• Annual Conference on Human and Language Technology
• /
• 2020.10a
• /
• pp.27-31
• /
• 2020

본 논문에서는 BERT가 합성된 새로운 Transformer 구조를 제안한 선행연구를 보완하기 위해 cardinality residual connection을 적용한 새로운 구조의 모델을 제안한다. Transformer의 인코더와 디코더의 셀프어텐션에 BERT를 각각 합성한 모델의 잔차연결을 수정하여 학습 속도와 번역 성능을 개선하고자 한다. 그리고 가중치를 다르게 부여하는 실험으로 어텐션을 선택하는 효과적인 방법을 제시하고 원문의 언어에 맞는 BERT를 사용하는 이유를 설명한다. IWSLT14 독일어-영어 말뭉치와 AI hub에서 제공하는 영어-한국어 말뭉치를 이용한 실험에서는 제안하는 방법의 모델이 기존 모델에 비해 더 나은 학습 속도와 번역 성능을 보였다.

• The Transactions of The Korean Institute of Electrical Engineers
• /
• v.64 no.1
• /
• pp.72-77
• /
• 2015

When connecting 6 lead wires from stator windings to the terminals of 3 phase induction motors for Y or ${\triangle}$ connection, it is feared that the polarities of windings could be reversed each other if the wire tags are lost or erased, resulting in inadmissibly high current to motors in case of starting. To protect motors against such situations, some test procedures are necessary during wire connection which need to be easy ways to electricians without particular tools except a general multi-tester and with less time-consuming in the field. This study focuses on a test measure to satisfy these requirements which is able to provide them a convenient procedure for winding polarity discrimination considering the field condition. Here, the proposed measure utilizes the residual magnetism of the rotor and checks the indication of voltage or current at windings which are induced by the residual flux of rotor when rotating it by hands with 3 stator windings connected in the form of I connection. Principle characteristics and experiment results for this method are analyzed in the view of the effectiveness and applicability for the winding polarity discrimination.

• Journal of Information Processing Systems
• /
• v.17 no.2
• /
• pp.385-398
• /
• 2021

The complexity of deep learning models affects the real-time performance of gesture recognition, thereby limiting the application of gesture recognition algorithms in actual scenarios. Hence, a residual learning neural network based on a deep convolutional neural network is proposed. First, small convolution kernels are used to extract the local details of gesture images. Subsequently, a shallow residual structure is built to share weights, thereby avoiding gradient disappearance or gradient explosion as the network layer deepens; consequently, the difficulty of model optimisation is simplified. Additional convolutional neural networks are used to accelerate the refinement of deep abstract features based on the spatial importance of the gesture feature distribution. Finally, a fully connected cascade softmax classifier is used to complete the gesture recognition. Compared with the dense connection multiplexing feature information network, the proposed algorithm is optimised in feature multiplexing to avoid performance fluctuations caused by feature redundancy. Experimental results from the ISOGD gesture dataset and Gesture dataset prove that the proposed algorithm affords a fast convergence speed and high accuracy.

• Steel and Composite Structures
• /
• v.37 no.6
• /
• pp.649-662
• /
• 2020

Although several types of rigid shear connectors have been developed particularly to increase load-carrying capacity, application is limited due to the complicated details of such connection. In this study, push-out tests were performed for specimens with hybrid shear connectors using headed studs and shear plates to identify the effects of each parameter on the structural performance of such shear connection. The test parameters included steel ratios of headed stud to shear plate, connection length, and embedded depth of shear plates. The peak strength and residual strength were estimated using various shear transfer mechanisms such as stud shear, concrete bearing, and shear friction. The hybrid shear connectors using shear plates and headed studs showed large load-carrying capacity and deformation capacity. The peak strength was predicted by the concrete bearing strength of the shear plates. The residual strength was sufficiently predicted by the stud shear strength of headed studs or by shear friction strength of dowel reinforcing bars. Further, the finite element analysis was performed to verify the shear transfer mechanism of the connection with hybrid shear connector.

• Structural Engineering and Mechanics
• /
• v.87 no.2
• /
• pp.151-164
• /
• 2023

The latest earthquake's costly repairs and economic disruption were brought on by excessive residual drift. Self-centering systems are one of the most efficient ways in the current generation of seismic resistance system to get rid of and reduce residual drift. The mechanics and behavior of the self-centering system in response to seismic forces were impacted by a number of important factors. The amount of post-tensioning (PT) force, which is often employed for the standing posture after an earthquake, is the first important component. The energy dissipater element is another one that has a significant impact on how the self-centering system behaves. Using the damper as a replaceable and affordable tool and fuse in self-centering frames has been recommended to boost energy absorption and dampening of structural systems during earthquakes. In this research, the self-centering steel moment frame connections are equipped with cushion flexural dampers (CFDs) as an energy dissipator system to increase energy absorption, post-yielding stiffness, and ease replacement after an earthquake. Also, it has been carefully considered how to reduce permanent deformations in the self-centering steel moment frames exposed to seismic loads while maintaining adequate stiffness, strength, and ductility. After confirming the FE model's findings with an earlier experimental PT connection, the behavior of the self-centering connection using CFD has been surveyed in this study. The FE modeling takes into account strands preloading as well as geometric and material nonlinearities. In addition to contact and sliding phenomena, gap opening and closing actions are included in the models. According to the findings, self-centering moment-resisting frames (SF-MRF) combined with CFD enhance post-yielding stiffness and energy absorption with the least amount of permeant deformation in a certain CFD thickness. The obtained findings demonstrate that the effective energy dissipation ratio (β), is increased to 0.25% while also lowering the residual drift to less than 0.5%. Also, this enhancement in the self-centering connection with CFD's seismic performance was attained with a respectable moment capacity to beam plastic moment capacity ratio.

