• The Korean Journal of Applied Statistics
• /
• v.32 no.5
• /
• pp.693-702
• /
• 2019

Deep learning has gained popularity for the classification and prediction task. Neural network layers become deeper as more data becomes available. Saturation is the phenomenon that the gradient of an activation function gets closer to 0 and can happen when the value of weight is too big. Increased importance has been placed on the issue of saturation which limits the ability of weight to learn. To resolve this problem, Glorot and Bengio (Proceedings of the Thirteenth International Conference on Artificial Intelligence and Statistics, 249-256, 2010) claimed that efficient neural network training is possible when data flows variously between layers. They argued that variance over the output of each layer and variance over input of each layer are equal. They proposed a method of initialization that the variance of the output of each layer and the variance of the input should be the same. In this paper, we propose a new method of establishing initialization by adopting truncated normal distribution and truncated cauchy distribution. We decide where to truncate the distribution while adapting the initialization method by Glorot and Bengio (2010). Variances are made over output and input equal that are then accomplished by setting variances equal to the variance of truncated distribution. It manipulates the distribution so that the initial values of weights would not grow so large and with values that simultaneously get close to zero. To compare the performance of our proposed method with existing methods, we conducted experiments on MNIST and CIFAR-10 data using DNN and CNN. Our proposed method outperformed existing methods in terms of accuracy.

• Journal of Intelligence and Information Systems
• /
• v.28 no.3
• /
• pp.185-207
• /
• 2022

Recently, with the development of deep learning technology, researches to apply a deep learning algorithm to analyze unstructured data such as text and images are being actively conducted. Text classification has been studied for a long time in academia and industry, and various attempts are being performed to utilize data characteristics to improve classification performance. In particular, a hierarchical relationship of labels has been utilized for hierarchical classification. However, the top-down approach mainly used for hierarchical classification has a limitation that misclassification at a higher level blocks the opportunity for correct classification at a lower level. Therefore, in this study, we propose a methodology for classifying hierarchical data using the autoencoder-based deeply supervised network that high-level classification does not block the low-level classification while considering the hierarchical relationship of labels. The proposed methodology adds a main classifier that predicts a low-level label to the autoencoder's latent variable and an auxiliary classifier that predicts a high-level label to the hidden layer of the autoencoder. As a result of experiments on 22,512 academic papers to evaluate the performance of the proposed methodology, it was confirmed that the proposed model showed superior classification accuracy and F1-score compared to the traditional supervised autoencoder and DNN model.

• Journal of the Korea Convergence Society
• /
• v.9 no.5
• /
• pp.1-6
• /
• 2018

In smart grid an accurate load forecasting is crucial in planning resources, which aids in improving its operation efficiency and reducing the dynamic uncertainties of energy systems. Research in this area has included the use of shallow neural networks and other machine learning techniques to solve this problem. Recent researches in the field of computer vision and speech recognition, have shown great promise for Deep Neural Networks (DNN). To improve the performance of daily electric peak load forecasting the paper presents a new deep neural network model which has the architecture of two multi-layer neural networks being serially connected. The proposed network model is progressively pre-learned layer by layer ahead of learning the whole network. For both one day and two day ahead peak load forecasting the proposed models are trained and tested using four years of hourly load data obtained from the Korea Power Exchange (KPX).

• The Journal of the Acoustical Society of Korea
• /
• v.36 no.4
• /
• pp.261-266
• /
• 2017

In this paper, we propose a vocal and nonvocal separation method which uses a combination of kernel model and LSTM (Long-Short Term Memory) networks. Conventional vocal and nonvocal separation methods estimate the vocal component even in sections where only non-vocal components exist. This causes a problem of the source estimation error. Therefore we combine the existing kernel based separation method with the vocal/nonvocal classification based on LSTM networks in order to overcome the limitation of the existing separation methods. We propose a parallel combined separation algorithm and series combined separation algorithm as combination structures. The experimental results verify that the proposed method achieves better separation performance than the conventional approaches.

• The Journal of the Korea Contents Association
• /
• v.21 no.10
• /
• pp.1-10
• /
• 2021

Log data accumulated in the Learning Management System (LMS) provide high-quality information for the learning process of students. Until now, various studies have been conducted to predict students' academic achievement using LMS log data. However, previous studies were based on relatively small sample sizes of students and courses, limiting the possibility of generalization. This study developed and validated a deep neural network model for the early prediction of academic achievement of college students using massive LMS log data. To this end, we used 78,466,385 cases of LMS log data and 165,846 cases of grade data. The proposed model predicted the excellent-grade students with a high level of accuracy from the beginning of the semester. Meanwhile, the prediction accuracy for the moderate and underachieving groups was relatively low, but the accuracy improved as the time points of the prediction were delayed. This study is meaningful in that we developed an early prediction model based on a deep neural network with sufficient accuracy for practical utilization by only using LMS log data.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.26 no.4
• /
• pp.503-509
• /
• 2022

The purpose of this study is to set the rate of change between the market price of the next day and the previous day to be predicted as the predicted value, and the market price for each section is generated by dividing the stock price ranking of the next day to be predicted at regular intervals, which is different from the previous papers that predict the market price. We would like to propose a new time series data prediction method that predicts the market price change rate of the final next day through a model using the rate of change as the predicted value. The change in the performance of the model according to the degree of subdivision of the predicted value and the type of input data was analyzed.

