• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.14 no.12
• /
• pp.4816-4834
• /
• 2020

This paper proposes transfer learning and fine-tuning techniques for a deep learning model to detect three distinct brain tumors from Magnetic Resonance Imaging (MRI) scans. In this work, the recent YOLOv4 model trained using a collection of 3064 T1-weighted Contrast-Enhanced (CE)-MRI scans that were pre-processed and labeled for the task. This work trained with the partial 29-layer YOLOv4-Tiny and fine-tuned to work optimally and run efficiently in most platforms with reliable performance. With the help of transfer learning, the model had initial leverage to train faster with pre-trained weights from the COCO dataset, generating a robust set of features required for brain tumor detection. The results yielded the highest mean average precision of 93.14%, a 90.34% precision, 88.58% recall, and 89.45% F1-Score outperforming other previous versions of the YOLO detection models and other studies that used bounding box detections for the same task like Faster R-CNN. As concluded, the YOLOv4-Tiny can work efficiently to detect brain tumors automatically at a rapid phase with the help of proper fine-tuning and transfer learning. This work contributes mainly to assist medical experts in the diagnostic process of brain tumors.

• The Korean Journal of Quaternary Research
• /
• v.19 no.2
• /
• pp.74-79
• /
• 2005

Recently, several attempts have been made to provide reasonable information on unusual severe weather phenomena such as tolerant heavy rains and very wild typhoons. Quantitative precipitation forecasts and probabilistic quantitative precipitation forecasts (QPFs and PQPFs, respectively) might be one of the most promising methodologies for early warning on the flesh floods because those diagnostic precipitation models require less computational resources than fine-mesh full-dynamics non-hydrostatic mesoscale model. The diagnostic rainfall model used in this study is the named QPM(Quantitative Precipitation Model), which calculates the rainfall by considering the effect of small-scale topography which is not treated in the mesoscale model. We examine the capability of probabilistic diagnostic rainfall model in terms of how well represented the observed several rainfall events and what is the most optimistic resolution of the mesoscale model in which diagnostic rainfall model is nested. Also, we examine the integration time to provide reasonable fine-mesh rainfall information. When we apply this QPM directly to 27 km mesh meso-scale model (called as M27-Q3), it takes about 15 min. while it takes about 87 min. to get the same resolution precipitation information with full dynamic downscaling method (called M27-9-3). The quality of precipitation forecast by M27-Q3 is quite comparable with the results of M27-9-3 with reasonable threshold value for precipitation. Based on a series of examination we may conclude that the proosed QPM has a capability to provide fine-mesh rainfall information in terms of time and accuracy compared to full dynamical fine-mesh meso-scale model.

• Structural Engineering and Mechanics
• /
• v.84 no.1
• /
• pp.113-127
• /
• 2022

Steel plate shear walls (SPSWs) are commonly utilized to provide lateral stiffness in high-rise structures. The simplified model is frequently used instead of the fine-scale model in the design of buildings with SPSWs. To predict the lateral strength of steel plate shear walls with diagonal stiffeners (DS-SPSWs), a simplified model is presented, namely the cross brace-strip model (CBSM). The bearing capacity and internal forces of columns for DS-SPSWs are calculated. In addition, a modification coefficient is introduced to account for the shear action of the thin plate. The feasibility of the CBSM is validated by comparing the numerical results with theoretical and experimental results. The numerical results from the CBSM and fine-scale model, which represent the bearing capacity of the DS-SPSW with varied stiffened plate dimensions, are in good accord with the theoretical values. The difference in bearing capacity between the CBSM and the fine-scale model is less than 1.35%. The errors of the bearing capacity from the CBSM are less than 5.67% when compared to the test results of the DS-SPSW. Furthermore, the shear and axial forces of CBSM agree with the results of the fine-scale model and theoretical analysis. As a result, the CBSM, which reflects the contribution of diagonal stiffeners to the lateral resistance of the SPSW as well as the effects on the shear and axial forces of the columns, can significantly improve the design accuracy and efficiency of buildings with DS-SPSWs.