• Journal of Information Processing Systems
• /
• v.17 no.6
• /
• pp.1115-1126
• /
• 2021

Plant disease is one of the most irritating problems for agriculture growers. Thus, timely detection of plant diseases is of high importance to practical value, and corresponding measures can be taken at the early stage of plant diseases. Therefore, numerous researchers have made unremitting efforts in plant disease identification. However, this problem was not solved effectively until the development of artificial intelligence and big data technologies, especially the wide application of deep learning models in different fields. Since the symptoms of plant diseases mainly appear visually on leaves, computer vision and machine learning technologies are effective and rapid methods for identifying various kinds of plant diseases. As one of the fruits with the highest nutritional value, apple production directly affects the quality of life, and it is important to prevent disease intrusion in advance for yield and taste. In this study, an improved deep residual network is proposed for apple leaf disease identification in a novel way, a global residual connection is added to the original residual network, and the local residual connection architecture is optimized. Including that 1,977 apple leaf disease images with three categories that are collected in this study, experimental results show that the proposed method has achieved 98.74% top-1 accuracy on the test set, outperforming the existing state-of-the-art models in apple leaf disease identification tasks, and proving the effectiveness of the proposed method.

• Computers and Concrete
• /
• v.25 no.4
• /
• pp.293-302
• /
• 2020

Connections play a significant role in strength of structures against earthquake-induced loads. According to the post-seismic reports, connection failure is a cause of overall failure in reinforced concrete (RC) structures. Connection failure results in a sudden increase in inter-story drift, followed by early and progressive failure across the entire structure. This article investigated the cyclic performance and behavioral improvement of shape-memory alloy-based connections (SMA-based connections). The novelty of the present work is focused on the effect of shape memory alloy bars is damage reduction, strain recoverability, and cracking distribution of the stated material in RC moment frames under seismic loads using 3D nonlinear static analyses. The present numerical study was verified using two experimental connections. Then, the performance of connections was studied using 14 models with different reinforcement details on a scale of 3:4. The response parameters under study included moment-rotation, secant stiffness, energy dissipation, strain of bar, and moment-curvature of the connection. The connections were simulated using LS-DYNA environment. The models with longitudinal SMA-based bars, as the main bars, could eliminate residual plastic rotations and thus reduce the demand for post-earthquake structural repairs. The flag-shaped stress-strain curve of SMA-based materials resulted in a very slight residual drift in such connections.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.14 no.6
• /
• pp.2480-2496
• /
• 2020

Considering that high-dose X-ray radiation during CT scans may bring potential risks to patients, in the medical imaging industry there has been increasing emphasis on low-dose CT. Due to complex statistical characteristics of noise found in low-dose CT images, many traditional methods are difficult to preserve structural details effectively while suppressing noise and artifacts. Inspired by the deep learning techniques, we propose a densely connected residual network (DCRN) for low-dose CT image noise cancelation, which combines the ideas of dense connection with residual learning. On one hand, dense connection maximizes information flow between layers in the network, which is beneficial to maintain structural details when denoising images. On the other hand, residual learning paired with batch normalization would allow for decreased training speed and better noise reduction performance in images. The experiments are performed on the 100 CT images selected from a public medical dataset-TCIA(The Cancer Imaging Archive). Compared with the other three competitive denoising algorithms, both subjective visual effect and objective evaluation indexes which include PSNR, RMSE, MAE and SSIM show that the proposed network can improve LDCT images quality more effectively while maintaining a low computational cost. In the objective evaluation indexes, the highest PSNR 33.67, RMSE 5.659, MAE 1.965 and SSIM 0.9434 are achieved by the proposed method. Especially for RMSE, compare with the best performing algorithm in the comparison algorithms, the proposed network increases it by 7 percentage points.

• Journal of Broadcast Engineering
• /
• v.24 no.4
• /
• pp.633-642
• /
• 2019

Recently, the convolution neural network (CNN) model at a single image super-resolution (SISR) have been very successful. The residual learning method can improve training stability and network performance in CNN. In this paper, we propose a SISR using recursive residual network architecture by introducing dense skip connections for learning nonlinear mapping from low-resolution input image to high-resolution target image. The proposed SISR method adopts a method of the recursive residual learning to mitigate the difficulty of the deep network training and remove unnecessary modules for easier to optimize in CNN layers because of the concise and compact recursive network via dense skip connection method. The proposed method not only alleviates the vanishing-gradient problem of a very deep network, but also get the outstanding performance with low complexity of neural network, which allows the neural network to perform training, thereby exhibiting improved performance of SISR method.

• Journal of the Korean Institute of Electrical and Electronic Material Engineers
• /
• v.32 no.3
• /
• pp.241-245
• /
• 2019

This work presents the response characteristics of a ZnO varistor to conductive EMP. An E1 pulse, standardized to MIL-STD-188-125-1, was applied to the varistors wherein the residual current and response times were measured with the applied E1 pulse current. Additionally, the response time was measured according to the length of the connection path. Consequently, the amplitude of the residual voltage through the ZnO varistors was increased with increasing amplitude of the applied E1 pulse current. As the length of the connection path increased, the operating response time and residual peak voltage also increased. These results indicate that the response characteristics of ZnO varistors can be applied to basic data to support the use of varistors as a protective measure against conductive EMP.