• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.24 no.2
• /
• pp.204-211
• /
• 2020

In this paper, we propose an optimal deep learning network architecture for main object location detection through weak supervised learning. The proposed network adds convolution blocks for improving the localization accuracy of the main object through weakly-supervised learning. The additional deep learning network consists of five additional blocks that add a composite product layer based on VGG-16. And the proposed network was trained by the method of weakly-supervised learning that does not require real location information for objects. In addition, Grad-CAM to compensate for the weakness of GAP in CAM, which is one of weak supervised learning methods, was used. The proposed network was tested through the CUB-200-2011 data set, we could obtain 50.13% in top-1 localization error. Also, the proposed network shows higher accuracy in detecting the main object than the existing method.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2020.11a
• /
• pp.118-122
• /
• 2020

문화재 영상 데이터는 방대한 양으로 인해 고해상도로 모두 저장이 어렵거나 시간이 지나 상대적으로 화질이 낮은 영상들이 다수 존재하기에 초해상화가 필요한 상황이 많다. 따라서 본 논문에서 처음으로 문화재 영상에 특화된 4 배 및 8 배 딥러닝 기반 초해상화 방식을 제안한다. 문화재 영상 데이터는 배경이 단조롭고 물체가 영상 중간에 위치한다는 특징이 있어 이를 고려해 중간 부분에서만 패치를 추출하는 방식을 적용하여 의미 있는 패치로 학습이 되도록 한다. 또 자연 영상 데이터 셋인 DIV2K 를 사용해 학습하는 방식과 직접 구성한 문화재 데이터 셋을 이용해 학습하는 방식, 그 둘을 적절히 함께 사용하여 학습하는 전이 학습 방법까지 세 가지로 학습하여 초해상화의 성능을 향상시키는 방법을 제안한다. 그 결과, 쌍삼차 보간법(Bicubic interpolation)보다 4 배 초해상화에서는 약 1.25dB, 8 배 초해상화에서는 약 1.26dB 의 성능 개선을 확인하였으며, 단순 DIV2K 로 학습한 방식보다는 4 배에서는 0.06dB, 8 배에서는 0.17dB 의 성능 개선을 확인하였다.

• Journal of Engineering Education Research
• /
• v.14 no.3
• /
• pp.9-14
• /
• 2011

This study was conducted to compare the learning styles and self-leadership between engineering college students with high and low team performance records. About 70% of students in high team performance group showed learning styles of converger and accommodator, whereas about 67% of students in low team performance group showed learning styles of accommodator and diverger. In regard to self-leadership, high team performance group showed higher level of self-leadership, especially self-observation, self-punishment, natural reward strategies, visualizing successful performance, self-talk, and evaluating beliefs and assumptions. It is recommended to provide the engineering students with the specialized training program to complement their learning styles and self-leadership strategies.

• Journal of The Korean Association of Information Education
• /
• v.12 no.2
• /
• pp.195-202
• /
• 2008

The programming learning by using a robot may be one of the most appropriate learning methods for enabling students to experience the creative learning of future society by avoiding the existing stereotyped style educational environment, and understand and improve algorithm which is the basic fundamental of mathematics and science. This study proposed four types of items of errors which may occur during robot programming by elementary school students, and made elementary school students in the fifth and sixth grades learn robot programming after developing the curriculum for the robot programming. Then, the study collected and classified errors that had occurred during the process of learning, and conducted a comparative analysis of computer-based programming language which had been previously studied. This study identified that robot programming in elementary school was shown superior to existing computer-based programming language as a creative learning method and tool through the field experience.

• Journal of the Korea Computer Graphics Society
• /
• v.27 no.3
• /
• pp.13-23
• /
• 2021

In this paper, we proposed a framework that generates the trajectory of a single rigid body based on its COM configuration and contact pose. Because we use a smaller input dimension than when we use a full body state, we can improve the learning time for reinforcement learning. Even with a 68% reduction in learning time (approximately two hours), the character trained by our network is more robust to external perturbations tolerating an external force of 1500 N which is about 7.5 times larger than the maximum magnitude from a previous approach. For this framework, we use centroidal dynamics to calculate the next configuration of the COM, and use reinforcement learning for obtaining a policy that gives us parameters for controlling the contact positions and forces.

