• Smart Media Journal
• /
• v.11 no.3
• /
• pp.62-73
• /
• 2022

Researches on the prediction of domestic apartment sales price have been continuously conducted, but it is not easy to accurately predict apartment prices because various characteristics are compounded. Prior to predicting apartment sales price, the analysis of major factors, influencing on sale prices, is of paramount importance to improve the accuracy of sales price. Therefore, this study aims to analyze what are the factors that affect the apartment sales price in Gwangju, which is currently showing a steady increase rate. With 6 years of Gwangju apartment transaction price and various social factor data, several maching learning techniques such as multiple regression analysis, random forest, and deep artificial neural network algorithms are applied to identify major factors in each model. The performances of each model are compared with RMSE (Root Mean Squared Error), MAE (Mean Absolute Error) and R² (coefficient of determination). The experiment shows that several factors such as 'contract year', 'applicable area', 'certificate of deposit', 'mortgage rate', 'leading index', 'producer price index', 'coincident composite index' are analyzed as main factors, affecting the sales price.

• Journal of the Korean Geosynthetics Society
• /
• v.21 no.2
• /
• pp.21-30
• /
• 2022

Shear strength is the most important indicator in the evaluation of rock slope stability. It is generally estimated by comparing the results of existing literature data, back analysis, experiments and etc. There are additional variables related to the state of discontinuity to consider in the shear strength of the rock slope. It is difficult to determine whether these variables exist through drilling, and it is also difficult to find an exact relationship with shear strength. In this study, the data calculated through back analysis were used. The relationship between previously considered variables was applied to deep learning and the possibility for estimating shear strength of rock slope was explored. For comparison, an existing simple linear regression model and a deep learning algorithm, a deep neural network(DNN) model, were used. Although each analysis model derived similar prediction results, the explanatory power of DNN was improved with a small differences.

• Journal of Intelligence and Information Systems
• /
• v.22 no.2
• /
• pp.127-142
• /
• 2016

Deep learning model is a kind of neural networks that allows multiple hidden layers. There are various deep learning architectures such as convolutional neural networks, deep belief networks and recurrent neural networks. Those have been applied to fields like computer vision, automatic speech recognition, natural language processing, audio recognition and bioinformatics where they have been shown to produce state-of-the-art results on various tasks. Among those architectures, convolutional neural networks and recurrent neural networks are classified as the supervised learning model. And in recent years, those supervised learning models have gained more popularity than unsupervised learning models such as deep belief networks, because supervised learning models have shown fashionable applications in such fields mentioned above. Deep learning models can be trained with backpropagation algorithm. Backpropagation is an abbreviation for "backward propagation of errors" and a common method of training artificial neural networks used in conjunction with an optimization method such as gradient descent. The method calculates the gradient of an error function with respect to all the weights in the network. The gradient is fed to the optimization method which in turn uses it to update the weights, in an attempt to minimize the error function. Convolutional neural networks use a special architecture which is particularly well-adapted to classify images. Using this architecture makes convolutional networks fast to train. This, in turn, helps us train deep, muti-layer networks, which are very good at classifying images. These days, deep convolutional networks are used in most neural networks for image recognition. Convolutional neural networks use three basic ideas: local receptive fields, shared weights, and pooling. By local receptive fields, we mean that each neuron in the first(or any) hidden layer will be connected to a small region of the input(or previous layer's) neurons. Shared weights mean that we're going to use the same weights and bias for each of the local receptive field. This means that all the neurons in the hidden layer detect exactly the same feature, just at different locations in the input image. In addition to the convolutional layers just described, convolutional neural networks also contain pooling layers. Pooling layers are usually used immediately after convolutional layers. What the pooling layers do is to simplify the information in the output from the convolutional layer. Recent convolutional network architectures have 10 to 20 hidden layers and billions of connections between units. Training deep learning networks has taken weeks several years ago, but thanks to progress in GPU and algorithm enhancement, training time has reduced to several hours. Neural networks with time-varying behavior are known as recurrent neural networks or RNNs. A recurrent neural network is a class of artificial neural network where connections between units form a directed cycle. This creates an internal state of the network which allows it to exhibit dynamic temporal behavior. Unlike feedforward neural networks, RNNs can use their internal memory to process arbitrary sequences of inputs. Early RNN models turned out to be very difficult to train, harder even than deep feedforward networks. The reason is the unstable gradient problem such as vanishing gradient and exploding gradient. The gradient can get smaller and smaller as it is propagated back through layers. This makes learning in early layers extremely slow. The problem actually gets worse in RNNs, since gradients aren't just propagated backward through layers, they're propagated backward through time. If the network runs for a long time, that can make the gradient extremely unstable and hard to learn from. It has been possible to incorporate an idea known as long short-term memory units (LSTMs) into RNNs. LSTMs make it much easier to get good results when training RNNs, and many recent papers make use of LSTMs or related ideas.