• Journal of the Korea institute for structural maintenance and inspection
• /
• v.24 no.3
• /
• pp.18-25
• /
• 2020

The unconfined compressive strength of lightweight treated soils strongly depends on mixing ratio. To characterize the relation between various LTS components and the unconfined compressive strength of LTS, extensive studies have been conducted, proposing normalized factor using regression models based on their experimental results. However, these results obtained from laboratory experiments do not expect consistent prediction accuracy due to complicated relation between materials and mix proportions. In this study, deep neural network model(Deep-LTS), which was based on experimental test results performed on various mixing conditions, was applied to predict the unconfined compressive strength. It was found that the unconfined compressive strength LTS at a given mixing ratio could be resonable estimated using proposed Deep-LTS.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2021.10a
• /
• pp.335-337
• /
• 2021

Flood damage is occurring all over the world, and the number of people living in flood-prone areas reached 86 million, a 25% increase compared to 2000. These floods cause enormous damage to life and property, and it is essential to decide on an appropriate evacuation in order to reduce the damage. Evacuation in anticipation of a flood also incurs a lot of cost, and if an evacuation is not performed due to an error in the flood prediction, a greater cost is incurred. Therefore, in this paper, we propose a flood risk determination model using the CNN model to enable evacuation at an appropriate time by using the time series data of precipitation and water level. Through this, it is thought that it can be utilized as an initial study to determine the time of flood evacuation to prevent unnecessary evacuation and to ensure that evacuation can be carried out at an appropriate time.

• Proceedings of the Korean Institute of Navigation and Port Research Conference
• /
• 2023.05a
• /
• pp.104-105
• /
• 2023

해양 관측과 위성 원격탐사를 이용하여 시공간적으로 다양하게 변하는 생태 어장 환경 및 선박 관련 자료를 획득할 수 있다. 이번 연구의 주요 목적은 야간 불빛 위성 자료를 이용하여 광범위한 해역에 대한 어선의 위치 분포를 파악하는 딥러닝 기반 모델을 제안하는 것이다. 제안한 모델의 정확성을 평가하기 위해 야간 조업 어선의 위치를 포함하고 있는 AIS(Automatic Identification System) 정보와 상호 비교 평가 하였다. 이를 위해, 먼저 AIS 자료를 획득 및 분석하는 방법을 소개한다. 해양안전종합시스템(General Information Center on Maritime Safety & Security, GICOMS)으로부터 제공받은 AIS 자료는 동적정보와 정적정보로 나뉜다. 동적 정보는 일별 자료로 구분되어있으며, 이 정보에는 해상이동업무식별번호(Maritime Mobile Service Identity, MMSI), 선박의 시간, 위도, 경도, 속력(Speed over Ground, SOG), 실침로(Course over Ground, COG), 선수방향(Heading) 등이 포함되어 있다. 정적정보는 1개의 파일로 구성되어 있으며, 선박명, 선종 코드, IMO Number, 호출부호, 제원(DimA, DimB, DimC, Dim D), 홀수, 추정 톤수 등이 포함되어 있다. 이번 연구에서는 선박의 정보에서 어선의 정보를 추출하여 비교 자료로 사용하였으며, 위성 자료는 구름의 영향이 없는 깨끗한 날짜의 영상 자료를 선별하여 사용하였다. 야간 불빛 위성 자료, 구름 정보 등을 이용하여 야간 조업 어선의 불빛을 감지하는 심층신경망(Deep Neural Network; DNN) 기반 모델을 제안하였다. 본 연구의결과는 야간 어선의 분포를 감시하고 한반도 인근 어장을 보호하는데 기여할 것으로 기대된다.

• Journal of IKEEE
• /
• v.24 no.4
• /
• pp.1086-1092
• /
• 2020

An autonomous agent for real world should be able to recognize text in scenes. With the advancement of deep learning, various DNN models have been utilized for transformation, feature extraction, and predictions. However, the existing state-of-the art STR (Scene Text Recognition) engines do not achieve the performance required for real world applications. In this paper, we introduce a performance-improvement method through an add-on composed of an OCR (Optical Character Recognition) engine and a classifier for STR engines. On instances from IC13 and IC15 datasets which a STR engine failed to recognize, our method recognizes 10.92% of unrecognized characters.