• 한국정보시스템학회지:정보시스템연구
• /
• 제30권4호
• /
• pp.203-226
• /
• 2021

Purpose The purpose of this study is to enhance the accuracy score of personal credit scoring using the convolutional neural networks and secure the transparency of the deep learning model using eXplainalbe Artifical Inteligence(XAI) technique. Design/methodology/approach This study built a classification model by using the convolutional neural networks(CNN) and applied a methodology that is transformation of numerical data to imaged data to apply CNN on personal credit data. Then layer-wise relevance propagation(LRP) was applied to model we constructed to find what variables are more influenced to the output value. Findings According to the empirical analysis result, this study confirmed that accuracy score by model using CNN is highest among other models using logistic regression, neural networks, and support vector machines. In addition, With the LRP that is one of the technique of XAI, variables that have a great influence on calculating the output value for each observation could be found.

• ETRI Journal
• /
• 제42권4호
• /
• pp.505-517
• /
• 2020

The increasing size and complexity of deep neural networks (DNNs) necessitate the development of efficient high-performance accelerators. An efficient memory structure and operating scheme provide an intuitive solution for high-performance accelerators along with dataflow control. Furthermore, the processing of various neural networks (NNs) requires a flexible memory architecture, programmable control scheme, and automated optimizations. We first propose an efficient architecture with flexibility while operating at a high frequency despite the large memory and PE-array sizes. We then improve the efficiency and usability of our architecture by automating the optimization algorithm. The experimental results show that the architecture increases the data reuse; a diagonal write path improves the performance by 1.44× on average across a wide range of NNs. The automated optimizations significantly enhance the performance from 3.8× to 14.79× and further provide usability. Therefore, automating the optimization as well as designing an efficient architecture is critical to realizing high-performance DNN accelerators.

• 한국멀티미디어학회논문지
• /
• 제21권11호
• /
• pp.1244-1253
• /
• 2018

The color of traffic light is sensitive to various illumination conditions. Especially it loses the hue information when oversaturation happens on the lighting area. This paper proposes a traffic light recognition method robust to these illumination variations. The method consists of two steps of traffic light detection and recognition. It just uses the intensity and saturation in the first step of traffic light detection. It delays the use of hue information until it reaches to the second step of recognizing the signal of traffic light. We utilized a deep learning technique in the second step. We designed a deep convolutional neural network(DCNN) which is composed of three convolutional networks and two fully connected networks. 12 video clips were used to evaluate the performance of the proposed method. Experimental results show the performance of traffic light detection reporting the precision of 93.9%, the recall of 91.6%, and the recognition accuracy of 89.4%. Considering that the maximum distance between the camera and traffic lights is 70m, the results shows that the proposed method is effective.

• 한국융합학회논문지
• /
• 제9권5호
• /
• pp.1-6
• /
• 2018

스마트그리드에서 정확한 단기 부하 예측을 통한 자원의 이용 계획은 에너지 시스템 운영의 불확실성을 줄이고 운영 효율을 높이는데 있어서 매우 중요하다. 단기 부하 예측에 얕은 신경회로망을 포함한 다수의 머신 러닝 기법이 적용되어왔지만 예측 정확도의 개선이 요구되고 있다. 최근에는 컴퓨터 비전이나 음성인식 분야에서 심층 신경회로망의 뛰어난 연구 결과로 인해 심층 신경회로망을 단기 전력수요 예측에 적용해 예측 정확도를 개선하려는 시도가 주목 받고 있다. 본 논문에서는 일별 전력 부하 첨두치를 예측하기 위한 다층신경회로망 구조의 심층 신경회로망 모델을 제안한다. 제안된 심층 신경회로망은 층별 학습이 선행된 후 전체 모델의 학습이 이루어진다. 한국전력거래소에서 얻은 4년 동안의 일별 전력 수요 데이터를 사용, 하루 및 이틀 앞선 전력수요 첨두치를 예측하는 심층 신경회로망 모델을 구축하고 예측 정확도를 비교, 평가한다.

• 센서학회지
• /
• 제30권5호
• /
• pp.290-294
• /
• 2021

Experts have designed popular and successful model architectures, which, however, were not the optimal option for different scenarios. Despite the remarkable performances achieved by deep neural networks, manually designed networks for classification tasks are the backbone of object detection. One major challenge is the ImageNet pre-training of the search space representation; moreover, the searched network incurs huge computational cost. Therefore, to overcome the obstacle of the pre-training process, we introduce a network adaptation technique using a pre-trained backbone model tested on ImageNet. The adaptation method can efficiently adapt the manually designed network on ImageNet to the new object-detection task. Neural architecture search (NAS) is adopted to adapt the architecture of the network. The adaptation is conducted on the MobileNetV2 network. The proposed NAS is tested using SSDLite detector. The results demonstrate increased performance compared to existing network architecture in terms of search cost, total number of adder arithmetics (Madds), and mean Average Precision(mAP). The total computational cost of the proposed NAS is much less than that of the State Of The Art (SOTA) NAS method.

