• International Journal of Internet, Broadcasting and Communication
• /
• v.9 no.4
• /
• pp.1-7
• /
• 2017

Clipart is artificial visual contents that are created using various tools such as Illustrator to highlight some information. Here, the style of the clipart plays a critical role in determining how it looks. However, previous studies on clipart are focused only on the object recognition [16], segmentation, and retrieval of clipart images using hand-craft image features. Recently, some clipart classification researches based on the style similarity using CNN have been proposed, however, they have used different CNN-models and experimented with different benchmark dataset so that it is very hard to compare their performances. This paper presents an experimental analysis of the clipart classification based on the style similarity with two well-known CNN-models (Inception Resnet V2 [13] and VGG-16 [14] and transfers learning with the same benchmark dataset (Microsoft Style Dataset 3.6K). From this experiment, we find out that the accuracy of Inception Resnet V2 is better than VGG for clipart style classification because of its deep nature and convolution map with various sizes in parallel. We also find out that the end-to-end training can improve the accuracy more than 20% in both CNN models.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2021.11a
• /
• pp.61-63
• /
• 2021

Machine learning model gained the popularity in recent years as multi-national companies have incorporated machine learning in their services. Such service is called machine learning as a service (MLaSS). Such services are provided to users based on charge-per-query which triggers the motivations for adversaries to steal the trained victim model to reduce the cost of using the service. Therefore, it is important for companies that provide MLaSS to protect their intellectual property (IP) against adversaries. It has been arms race between the attack and defence in a context of the privacy of machine learning models. In this paper, we provide a comprehensive study of recent development in protecting privacy of machine learning models.

• Journal of Magnetics
• /
• v.17 no.1
• /
• pp.1-5
• /
• 2012

The coercivity of a single 27 nm-spherical barium ferrite (S-BaFe) particle was simulated using three models: 1) Gibbs free energy (GFE), 2) Landau-Lifshitz-Gilbert (LLG), and 3) Stoner-Wohlfarth (S-W). Spherically and hexagonally shaped particles were used in the GFE and LLG simulations to investigate coercivity with the different shape anisotropies. The effect of shape was not included in the S-W model. It was found that the models using a spherical shape resulted in a coercivity higher than the models using the hexagonal shape with both shapes having the same diameter. The coercivity estimated with the S-W model was approximately the same as that for the spherical-shape models, which indicates that spherical shape has no significant effect on the particle's coercivity at nanoscale.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2023.07a
• /
• pp.371-372
• /
• 2023

본 논문에서는 대규모 언어 모델(Large Language Models)을 기반으로 한 입학 상담용 챗봇을 설계하였다. 입시 전문 LLM은 Polyglot-ko 5.8B을 베이스 모델로 대학의 입시 관련 데이터를 수집, 가공한 후 데이터 증강을 하여 파인튜닝 하였다. 또한, 모델 성능 향상을 위해 RLHF의 후 공정을 진행하였다. 제안 챗봇은 생성한 입시 LLM을 기반으로 웹브라우저를 통해 접근하여 입시 상담 자동 응답 서비스를 활용할 수 있다.

• JSTS:Journal of Semiconductor Technology and Science
• /
• v.15 no.1
• /
• pp.145-149
• /
• 2015

A reconfigurable lighting engine for widely used lighting models is proposed for low-power GPU shaders. Conventionally, lighting operations that involve many complex arithmetic operations were calculated by the shader programs on the GPU, which led to a significant energy overhead. In this letter, we propose a lighting engine to improve the energy-efficiency by supporting the widely used advanced lighting models in hardware. It supports the Blinn-Phong, Oren-Nayar, and Cook-Torrance models, by exploiting the logarithmic arithmetic and optimizing the trigonometric function evaluations for the energy-efficiency. Experimental results demonstrate 12.7%, 42.5%, and 35.5% reductions in terms of power-delay product from the shader program implementations for each lighting model. Moreover, our work shows 10.1% higher energy-efficiency for the Blinn-Phong model compared to the prior art.

• Transactions of the Korean Society of Machine Tool Engineers
• /
• v.13 no.3
• /
• pp.30-36
• /
• 2004

The exchange of model design data among heterogeneous CAD systems is very difficult because each CAD system has different data structures suitable for its own functions. STEP represents product information in a common computer-interpretable form that is required to remain complete and consistent when the product information is needed to be exchanged among different computer systems. However, STEP has complex architecture to represent point, line, curve and vectors of element. Moreover it can't represent geometry data of feature based models. In this study, a structure of XML document that represents geometry data of feature based models as neutral format has been developed. To use the developed XML document, a converter also has been developed to exchange modules so that it can exchange feature based data models among heterogeneous CAD systems. Developed XML document and Converter have been applied to commercial CAD systems.

• Proceedings of the Korean Society for Bioinformatics Conference
• /
• 2003.10a
• /
• pp.179-182
• /
• 2003

The Development of promoter recognition systems is a interesting problem in computational biology. In this paper, we introduce a intelligent system fur promoter recognition with multiple decision models using artificial neural networks. We have trained this models with 1871 human promoter sequences and 5230exon and intron sequences. Our system is found to perform better than other promoter finding systems insensitivity and specificity measures. We have tested our system with Chromosome 22 dataset.

• The Journal of Korean Society for Radiation Therapy
• /
• v.2 no.1
• /
• pp.87-92
• /
• 1987

The authors have intended to measure intrinsic dose distribution by Farmer dosimeter in irregularly shaped fields such as L, M, T,-shape model in order to determine dose inhomogeneity in those models. We made 2 off-axis points in each model and measured the depth dose at 1.5,5, and 9cm below surface. The results showed $1-3\%$ dose discrepancy between 2 points. We also measured the depth dose by geometric approximation and computer calculation in those models, and came to the conclusion that computer calculation using Clarkson's principle is simpler and the measurements are to the ideal data obtained by the experiment in those three models of irregularly shaped fields than those of geometric approximation method.

• Journal of Biosystems Engineering
• /
• v.15 no.4
• /
• pp.290-297
• /
• 1990

The purposes of this study were to perform theoretical analysis of an electronic-hydraulic hitch control system for position and draft control of tractor implements and to investigate the performance of the control system through computer simulation. Computer simulation models which could predict the responses of the system to the step and sinusoidal inputs in position and draft controls were developed using the simulation package "TUTSIM". The effects of control mode, hydraulic flow rate, deadband, and proportional constant on control performance of the system were investigated. The simulated results were compard with the experimental ones to verify the simulation models. The simulation models appeared to be a useful means for the analysis and the design of the electronic-hydraulic hitch control system.

• Journal of Computing Science and Engineering
• /
• v.5 no.4
• /
• pp.288-293
• /
• 2011

To resolve ambiguities in speech act classification, various machine learning models have been proposed over the past 10 years. In this paper, we review these machine learning models and present the results of experimental comparison of three representative models, namely the decision tree, the support vector machine (SVM), and the maximum entropy model (MEM). In experiments with a goal-oriented dialogue corpus in the schedule management domain, we found that the MEM has lighter hardware requirements, whereas the SVM has better performance characteristics.