• 한국음향학회:학술대회논문집
• /
• 한국음향학회 1997년도 영남지회 학술발표회 논문집 Acoustic Society of Korean Youngnam Chapter Symposium Proceedings
• /
• pp.27-34
• /
• 1997

In this study , we tried to test the classification performance of a neural netow and thereby to examine its applicability to the signals distorted by a shallow water einvironment . We conducted an acoustic experiment iin a shallow sea near Pohang, Korea in which water depth is about 60m. The signals, on which the network has been tested, is ilinear frequency modulated ones centered on one of the frequencies, 200, 400, 600 and 800 Hz, each being swept up or down with bandwidth 100Hz. we considered two transforms, STFT(short-time Fourier transform) and PWVD (pseudo Wigner-Ville distribution), form which power spectra were derived. The training signals were simulated using an acoutic model based on the Fourier synthesis scheme. When the network has been trained on the measured signals of center frequency 600Hz,it gave a little better results than that trained onthe simulated . With the center frequencies varied, the overall performance reached over 90% except one case of center frequency 800Hz. With the feature extraction techniques(STFT and PWVD) varied,the network showed performance comparable to each other . In conclusion , the signals which have been simulated with water depth were successully applied to training a neural network, and the trained network performed well in classifying the signals distorted by a surrounding environment and corrupted by noise.

• Korean Journal of Mathematics
• /
• 제28권4호
• /
• pp.943-953
• /
• 2020

A university grades input system requires for professors to enter the normalized total scores for the letter grades and to input the scores from six fields such as Midterm, Final, Quiz which sum up to the total. All six fields have specified bounds which add up to 100. Professors should scale in the total scores to match up the letter grades and scale in every field of each student's original scores within the bounds to sum up to the scaled total score. We solve this problem by a novel design of simple shallow neural network.

• Journal of Information Processing Systems
• /
• 제17권2호
• /
• 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.

• The Journal of the Acoustical Society of Korea
• /
• 제17권1E호
• /
• pp.54-65
• /
• 1998

This study attempts to test the classifying performance of a neural network and thereby examine its applicability to the signals distorted in a shallow water environment. Linear frequency modulated(LFM) signals are simulated by using an acoustic model and also measured through sea experiment. The network is constructed to have three layers and trained on both data sets. To get normalized power spectra as feature vectors, the study considers the three transforms : shot-time Fourier transform (STFT), wavelet transform (WT) and pseudo Wigner-Ville distribution (PWVD). After trained on the simulated signals over water depth, the network gives over 95% performance with the signal to noise ratio (SNR) being up to-10 dB. Among the transforms, the PWVD presents the best performance particularly in a highly noisy condition. The network performs worse with the summer sound speed profile than with the winter profile. It is also expected to present much different performance by the variation of bottom property. When the network is trained on the measured signals, it gives a little better results than that trained on the simulated data. In conclusion, the simulated signals are successfully applied to training a network, and the trained network performs well in classifying the signals distorted by a surrounding environment and corrupted by noise.

• 한국수자원학회논문집
• /
• 제56권12호
• /
• pp.939-953
• /
• 2023

천수방정식(shallow water equations, SWE)은 물의 거동을 수치적으로 해석하기 위한 지배방정식으로 수리수문 분야에 널리 활용되고 있으며, 비선형 연립방정식으로 일반적으로 수치적으로 해석할 수 있다. 하지만 기존의 여러 수치 해석법은 격자망 생성에 민감하며 복잡한 지형에서의 해석에 한계가 발생할 수 있다. 이러한 한계점을 극복하기 위하여 본 연구에서는 물리 정보 신경망(Physics-Informed Neural Networks, PINNs)을 사용하고자 하였다. PINNs은 물리 법칙을 신경망에 직접적으로 도입하여 지배방정식을 해석하고자 하는 기법이며 지배 방정식에 대한 물리적, 수학적 정보를 손실함수로 변환하여 최적화하고 해를 산정할 수 있다. 본 연구에서는 지배방정식을 PINNs 구조 내에서 사용할 수 있도록 신경망 구조, 학습 전략, 데이터 생성 기술과 같은 포괄적인 방법론을 제시하고 결과를 ANN 기법과 비교하였다. 물리적 사전지식이 반영되지 않은 ANN과 달리 PINNs은 천수방정식에 대하여 매우 정확한 수치적 솔루션을 효과적으로 제공하는 것으로 나타났다. 따라서 PINNs은 지배방정식의 수치해석적 연구에 많은 잠재력이 있는 것으로 판단되며, 정확하고 효율적인 천수방정식의 솔루션을 위한 기법으로 활용될 수 있을 것으로 기대된다.