• 한국멀티미디어학회논문지
• /
• 제20권5호
• /
• pp.769-781
• /
• 2017

Deep learning networks like Convolutional Neural Networks (CNNs) show successful performances in many computer vision applications such as image classification, object detection, and so on. For implementation of deep learning networks in embedded system with limited processing power and memory, deep learning network may need to be simplified. However, simplified deep learning network cannot learn every possible scene. One realistic strategy for embedded deep learning network is to construct a simplified deep learning network model optimized for the scene images of the installation place. Then, automatic training will be necessitated for commercialization. In this paper, as an intermediate step toward automatic training under fisheye camera environments, we study more precise human localization in fisheye images, and propose an accurate human localization method, Automatic Ground-Truth Labelling Method (AGTLM). AGTLM first localizes candidate human object bounding boxes by utilizing GoogLeNet-LSTM approach, and after reassurance process by GoogLeNet-based CNN network, finally refines them more correctly and precisely(tightly) by applying saliency object detection technique. The performance improvement of the proposed human localization method, AGTLM with respect to accuracy and tightness is shown through several experiments.

• 한국항해항만학회:학술대회논문집
• /
• 한국항해항만학회 2022년도 추계학술대회
• /
• pp.170-171
• /
• 2022

A ship's maneuvering motion model is important in a computer simulation, especially under the trend of intelligent navigation. This model is usually constructed by the hydrodynamic parameters of the ship which are generated by the principles of hydrodynamics. Ship's motion model is a nonlinear function. By using this function, ships' motion elements can be calculated, then the ship's trajectory can be predicted. Deeping neural networks can construct any linear or non-linear equation theoretically if there have enough and sufficient training data. This study constructs some kinds of deep Networks and trains this network by real ship motion data, and chooses the best one of the networks, uses real data to train it, then uses it to predict the ship's trajectory, getting some conclusions and experiences.

• 한국융합학회논문지
• /
• 제9권6호
• /
• pp.39-44
• /
• 2018

심층 신경회로망은 적합한 수학적 모델에 대한 어떠한 가정 없이 데이터로부터 유용한 정보를 추출해서 예측에 필요한 입출력 관계를 정의할 수 있기 때문에 최근 시계열 예측 분야에서 주목 받고 있다. 본 논문에서는 주가의 일별 종가를 예측하기 위한 심층 신경회로망 모델을 제안한다. 제안된 심층 신경회로망은 예측 정밀도를 높이기 위해 단일 층의 오토인코더와 4층의 신경회로망이 결합된 구조를 갖는다. 오토인코더 층은 주가 예측에 필요한 최적의 입력 특징을 추출하고 4층의 신경회로망은 추출된 특징을 사용해 주가 예측에 필요한 동특성을 반영하여 주가를 출력한다. 제안된 심층 신경회로망의 학습은 층별로 단계적으로 이뤄지며 최종 단계에서 전체 심층 신경회로망에 대해 한 번 더 학습이 실행된다. 본 논문에 제안된 방법으로 KOrea composite Stock Price Index (KOSPI) 일별 종가를 예측하는 심층 신경회로망을 구현하고 기존 방법과 예측 정확도를 비교, 평가한다.

• 한국산업정보학회논문지
• /
• 제23권4호
• /
• pp.81-92
• /
• 2018

Bitcoin is a cryptographic digital currency and has been given a significant amount of attention in literature since it was first introduced by Satoshi Nakamoto in 2009. It has become an outstanding digital currency with a current market capitalization of approximately $60 billion. By 2019, it is expected to have over 5 million users. Nowadays, investing in Bitcoin is popular, and along with the advantages and disadvantages of Bitcoin, learning how to forecast is important for investors in their decision-making so that they are able to anticipate problems and earn a profit. However, most investors are reluctant to invest in bitcoin because it often fluctuates and is unpredictable, which may cost a lot of money. In this paper, we focus on solving the Bitcoin forecasting prediction problem based on deep learning structures and neural decomposition. First, we propose a deep learning-based framework for the bitcoin forecasting problem with deep feed forward neural network. Forecasting is a time-dependent data type; thus, to extract the information from the data requires decomposition as the feature extraction technique. Based on the results of the experiment, the use of neural decomposition and deep neural networks allows for accurate predictions of around 89%.

• Journal of Information Processing Systems
• /
• 제17권3호
• /
• pp.556-570
• /
• 2021

The existing video expression recognition methods mainly focus on the spatial feature extraction of video expression images, but tend to ignore the dynamic features of video sequences. To solve this problem, a multi-mode convolution neural network method is proposed to effectively improve the performance of facial expression recognition in video. Firstly, OpenFace 2.0 is used to detect face images in video, and two deep convolution neural networks are used to extract spatiotemporal expression features. Furthermore, spatial convolution neural network is used to extract the spatial information features of each static expression image, and the dynamic information feature is extracted from the optical flow information of multiple expression images based on temporal convolution neural network. Then, the spatiotemporal features learned by the two deep convolution neural networks are fused by multiplication. Finally, the fused features are input into support vector machine to realize the facial expression classification. Experimental results show that the recognition accuracy of the proposed method can reach 64.57% and 60.89%, respectively on RML and Baum-ls datasets. It is better than that of other contrast methods.