• Annual Conference on Human and Language Technology
• /
• 2018.10a
• /
• pp.532-536
• /
• 2018

심층인공신경망을 이용한 대화 모델링 연구가 활발하게 진행되고 있다. 본 논문에서는 대화에서 발화의 감정과 화행을 분류하기 위해 멀티태스크(multitask) 학습을 이용한 End-to-End 시스템을 제안한다. 우리는 감정과 화행을 동시에 분류하는 시스템을 개발하기 위해 멀티태스크 학습을 수행한다. 또한 불균형 범주 분류를 위해 계단식분류(cascaded classification) 구조를 사용하였다. 일상대화 데이터셋을 사용하여 실험을 수행하였고 macro average precision으로 성능을 측정하여 감정 분류 60.43%, 화행 분류 74.29%를 각각 달성하였다. 이는 baseline 모델 대비 각각 29.00%, 1.54% 향상된 성능이다. 본 논문에서는 제안하는 구조를 이용하여, 발화의 감정 및 화행 분류가 End-to-End 방식으로 모델링 가능함을 보였다. 그리고, 두 분류 문제를 하나의 구조로 적절히 학습하기 위한 방법과 분류 문제에서의 범주 불균형 문제를 해결하기 위한 분류 방법을 제시하였다.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2018.11a
• /
• pp.53-54
• /
• 2018

주어진 영상으로부터의 3 차원 얼굴 모델링은 얼굴 분석, 애니메이션, 생체 인식 등의 많은 컴퓨터비전 및 그래픽스 응용분야에서 중요한 역할을 하고 있다. 그 중에서도 헤어 영역은 얼굴에 비해 모양의 다양성과 모델의 복잡도가 현저히 높다. 기존의 연구는 주로 얼굴 영역에 한정한 3 차원 얼굴 모델링을 중심으로 이루어졌지만 헤어 모델링은 중요하게 다루지 않고 있는 경우가 많다. 본 논문에서는 심층인공신경망의 일종인 FCN (fully connected network)을 이용하여 인물 영상에서 헤어 부분을 영역화하고 그와 가장 유사한 3D 헤어 모델을 데이터베이스에서 검색하여 3 차원 얼굴 모델에 증강함으로써 완전한 얼굴 모델링을 수행하는 방법을 제안한다. 이는 FCN 을 이용하여 다양한 인물 영상에 대하여 네트워크 학습을 수행하는 과정과 3D 헤어 데이터베이스의 구축 과정을 포함한다. 실험 결과 적절한 수준의 헤어 모델이 3 차원 얼굴 모델링 결과물에 증강됨을 확인하였다.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2018.07a
• /
• pp.119-120
• /
• 2018

본 논문에서는 음악을 듣고 어떤 음악인지 인식하고 판별하는 음원분류 시스템과 해당 기술 구현을 딥러닝을 통해 적용하도록 제안하였다. 제안한 시스템은 인공심층신경망을 통해 음원파일을 여러 음원 특징 추출 모델에 따라 검출된 특징들을 학습하여 해당 음원의 고유한 보컬이나 반주의 특색 등을 찾아내어 이를 인식할 수 있도록 구현하였다. 이를 통해, 기존의 Fingerprint 방식의 데이터베이스 검색 시스템과는 다른 접근방식으로 보다 사람이 음악을 기억하는 방법에 가깝도록 구현하여 능동성과 유연성을 개선하고 다양한 응용분야로 활용할 수 있는 시스템을 제안하였다.