• Journal of Oriental Neuropsychiatry
• /
• v.19 no.3
• /
• pp.85-100
• /
• 2008

Objective: This experiment was designed to investigate the effect of the KSKH hot water extract & ultra-fine powder on microglia and memory deficit model. Method: The effects of the KSKH hot water extract on expression of IL-1$\beta$, IL-6, TNF-$\alpha$ mRNA and production of IL-1$\beta$, IL-6, TNF-$\alpha$ in BV2 microglial cell line treated by lipopolysacchaide(LPS) were investigated. The effects of the KSKH hot water extract & ultra-fine-fine powder on the behavior of the memory deficit mice induced by scopolamine and AChE in serum of the memory deficit mice induced by scopolamine were investigated. Results: 1. The KSKH hot water extract suppressed the expression of IL-1$\beta$, IL-6, TNF-$\alpha$ mRNA in BV2 microglial cell line treated by LPS. 2. The KSKH hot water extract suppressed the production of IL-1$\beta$, IL-6, TNF-$\alpha$ in 100$\mu g/m\ell$ concentration of BV2 microglial cell line culture supernatant. 3. The KSKH hot water extract & ultra-fine powder decreased AChE activation significantly in the serum of the memory deficit mice induced by scopolamine. 4. The KSKH hot water extract & ultra-fine powder showed significant effect on memory impairment in the stop-through latency type of Morris water maze test. Conclusions: This experiment shows that the KSKH hot water extract & ultra-fine powder might be effective for the prevention and treatment of amnesia and Alzheimer's disease. Investigation into the clinical use of the KSKH hot water extract & ultra-fine powder for amnesia and Alzheimer's disease is suggested for future research.

• Journal of Communications and Networks
• /
• v.3 no.4
• /
• pp.383-395
• /
• 2001

Recent measurement and simulation studies have revealed that wide area network traffic displays complex statistical characteristics-possibly multifractal scaling-on fine timescales, in addition to the well-known properly of self-similar scaling on coarser timescales. In this paper we investigate the performance and network engineering significance of these fine timescale features using measured TCP anti MPEG2 video traces, queueing simulations and analytical arguments. We demonstrate that the fine timescale features can affect performance substantially at low and intermediate utilizations, while the longer timescale self-similarity is important at intermediate and high utilizations. We relate the fine timescale structure in the measured TCP traces to flow controls, and show that UDP traffic-which is not flow controlled-lacks such fine timescale structure. Likewise we relate the fine timescale structure in video MPEG2 traces to sub-frame encoding. We show that it is possibly to construct a relatively parsimonious multi-fractal cascade model of fine timescale features that matches the queueing performance of both the TCP and video traces. We outline an analytical method ta estimate performance for traffic that is self-similar on coarse timescales and multi-fractal on fine timescales, and show that the engineering problem of setting safe operating points for planning or admission controls can be significantly influenced by fine timescale fluctuations in network traffic. The work reported here can be used to model the relevant characteristics of wide area traffic across a full range of engineering timescales, and can be the basis of more accurate network performance analysis and engineering.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.22 no.4
• /
• pp.595-601
• /
• 2018

Recently, as the fine dust level has risen rapidly, there is a great interest. Exposure to fine dust is associated with the development of respiratory and cardiovascular diseases and has been reported to increase death rate. In addition, there exist damage to fine dusts continues at industrial sites. However, exposure to fine dust is inevitable in modern life. Therefore, predicting and minimizing exposure to fine dust is the most efficient way to reduce health and industrial damages. Existing fine dust prediction model is estimated as good, normal, poor, and very bad, depending on the concentration range of the fine dust rather than the concentration value. In this paper, we study and implement to predict the PM10 level by applying the Artificial neural network algorithm and the K-Nearest Neighbor algorithm, which are machine learning algorithms, using the actual weather and air quality data.