• Steel and Composite Structures
• /
• 제49권2호
• /
• pp.231-244
• /
• 2023

Many of the recent investigations in the field of geotechnical engineering focused on the bearing capacity theories of multilayered soil. A number of factors affect the bearing capacity of the soil, such as soil properties, applied overburden stress, soil layer thickness beneath the footing, and type of design analysis. An extensive number of finite element model (FEM) simulation was performed on a prototype slope with various abovementioned terms. Furthermore, several non-linear artificial intelligence (AI) models are developed, and the best possible neural network system is presented. The data set is from 3443 measured full-scale finite element modeling (FEM) results of a circular shallow footing analysis placed on layered cohesionless soil. The result is used for both training (75% selected randomly) and testing (25% selected randomly) the models. The results from the predicted models are evaluated and compared using different statistical indices (R² and RMSE) and the most accurate model BBO (R²=0.9481, RMSE=4.71878 for training and R²=0.94355, RMSE=5.1338 for testing) and TLBO (R²=0.948, RMSE=4.70822 for training and R²=0.94341, RMSE=5.13991 for testing) are presented as a simple, applicable formula.

• 한국수자원학회:학술대회논문집
• /
• 한국수자원학회 2023년도 학술발표회
• /
• pp.171-171
• /
• 2023

This study investigates the capability of Physics-Informed Neural Networks (PINNs) for solving the solution of partial differential equations. Particularly, the 1D Saint-Venant Equations (SVEs) were considered, which describe the movement of water in a domain with shallow depth compared to its horizontal extent, and are widely adopted in hydrodynamics, river, and coastal engineering. The core contribution of this work is to combine the robustness of neural networks with the physical constraints of the SVEs. The PINNs method utilized a neural network to approximate the solutions of SVEs, while also enforcing the underlying physical principles of the equations. This allows for a more effective and reliable solution, especially in areas with complex geometry and varying bathymetry. To validate the robustness of the PINNs method, numerical experiments were conducted on several benchmark problems. The results show that the PINNs could be achieved high accuracy when compared with the solution from the numerical solution. Overall, this study demonstrates the potential of using PINNs and highlights the benefits of integrating neural network and physics information for improved efficiency and accuracy in solving SVEs.

• 한국정보기술학회논문지
• /
• 제16권11호
• /
• pp.11-21
• /
• 2018

본 논문에서는 사람을 대신하여 분류, 예측 하는 딥러닝 기술을 활용하여 목소리를 통해 남녀노소를 분류하는 연구를 수행한다. 연구과정은 기존 신경망 기반의 사운드 분류 연구를 살펴보고 목소리 분류를 위한 개선된 신경망을 제안한다. 기존 연구에서는 도시 데이터를 이용해 사운드를 분류하는 연구를 진행하였으나, 얕은 신경망으로 인한 성능 저하가 나타났으며 다른 소리 데이터에 대해서도 좋은 성능을 보이지 못했다. 이에 본 논문에서는 목소리 데이터를 전처리하여 특징값을 추출한 뒤 추출된 특징값을 기존 사운드 분류 신경망과 제안하는 신경망에 입력하여 목소리를 분류하고 두 신경망의 분류 성능을 비교 평가한다. 본 논문의 신경망은 망을 더 깊고 넓게 구성함으로써 보다 개선된 딥러닝 학습이 이루어지도록 하였다. 성능 결과로는 기존 연구와 본 연구의 신경망에서 각각 84.8%, 91.4%로 제안하는 신경망에서 약 6% 더 높은 정확도를 보였다.

• 한국암반공학회:학술대회논문집
• /
• 한국암반공학회 2008년도 국제학술회의
• /
• pp.43-51
• /
• 2008

Currently an increasing number of urban tunnels with small overburden are excavated according to the principle of the New Austrian Tunneling Method (NATM). For rational management of tunnels from planning to construction and maintenance stages, prediction, control and monitoring of displacements of and around the tunnel have to be performed with high accuracy. Computational method tools, such as finite element method, have been and are indispensable tool for tunnel engineers for many years. It is, however, a commonly acknowledged fact that determination of input parameters, especially material properties exhibiting nonlinear stress-strain relationship, is not an easy task even for an experienced engineer. Use and application of the acquired tunnel information is important for prediction accuracy and improvement of tunnel behavior on construction. Artificial Neural Network (ANN) model is a form of artificial intelligence that attempts to mimic behavior of human brain and nervous system. The main objective of this paper is to perform the deformation analysis in NATM tunnel by means of numerical simulation and artificial neural network (ANN) with field database. Developed ANN model can achieve a high level of prediction accuracy.

• 한국수자원학회:학술대회논문집
• /
• 한국수자원학회 2018년도 학술발표회
• /
• pp.157-157
• /
• 2018

A reliable and accurate downscaling model which can provide climate change information, obtained from global climate models (GCMs), at finer resolution has been always of great interest to researchers. In order to achieve this model, linear methods widely have been studied in the past decades. However, nonlinear methods also can be potentially beneficial to solve downscaling problem. Therefore, this study explored the applicability of some nonlinear machine learning techniques such as neural network (NN), extreme learning machine (ELM), and ELM autoencoder (ELM-AE) as well as a linear method, least absolute shrinkage and selection operator (LASSO), to build a reliable temperature downscaling model. ELM is an efficient learning algorithm for generalized single layer feed-forward neural networks (SLFNs). Its excellent training speed and good generalization capability make ELM an efficient solution for SLFNs compared to traditional time-consuming learning methods like back propagation (BP). However, due to its shallow architecture, ELM may not capture all of nonlinear relationships between input features. To address this issue, ELM-AE was tested in the current study for temperature downscaling.