A5101397414- Neural Networks and Applications
- Model Reduction and Neural Networks
- Electronic Packaging and Soldering Technologies
- Advanced Frequency and Time Standards
- Advanced Radiotherapy Techniques
- Indoor and Outdoor Localization Technologies
- Stochastic Gradient Optimization Techniques
- Network Time Synchronization Technologies
- Machine Learning and ELM
- Sparse and Compressive Sensing Techniques
- Plasmonic and Surface Plasmon Research
- Advanced Biosensing Techniques and Applications
- Target Tracking and Data Fusion in Sensor Networks
- Advancements in PLL and VCO Technologies
- Tropical and Extratropical Cyclones Research
- Advanced Welding Techniques Analysis
- Neural Networks and Reservoir Computing
- Advanced Multi-Objective Optimization Algorithms
- Radiation Therapy and Dosimetry
- Face and Expression Recognition
- Advanced Neural Network Applications
- Anaerobic Digestion and Biogas Production
- Wastewater Treatment and Nitrogen Removal
- Speech and Audio Processing
- Lung Cancer Diagnosis and Treatment
Yonsei University
2023-2024
John Brown University
2024
Samsung (South Korea)
2024
Korea Advanced Institute of Science and Technology
2022-2023
National Institute for Mathematical Sciences
2022
Catholic University of Korea
2013-2017
Korea Atomic Energy Research Institute
2017
Uijeongbu St. Mary's Hospital
2017
Seoul St. Mary's Hospital
2013-2016
The University of Texas at Austin
2012-2015
This study aims to improve the accuracy and interpretability of flood susceptibility mapping (FSM) by integrating automated machine learning (AutoML) with explainable artificial intelligence (XAI) frameworks. Using spatial data from Seoul, South Korea, ten topographic environmental factors were selected as model inputs. Initially, Tree-based Pipeline Optimization Tool (TPOT), an evolutionary AutoML algorithm, was employed construct baseline models using Gradient Boosting (GB), Random Forest...
In this work, we demonstrate a novel approach to superior multilevel-cell (MLC) ferroelectric field-effect transistor (FeFET) with large memory window (MW) and negligible <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${V}_{T}$ </tex-math></inline-formula> variation toward MLC operation. We realized high ferroelectricity in relatively thick HZO (FE) layer for FeFET MW [MW notation="LaTeX">$\propto $...
We demonstrate efficient optical directional beaming using an array of sub-wavelength patterns on a metallic surface. Specifically, sized slit placed next to periodic grating is designed and optimized realize maximum coupling efficiency radiation into leaky-wave plasmonic mode. Collective scattering from the corrugations forming synthesized radiate towards desired direction, achieved through tailoring design parameters with simple analytical model. also prove that directivity can be further...
In radiation therapy, accurate radiotherapy treatment plan (RTP) reproduction is necessary to optimize the clinical results. Thus, attempts have recently been made ensure high RTP reproducibility using image-guide therapy (IGRT) technology. However, use of digital X-ray equipment requires extended quality assurance (QA) for those devices, since IGRT device determines precision intensity-modulated therapy. The study described in this paper was focused on developing a multi-energy PbO...
The control rod assembly controls reactor power by adjusting its position during normal operation and shuts down chain reactions free drop under scram conditions. Therefore, the performance of is important for safety a nuclear reactor. In this study, conceptually designed prototype generation IV sodium-cooled fast that being developed at Korea Atomic Energy Research Institute as next-generation was experimentally investigated. For test, test facility procedure were established first, several...
Extreme learning machine (ELM) is a methodology for solving partial differential equations (PDEs) using single hidden layer feed-forward neural network. It presets the weight/bias coefficients in with random values, which remain fixed throughout computation, and uses linear least squares method training parameters of output known to be much faster than Physics informed networks. However, classical ELM still computationally expensive when high level representation desired solution as this...
We present the design and integration of a two-dimensional (2D) plasmonic nanogratings structure on electrode colloidal quantum dot-based light-emitting diodes (QDLEDs) as compact light source towards arrayed on-chip imaging tumor cells. Colloidal dots (QDs) were used emission layer due to their unique capabilities, including multicolor emission, narrow bandwidth, tunable wavelengths, compatibility with silicon fabrication. The nanograting, based metal-dielectric-metal waveguide, aims...
기상청은 적설하중을 계측하지 않기 때문에 눈 단위중량을 고려하여 심적설을 적설하중으로 환산하는 방식이 필요하다. 세계 각국의 건축물구조기준에서는 서로 다른 이용하여 환산하고 있었다. 본 연구에서는 강수량, 신적설 및 일평균풍속을 신적설의 산정하였다. 또한 미계측 지점의 보수적으로 추정하기 위하여 최대가능적설하중 산정 방안을 제안하였다. 최대가능적설하중의 정밀도를 개선하기 위해서는 국지적 분포 변화를 고려할 수 있는 방안이 필요하였다. 최대가능적설하중을 원보험용 적설재해지도를 작성하였다. 동해안과 서해안 지역의 재해등급이 높은 특성이 나타나며, 우리나라 대부분 재해등급은 매우 낮은 것으로 나타났다. Korea needs to take snow unit weight into account transform depth loads because Korean Meteorological Agency does not gauge loads. In world wide building...