• Proceedings of the Korean Institute of Building Construction Conference
• /
• 2022.11a
• /
• pp.57-58
• /
• 2022

Concrete mix design is used as essential data for the quality of concrete, analysis of structures, and stable use of sustainable structures. However, since most of the formulation design is established based on the experience of experts, there is a lack of data to base it on. are suffering Accordingly, in this study, the purpose of this study is to build a predictive model to use the concrete mixing factor as basic data for calculation using the DNN technique. As for the data set for DNN model learning, OPC and ternary concrete data were collected according to the presence or absence of admixture, respectively, and the model was separated for OPC and ternary concrete, and training was carried out. In addition, by varying the number of hidden layers of the DNN model, the prediction performance was evaluated according to the model structure. The higher the number of hidden layers in the model, the higher the predictive performance for the prediction of the mixing elements except for the compressive strength factor set as the output value, and the ternary concrete model showed higher performance than the OPC. This is expected because the data set used when training the model also affected the training.

• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.18 no.4
• /
• pp.44-57
• /
• 2019

Various studies have been conducted to solve traffic congestions in many metropolitan cities through accurate traffic flow prediction. Most studies are based on the assumption that past traffic patterns repeat in the future. Models based on such an assumption fall short in case irregular traffic patterns abruptly occur. Instead, the approaches such as predicting traffic pattern through big data analytics and artificial intelligence have emerged. Specifically, deep learning algorithms such as RNN have been prevalent for tackling the problems of predicting temporal traffic flow as a time series. However, these algorithms do not perform well in terms of long-term prediction. In this paper, we take into account various external factors that may affect the traffic flows. We model the correlation between the multi-dimensional context information with temporal traffic speed pattern using deep neural networks. Our model trained with the traffic data from TOPIS system by Seoul, Korea can predict traffic speed on a specific date with the accuracy reaching nearly 90%. We expect that the accuracy can be improved further by taking into account additional factors such as accidents and constructions for the prediction.

• Journal of the Korea Society of Computer and Information
• /
• v.27 no.12
• /
• pp.59-68
• /
• 2022

This paper presents a method for 1:1 verification by comparing the similarity between the given real product image and the drawing image. The proposed method combines two existing CNN-based deep learning models to construct a Siamese Network. After extracting the feature vector of the image through the FC (Fully Connected) Layer of each network and comparing the similarity, if the real product image and the drawing image (front view, left and right side view, top view, etc) are the same product, the similarity is set to 1 for learning and, if it is a different product, the similarity is set to 0. The test (inference) model is a deep learning model that queries the real product image and the drawing image in pairs to determine whether the pair is the same product or not. In the proposed model, through a comparison of the similarity between the real product image and the drawing image, if the similarity is greater than or equal to a threshold value (Threshold: 0.5), it is determined that the product is the same, and if it is less than or equal to, it is determined that the product is a different product. The proposed model showed an accuracy of about 71.8% for a query to a product (positive: positive) with the same drawing as the real product, and an accuracy of about 83.1% for a query to a different product (positive: negative). In the future, we plan to conduct a study to improve the matching accuracy between the real product image and the drawing image by combining the parameter optimization study with the proposed model and adding processes such as data purification.

• Journal of the Korean Geosynthetics Society
• /
• v.22 no.2
• /
• pp.1-12
• /
• 2023

Research on the causes of landslides and prediction of vulnerable areas is being conducted globally. This study aims to predict the effective soil depth, a critical element in analyzing and forecasting landslide disasters, using topographic information. Topographic data from various institutions were collected and assigned as attribute information to a 100 m × 100 m grid, which was then reduced through data grading. The study predicted effective soil depth for two cases: three depths (shallow, normal, deep) and five depths (very shallow, shallow, normal, deep, very deep). Three classification models, including K-Nearest Neighbor, Random Forest, and Deep Artificial Neural Network, were used, and their performance was evaluated by calculating accuracy, precision, recall, and F1-score. Results showed that the performance was in the high 50% to early 70% range, with the accuracy of the three classification criteria being about 5% higher than the five criteria. Although the grading criteria and classification model's performance presented in this study are still insufficient, the application of the classification model is possible in predicting the effective soil depth. This study suggests the possibility of predicting more reliable values than the current effective soil depth, which assumes a large area uniformly.