image patterns and characteristics. On the other hand, obtaining the necessary amount of data for coated bolts is difficult, making training time-consuming. In this paper, resorting to the same VGG16 model as in a previous study, transfer learning was applied to decrease the training time and achieve the same or better accuracy with fewer data. The classifier was trained, considering the number of training data for this study and its similarity with ImageNet data. In conjunction with the fully connected layer, the highest accuracy was achieved (95%). To enhance the performance further, the last convolution layer and the classifier were fine-tuned, which resulted in a 2% increase in accuracy (97%). This shows that the learning time can be reduced by transfer learning and fine-tuning while maintaining a high screening accuracy.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.24 no.6B
• /
• pp.1048-1058
• /
• 1999

This paper compares some architectures to achieve the optimized WDM architecture for all optical transport network, the comparison is presented in terms of the number of required wavelength and LT. These architecture types are PPWDM, SHWDM, DHWDM and fully optical WDM. Topology is a static ring network where the routing pattern is fixed and traffic pattern has uniform demand. This paper also proposes an algorithm for the wavelength assignment for a folly optical WDM ring network which has full mesh traffic pattern. The algorithm is based on heuristic algorithm which assigns traffic connections according to their respective shortest path. Traffic described here that is to be passed through can be routed directly within the optical layer instead of having the higher layer to handle it.