• Journal of Internet Computing and Services
• /
• v.18 no.6
• /
• pp.25-34
• /
• 2017

Autoencoder is a type of deep learning method where input layer and output layer are the same, and effectively extracts and restores characteristics of input vector using constraints of hidden layer. In this paper, we propose methods of Autoencoders to remove a natural background image which is a noise to the CAPTCHA and recover only a numerical images by applying various autoencoder models to a region where one number of CAPTCHA images and a natural background are mixed. The suitability of the reconstructed image is verified by using the softmax function with the output of the autoencoder as an input. And also, we compared the proposed methods with the other method and showed that our methods are superior than others.

• Journal of Korea Multimedia Society
• /
• v.24 no.10
• /
• pp.1414-1424
• /
• 2021

In this paper, the importance of input factors of a DNN (Deep Neural Network) PM_2.5 forecasting model using LRP(Layer-wise Relevance Propagation) is analyzed, and forecasting performance is improved. Input factor importance analysis is performed by dividing the learning data into time and PM_2.5 concentration. As a result, in the low concentration patterns, the importance of weather factors such as temperature, atmospheric pressure, and solar radiation is high, and in the high concentration patterns, the importance of air quality factors such as PM_2.5, CO, and NO₂ is high. As a result of analysis by time, the importance of the measurement factors is high in the case of the forecast for the day, and the importance of the forecast factors increases in the forecast for tomorrow and the day after tomorrow. In addition, date, temperature, humidity, and atmospheric pressure all show high importance regardless of time and concentration. Based on the importance of these factors, the LRP_DNN prediction model is developed. As a result, the ACC(accuracy) and POD(probability of detection) are improved by up to 5%, and the FAR(false alarm rate) is improved by up to 9% compared to the previous DNN model.

• Geomechanics and Engineering
• /
• v.34 no.6
• /
• pp.665-681
• /
• 2023

Liquefaction is one of the most devastating geotechnical phenomena that severely damage vital structures and lifelines. Before constructing structures on problematic ground, it is necessary to improve the site and solve the geotechnical problem. Among ground improvement methods dealing with liquefaction, gravel drain (GD) columns and deep soil mixing (DSM) columns are popular. In this study, the results of a series of seismic experiments in a 1g environment on a structure located over liquefiable ground with different thicknesses reinforced with GD and DSM techniques were presented. The dynamic response of the reinforced ground system was investigated based on the parameters of subsidence rate, excess pore water pressure ratio, and maximum acceleration. The time history of the input acceleration was applied harmonically with an acceleration range of 0.2g and at frequencies of 1, 2, and 3 Hz. The results show that the thickness of the liquefiable layer and the frequency of the input motion have a significant impact on the effectiveness of the improvement method and all responses. Among the two techniques used, DSM in thick liquefied layers was much more efficient than GD in controlling the subsidence and rupture of the soil under the foundation. Maximum settlement values, settlement rate, and foundation rotation in the thicker liquefied layer at the 1-Hz input frequency were higher than at other frequencies. At low thicknesses, the dynamic behavior of the GD was closer to that of the DSM.

• Nutrition Research and Practice
• /
• v.17 no.6
• /
• pp.1255-1266
• /
• 2023

BACKGROUND/OBJECTIVES: This study aimed to predict the association between nutritional intake and diabetes mellitus (DM) by developing an artificial neural network (ANN) model for older adults. SUBJECTS/METHODS: Participants aged over 65 years from the 7th (2016-2018) Korea National Health and Nutrition Examination Survey were included. The diagnostic criteria of DM were set as output variables, while various nutritional intakes were set as input variables. An ANN model comprising one input layer with 16 nodes, one hidden layer with 12 nodes, and one output layer with one node was implemented in the MATLAB^® programming language. A sensitivity analysis was conducted to determine the relative importance of the input variables in predicting the output. RESULTS: Our DM-predicting neural network model exhibited relatively high accuracy (81.3%) with 11 nutrient inputs, namely, thiamin, carbohydrates, potassium, energy, cholesterol, sugar, vitamin A, riboflavin, protein, vitamin C, and fat. CONCLUSIONS: In this study, the neural network sensitivity analysis method based on nutrient intake demonstrated a relatively accurate classification and prediction of DM in the older population.