• The KIPS Transactions:PartB
• /
• v.16B no.6
• /
• pp.479-488
• /
• 2009

This paper shows a reinforced construction scheme of training data for improvement of text classification by automatic search of boundary category. The documents laid on boundary area are usually misclassified as they are including multiple topics and features. which is the main factor that we focus on. In this paper, we propose an automated exploring methodology of optimal boundary category based on previous research. We consider the boundary area among target categories to new category to be required training, which are then added to the target category sementically. In experiments, we applied our method to complex documents by intentionally making errors in training process. The experimental results show that our system has high accuracy and reliability in noisy environment.

• Journal of the Korea Convergence Society
• /
• v.12 no.10
• /
• pp.81-88
• /
• 2021

Convertible bonds are financial products that contain the nature of both bonds and shares, which are generally issued by companies with lower credit ratings to increase liquidity. Conversion bonds rely on qualitative judgment in the past, although decision-making on whether and when to exercise the right to convert is the most important issue. Therefore, this paper proposes to apply artificial neural network techniques to scientifically determine the exercise of conversion rights. We distinguish between a total of 1,800 learning data published in the past and 200 predictive experimental data and build an artificial neural network learning model. As a result, the parity performance in most groups was excellent, achieving an average excess of about 10% or more. In particular, groups 3-6 recorded an average excess of about 20% and group 6 recorded an average excess of about 37%. This paper is meaningful in that it focused on solving decision problems by converging and applying machine learning techniques, a representative technology of the fourth industry, to the financial sector.

• The Journal of the Korea Contents Association
• /
• v.18 no.2
• /
• pp.427-437
• /
• 2018

This study explored the effectiveness of the online English learning program contents for elementary school students. The study used the online English learning program served by Gyeonggi province office of education. 107 students attending P elementary school in K city volunteered for the program. After studying English via the website for almost one year, they were asked to respond the questionnaires related to the contents of the online English program. Since the research investigated that the relations of students' grades, the time for the study, their diagnostic test scores, and the effectiveness of the contents, the survey responses were analyzed with Spearman correlation. As a result, older students thought that the type of problem-solving, the type of performing a task, WBI (Web Based Instruction) were not efficacious. Also, these types of online English program were chosen as ineffective from the students at the higher level. Whereas the type of private lesson, the lessons based on a story, and the type of animation were preferred to the students who spent longer time on the website. This highlights the need to consider the students' characteristics such as students' grades, the time for the study, and their English level when developing the contents of the online English learning program.

• Journal of IKEEE
• /
• v.24 no.2
• /
• pp.514-522
• /
• 2020

In Continuous Learning environment, we identify that the Catastrophic Forgetting phenomenon, which forgets the information of previously learned data, occurs easily between data having different domains. To control this phenomenon, we introduce how to measure the relationship between previously learned data and newly learned data through the distribution of the neural network's output, and how to use these measurements to mitigate the Catastrophic Forcing phenomenon. MNIST and EMNIST data were used for evaluation, and experiments showed an average 22.37% improvement in accuracy for previous data.

• Journal of Internet Computing and Services
• /
• v.20 no.3
• /
• pp.33-41
• /
• 2019

In this paper, we propose a method to learn the redesigned class with background and object for semantic segmentation of indoor scene image. Semantic image segmentation is a technique that divides meaningful parts of an image, such as walls and beds, into pixels. Previous work of semantic image segmentation has proposed methods of learning various object classes of images through neural networks, and it has been pointed out that there is insufficient accuracy compared to long learning time. However, in the problem of separating objects and backgrounds, there is no need to learn various object classes. So we concentrate on separating objects and backgrounds, and propose method to learn after class simplification. The accuracy of the proposed learning method is about 5 ~ 12% higher than the existing methods. In addition, the learning time is reduced by about 14 ~ 60 minutes when the class is configured differently In the same environment, and it shows that it is possible to efficiently learn about the problem of separating the object and the background.