• The Journal of Information Systems
• /
• v.30 no.4
• /
• pp.203-226
• /
• 2021

Purpose The purpose of this study is to enhance the accuracy score of personal credit scoring using the convolutional neural networks and secure the transparency of the deep learning model using eXplainalbe Artifical Inteligence(XAI) technique. Design/methodology/approach This study built a classification model by using the convolutional neural networks(CNN) and applied a methodology that is transformation of numerical data to imaged data to apply CNN on personal credit data. Then layer-wise relevance propagation(LRP) was applied to model we constructed to find what variables are more influenced to the output value. Findings According to the empirical analysis result, this study confirmed that accuracy score by model using CNN is highest among other models using logistic regression, neural networks, and support vector machines. In addition, With the LRP that is one of the technique of XAI, variables that have a great influence on calculating the output value for each observation could be found.

• Journal of the Korea Society of Computer and Information
• /
• v.29 no.7
• /
• pp.11-20
• /
• 2024

In this paper, we propose the BAG (Bit-width Aware Generator) and the Intermediate Layer Knowledge Distillation using Channel-wise Attention to reduce the knowledge gap between a quantized network, a full-precision network, and a generator in GDFQ (Generative Data-Free Quantization). Since the generator in GDFQ is only trained by the feedback from the full-precision network, the gap resulting in decreased capability due to low bit-width of the quantized network has no effect on training the generator. To alleviate this problem, BAG is quantized with same bit-width of the quantized network, and it can generate synthetic images, which are effectively used for training the quantized network. Typically, the knowledge gap between the quantized network and the full-precision network is also important. To resolve this, we compute channel-wise attention of outputs of convolutional layers, and minimize the loss function as the distance of them. As the result, the quantized network can learn which channels to focus on more from mimicking the full-precision network. To prove the efficiency of proposed methods, we quantize the network trained on CIFAR-100 with 3 bit-width weights and activations, and train it and the generator with our method. As the result, we achieve 56.14% Top-1 Accuracy and increase 3.4% higher accuracy compared to our baseline AdaDFQ.

• Wind and Structures
• /
• v.12 no.3
• /
• pp.219-237
• /
• 2009

This study examines the accuracy of large-eddy simulation (LES) to simulate the flow around a large irregular sloping complex terrain. Typically, real built up environments are surrounded by complex terrain geometries with many features. The complex terrain surrounding The Hong Kong University of Science and Technology campus was modelled and the flow over an uphill slope was simulated. The simulated results, including mean velocity profiles and turbulence intensities, were compared with the flow characteristics measured in a wind tunnel model test. Given the size of the domain and the corresponding constraints on the resolution of the simulation, the mean velocity components within the boundary layer flow, especially in the stream-wise direction were found to be reasonably well replicated by the LES. The turbulence intensity values were found to differ from the wind tunnel results in the building recirculation zones, mostly due to the constraints placed on spatial and temporal resolutions. Based on the validated mean velocity profile results, the flow-structure interactions around these buildings and the surrounding terrain were examined.

• Advanced Composite Materials
• /
• v.16 no.3
• /
• pp.223-244
• /
• 2007

This study experimentally and numerically investigated the tensile damage progress in stitched laminates. In particular, it focused on the effects of stitching on the damage progress. First, we experimentally confirmed that ply cracks and delamination appeared under load regardless of stitching. We then performed damage-extension simulation for stitched laminates using a layer-wise finite element model with stitch threads as beam elements, in which the damage (ply cracks and delamination) was represented by cohesive elements. A detailed comparison between observation and the simulated results confirmed that stitching had little effect on the onset and accumulation of ply cracks. Furthermore, we demonstrated that the stitch threads significantly suppressed the extension of the delamination.

• Computational Structural Engineering
• /
• v.9 no.2
• /
• pp.173-182
• /
• 1996

The present study proposes a detection technique for delaminations in a laminated compoiste structure. the proposed technique optimizes the spatial distribution of harmonic excitation so as to magnify the difference in response between the delaminated and intact structures. The technique is evaluated by numerical simulation of two-layered aluminum beams. Effects of measurement and geometric noises are included in the analysis. A finite element model for a delaminated beam, based on the layer-wise laminated plate theory in conjunction with a step function to simulate ddelaminations, is used.

• Journal of Korea Multimedia Society
• /
• v.25 no.6
• /
• pp.819-832
• /
• 2022

Cross-domain recommender system transfers knowledge across different domains to improve the recommendation performance in a target domain that has a relatively sparse model. However, they suffer from the "negative transfer" in which transferred knowledge operates as noise. This paper proposes a novel Multiple Fusion-based Deep Cross-Domain Recommendation named MFDCR. We exploit Doc2Vec, one of the famous word embedding techniques, to fuse data user-wise and transfer knowledge across multi-domains. It alleviates the "negative transfer" problem. Additionally, we introduce a simple multi-layer perception to learn the user-item interactions and predict the possibility of preferring items by users. Extensive experiments with three domain datasets from one of the most famous services Amazon demonstrate that MFDCR outperforms recent single and cross-domain recommendation algorithms. Furthermore, experimental results show that MFDCR can address the problem of "negative transfer" and improve recommendation performance for multiple domains simultaneously. In addition, we show that our approach is efficient in extending toward more domains.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2020.07a
• /
• pp.540-543
• /
• 2020