• Proceedings of the Korean Operations and Management Science Society Conference
• /
• 2004.05a
• /
• pp.52-55
• /
• 2004

VMS (virtual manufacturing system) may be defined as a transparent interface/control mechanism to support human decision-making via simulation and monitoring of real operating situation through modeling of all activities in RMS (real manufacturing system). The three main layers in VMS are business process layer, manufacturing execution layer, and facility operation layer, and each layer is represented by a specific software system having its own input modeler module. The current version of these input modelers has been implemented based on its own 'local' framework, and as a result, there are no information sharing mechanism, nor a common user view among them. Proposed in this paper is a unified modeler framework covering the three VMS layers, in which the concept of PPR (product-process-resource) model is employed as a common semantics framework and a 2D graphic network model is used as a syntax framework. For this purpose, abstract class PPRObject and GraphicObject are defined and then a subclass is inherited from the abstract class for each application layer. This feature would make it easier to develop and maintain the individual software systems. For information sharing, XML is used as a common data format.

• Journal of Environmental Impact Assessment
• /
• v.12 no.4
• /
• pp.271-279
• /
• 2003

Temperature and humidity forecasting have been performed using artificial neural networks model(ANN). We composed ANN with multi-layer perceptron which is 2 input layers, 2 hidden layers and 1 output layer. Back propagation algorithm was used to train the ANN. 6 nodes and 12 nodes in the middle layers were appropriate to the temperature model for training. And 9 nodes and 6 nodes were also appropriate to the humidity model respectively. 90% of the all data was used learning set, and the extra 10% was used to model verification. In the case of temperature, average temperature before 15 minute and humidity at present constituted input layer, and temperature at present constituted out-layer and humidity model was vice versa. The sensitivity analysis revealed that previous value data contributed to forecasting target value than the other variable. Temperature was pseudo-linearly related to the previous 15 minute average value. We confirmed that ANN with multi-layer perceptron could support pollutant dispersion model by computing meterological data at real time.

• The Korean Journal of Health Service Management
• /
• v.5 no.1
• /
• pp.1-13
• /
• 2011

The purpose of this paper was to evaluate performance of information system for one national university hospital in order to identify the factors influencing performance of information system. KPIs were collected for 181 users of information system (41 doctors, 104 nurses, and 11 medical supporting staffs, and 25 administrative staffs) from August 10 to 24, 2010. The results were as follows: Average performance score for input layer was 3.16; average performance score for process layer was 3.35; and average performance score for business layer was 3.57. Scores for input layer was lowest for nurses and scores for process and business layer were lowest for doctors. Results from the path analysis showed that system quality, demographic characteristics, and security significantly influenced management process but these factors except demographic characteristics influenced user satisfaction; and management process also significantly influenced user satisfaction.

• Earthquakes and Structures
• /
• v.16 no.1
• /
• pp.109-118
• /
• 2019

This study, analyses the seismic response for a rigid massless square foundation resting on a viscoelastic soil layer limited by rigid bedrock. The foundation is subjected either to externally applied forces or to obliquely incident seismic body or surface harmonic seismic waves P, SV and SH. A 3-D frequency domain BEM formulation in conjunction with the thin layer method (TLM) is adapted here for the solution of elastodynamic problems and used for obtained the seismic response. The mathematical approach is based on the method of integral equations in the frequency domain using the formalism of Green's functions (Kausel and Peck 1982) for layered soil, the impedance functions are calculated by the compatibility condition. In this study, The key step is the characterization of the soil-foundation interaction with the input motion matrix. For each frequency the impedance matrix connects the applied forces to the resulting displacement, and the input motion matrix connects the displacement vector of the foundation to amplitudes of the free field motion. This approach has been applied to analyze the effect of soil-structure interaction on the seismic response of the foundation resting on a viscoelastic soil layer limited by rigid bedrock.

• Proceedings of the Acoustical Society of Korea Conference
• /
• 1993.06a
• /
• pp.117-120
• /
• 1993

In this paper, we explore the possibility of URAN(Universally Reconstructable Artificial Neural-network) VLSI chip for speech recognition. URAN, a newly developed analog-digital hybrid neural chip, is discussed in respects to its input, output, and weight accuracy and their relations to its performance on speaker independent digit recognition. Multi-layer perceptron(MLP) nets including a large frame input layer are used to recognize a digit syllable at a forward retrieval. The simulation results using the full and limited floating precision computations for the input, output, and weight variables of the network give the comparable classification performance. An MLP with piecewise linear hidden and output units is also trained successfully using low accuracy computation.

• Proceedings of the KIEE Conference
• /
• 1989.07a
• /
• pp.166-169
• /
• 1989

A study for the recognition of the Korean characters by a neural network is presented. To reduce the dimension of the input image data, DC components are extracted from each input image and used as input to the neural net. A multi-layer perceptron with one hidden layer was trained with back-error propagation training algorithm. Its performance is tested for 24 ${\times}$ 24 binary images of Korean characters and the results of several experiments are presented.