• Journal of the Korea Convergence Society
• /
• v.10 no.8
• /
• pp.171-179
• /
• 2019

This study was to develop the health education program model of young children based on e-book using QR code to prevent and manage fine dust. We developed program to prevent and manage fine dust for young children' health through in-depth discussions with 2 young children's health experts and 3 young children education experts, and developed the e-book using QR code. The program model for fine dust's prevention education based on e-book using QR code was verified for suitability as a field adaptation and teaching medium for 5-year old. The results of this study are as follows: First, the program model for fine dust's prevention education based on e-book using QR code was developed through 3 stages of appreciation activity-central activity-finishing activity. Second, the e-book using QR code is a useful teaching medium applicable to fine dust's prevention education program. In conclusion, the e-book-based fine dust's prevention education program using QR code will be used as an appropriate education method for fine dust's prevention and health management activities in the field of education for young children.

• Proceedings of The Korean Society of Agricultural and Forest Meteorology Conference
• /
• 2011.11a
• /
• pp.20-27
• /
• 2011

A system for fine-resolution long-range weather forecast is introduced in this study. The system is basically consisted of a global-scale coupled general circulation model (CGCM) and Weather Research and Forecast (WRF) regional model. The system makes use of a data assimilation method in order to reduce the initial shock or drift that occurs at the beginning of coupling due to imbalance between model dynamics and observed initial condition. The long-range predictions are produced in the system based on a non-linear ensemble method. At the same time, the model bias are eliminated by estimating the difference between hindcast model climate and observation. In this research, the predictability of the forecast system is studied, and it is illustrated that the system can be effectively used for the high resolution long-term weather prediction. Also, using the system, fine-resolution climatological data has been produced with high degree of accuracy. It is proved that the production of agrometeorological variables that are not intensively observed are also possible.

• Transactions on Semiconductor Engineering
• /
• v.1 no.1
• /
• pp.23-30
• /
• 2023

With the recent development of Generative AI technology by IT giants, the size of the transformer model is increasing exponentially over trillion won. In order to continuously enable these AI services, it is essential to reduce the weight of the model. In this paper, we find a hardware-friendly structured pruning pattern and propose a lightweight method of the transformer model. Since compression proceeds by utilizing the characteristics of the model algorithm, the size of the model can be reduced and performance can be maintained as much as possible. Experiments show that the structured pruning proposed when pruning GPT-2 and BERT language models shows almost similar performance to fine-grained pruning even in highly sparse regions. This approach reduces model parameters by 80% and allows hardware acceleration in structured form with 0.003% accuracy loss compared to fine-tuned pruning.

• Journal of Korea Society of Industrial Information Systems
• /
• v.29 no.3
• /
• pp.1-12
• /
• 2024

In this paper, we fine-tuned KoAlpaca, a large-scale Korean language model, and implemented a blog text generation system utilizing it. Blogs on social media platforms are widely used as a marketing tool for businesses. We constructed training data of positive reviews through emotion analysis and refinement of collected sponsored blog texts and applied QLoRA for the lightweight training of KoAlpaca. QLoRA is a fine-tuning approach that significantly reduces the memory usage required for training, with experiments in an environment with a parameter size of 12.8B showing up to a 58.8% decrease in memory usage compared to LoRA. To evaluate the generative performance of the fine-tuned model, texts generated from 100 inputs not included in the training data produced on average more than twice the number of words compared to the pre-trained model, with texts of positive sentiment also appearing more than twice as often. In a survey conducted for qualitative evaluation of generative performance, responses indicated that the fine-tuned model's generated outputs were more relevant to the given topics on average 77.5% of the time. This demonstrates that the positive review generation language model for sponsored content in this paper can enhance the efficiency of time management for content creation and ensure consistent marketing effects. However, to reduce the generation of content that deviates from the category of positive reviews due to elements of the pre-trained model, we plan to proceed with fine-tuning using the augmentation of training data.