기존의 초해상도 딥러닝 기법은 모델의 깊이가 깊어지면서, 좋은 성능을 내지만 점점 더 복잡해지고 있고, 실제로 사용하는데 있어 많은 시간을 요구한다. 이를 해결하기 위해, 우리는 딥러닝 모델의 가중치를 양자화 하여 추론시간을 줄이고자 한다. 초해상도 모델은 feature extraction, non-linear mapping, reconstruction 세 부분으로 나누어져 있으며, 레이어 사이에 많은 skip-connection 이 존재하는 특징이 있다. 따라서 양자화 시 최종 성능 하락에 미치는 영향력이 레이어 별로 다르며, 이를 감안하여 강화학습으로 레이어 별 최적 bit 를 찾아 성능 하락을 최소화한다. 본 논문에서는 Skip-connection 이 많이 존재하는 MSRN 을 사용하였으며, 결과에서 feature extraction, reconstruction 부분과 블록 내 특정 위치의 레이어가 항상 높은 bit 를 가짐을 알 수 있다. 기존에 영상 분류에 한정되어 사용되었던 혼합 bit 양자화를 사용하여 초해상도 딥러닝 기법의 모델 사이즈를 줄인 최초의 논문이며, 제안 방법은 모바일 등 제한된 환경에 적용 가능할 것으로 생각된다.

• International Journal of Aerospace System Engineering
• /
• v.3 no.1
• /
• pp.10-16
• /
• 2016

A trigonometric Layerwise higher order shear deformation theory (TLHSDT) is developed and implemented for free vibration and buckling analysis of laminated composite and sandwich plates by analytical and finite element formulation. The present model assumes parabolic variation of out-plane stresses through the depth of the plate and also accomplish the zero transverse shear stresses over the surface of the plate. Thus a need of shear correction factor is obviated. The present zigzag model able to meet the transverse shear stress continuity and zigzag form of in-plane displacement continuity at the plate interfaces. Hence, botheration of shear correction coefficient is neglected. In the case of analytical method, the governing differential equation and boundary conditions are obtained from the principle of virtual work. For the finite element formulation, an efficient eight noded $C^0$ continuous isoparametric serendipity element is established and employed to examine the dynamic analysis. Like FSDT, the considered mathematical model possesses similar number of variables and which decides the present models computationally more effective. Several numerical predictions are carried out and results are compared with those of other existing numerical approaches.

• Advances in aircraft and spacecraft science
• /
• v.3 no.3
• /
• pp.331-349
• /
• 2016

Stiff and light fibre reinforced composites as used in air- and space-craft applications tend to high sound emission. Therefore, the damping properties are essential for the entire structural and acoustic engineering. Viscous damping is an established and reasonably linear model of the dissipation behaviour. Commonly, it is assumed to be isotropic and constant over all modes. For anisotropic materials it depends on the fibre orientation as well as the elastic and thermal material properties. To portray the orthogonal anisotropic behaviour, a model for unidirectional fibre reinforced plastics (frp) has been developed based on the classical laminate theory by ADAMS and BACON starting in 1973. Their approach includes three damping coefficients - for longitudinal damping in fibre direction, damping transversal to the fibres and shear based dissipation. The damping of a laminate is then accumulated layer wise including the anisotropic stiffness. So far, the model has been applied mainly to thermoset matrix materials. In this study, an experimental parameter estimation for different thermoplastic frp with angle ply and cross ply layups was carried out by measuring free vibrations of cantilever beams. The results show potential and limits of the ADAMS/BACON damping criterion. In addition, a possibility of modelling the anisotropic damping is shown. The implementation in standard FEA software is used to study the influence of boundary conditions on the damping properties and numerically estimate the radiated sound power of thin-walled frp parts.

• Journal of Powder Materials
• /
• v.24 no.6
• /
• pp.494-507
• /
• 2017

A three-dimensional physical part can be fabricated from a three-dimensional digital model in a layer-wise manner via additive manufacturing (AM) technology, which is different from the conventional subtractive manufacturing technology. Numerous studies have been conducted to take advantage of the AM opportunities to penetrate bespoke custom product markets, functional engineering part markets, volatile low-volume markets, and spare part markets. Nevertheless, materials issues, machines issues, product issues, and qualification/certification issues still prevent the AM technology from being extensively adopted in industries. The present study briefly reviews the standard classification, technological structures, industrial applications, technological advances, and qualification/certification activities of the AM technology. The economics, productivity, quality, and reliability of the AM technology should be further improved to pass through the technology adoption lifecycle of innovation technology. The AM technology is continuously evolving through the introduction of PM materials, hybridization of AM and conventional manufacturing technologies, adoption of process diagnostics and control systems, and enhanced standardization of the whole lifecycle qualification and certification methodology.