As deep neural networks (DNNs) become deeper, the training time increases. In this perspective, multi-GPU parallel computing has a key tool in accelerating of DNNs. article, we introduce novel methodology to construct network that can utilize multiple GPUs simultaneously from given DNN. We observe layers DNN be interpreted as steps time-dependent problem and parallelized by emulating parallel-in-time algorithm called parareal. The parareal consists fine structures which implemented coarse...
본 논문에서는 발신자가 송출한 신호를 이용하여 TDOA(Time Difference of Arrival) 방식으로 발신자의 위치를 추정할 때, BLUE(Best Linear Unbiased Estimator) 추정기와 이의 CRLB(Crammer-Rao Lower Bound)를 닫힌 해 형태로 구하였다. 3 개의 기준국 또는 센서를 사용하여 2차원의 발신자 BLUE 추정기를 CRLB를 구하기 위해서 위치에 대한 기준 설정한 후 이를 1 차 Taylor 급수로부터 유도된 근사회된 TDOA 쌍곡선 방정식을 사용하였다. 논문에서 근사화를 통해 구해진 유도식은 각 센서에서의 TOA(Time 측정 잡음이 서로 상관관계가 없고 독립적이라는 가정하에서, 백색 가우시안 잡음에 대해서뿐만 아니라 평균이 제로인 모든 대해서 적용할 수 있다. In this paper, we derived a closed-form equation Best Estimator (BLUE) and its...
본 연구에서는 적충 구조를 이용하여 누설전류를 저감 시키는 기술을 적용하여 입자침전법을 이용한 방사선 영상 센서의 변환 물질을 개발하였다. 이는 디지털 검출기의 두 가지 방식 중 하나인 직접방식에 사용되는 핵심 소자로 기존의 비정질셀레늄(a-Se)을 대체하여 더욱 효율이 높은 후보 물질들이 연구되어지는 가운데 태양전지와 반도체 분야에서 이미 많이 사용되어온 이종접합을 이용해 누설 전류를 시키는데 그 목적이 있다. 연구에서 입자 침전법 제조방법은 검출 물질 제작이 용이하고 수율과 대면적의 검출기 제작에 적합하나 전류가 의료 영상에 있어서 문제가 되어 오고 이러한 단점을 보완하기 위해 저감시킨다면 PIB법을 간편하게 향상된 효율의 검출기를 제작 할 수 있다고 사료 진다. 전류와 민감도에 대한 전기적 신호를 측정하여 제작된 구조의 물질의 특성 평가가 이루어 졌다. In this study, the basic research verifying possibility of...
The increasing demand for higher memory capacity has led to a greater emphasis on high-density DIMM such as 3-dimension stacked (3DS) 128GB products compared the current mainstream 64GB DIMM. However, adoption of 3DS-DDR5 necessitates use technologies like through-silicon via (TSV), which can lead potential drawbacks price escalation and additional read latency. To overcome these challenges, there exists need explore development high-capacity using monolithic die approach. Furthermore, high...
We introduce a novel two-level overlapping additive Schwarz preconditioner for accelerating the training of scientific machine learning applications. The design proposed is motivated by nonlinear preconditioner. neural network parameters are decomposed into groups (subdomains) with regions. In addition, network's feed-forward structure indirectly imposed through subdomain-wise synchronization strategy and coarse-level step. Through series numerical experiments, which consider...
<title>Abstract</title> The performance of neural networks has been significantly improved by increasing the number channels in convolutional layers. However, this increase comes with a higher computational cost, resulting numerous studies focused on reducing it. One promising approach to address issue is group convolution, which effectively reduces cost grouping channels. best our knowledge, there no theoretical analysis how well convolution approximates standard convolution. In paper, we...
To understand the population changes of Mongolian racerunner (Eremias argus) inhabiting Baramarae Beach SPA (Special Protection Area) in Taeanhaean National Park, monitoring was conducted May, August, and October from 2009 to 2023. The results showed that racerunners this area numbered between 48 80 individuals, their numbers did not differ significantly seasons. predict fluctuations over next five years, moving average (MA) ARIMA analyses were performed. indicated number likely remain...
We design two classes of ultra-fast meta-solvers for linear systems arising after discretizing PDEs by combining neural operators with either simple iterative solvers, e.g., Jacobi and Gauss-Seidel, or Krylov methods, GMRES BiCGStab, using the trunk basis DeepONet as a coarse preconditioner. The idea is to leverage spectral bias networks account lower part spectrum in error distribution while upper handled easily inexpensively relaxation methods fine-scale preconditioners. create pareto...
The effect of flip chip bonding parameters on formation intermetallic compounds (IMCs) between Au stud bumps and Sn-3.5Ag solder was investigated. In this study, temperature performed at <TEX>$260^{\circ}C$</TEX> <TEX>$300^{\circ}C$</TEX> with various times 5, 10, 20 sec. AuSn, <TEX>$AuSn_2$</TEX> <TEX>$AuSn_4$</TEX> IMCs were formed the interface joints (Au, Cu)<TEX>$_6Sn_5$</TEX> IMC observed near Cu pad side in joint. At <TEX>$260^{\circ}C$</TEX>, dominant joint compared to other Au-Sn as...