• Annual Conference on Human and Language Technology
• /
• 2018.10a
• /
• pp.453-458
• /
• 2018

본 논문은 한국어 자소를 인식 단위로 사용한 hybrid CTC-Attention 모델 기반 end-to-end speech recognition을 제안한다. End-to-end speech recognition은 기존에 사용된 DNN-HMM 기반 음향 모델과 N-gram 기반 언어 모델, WFST를 이용한 decoding network라는 여러 개의 모듈로 이루어진 과정을 하나의 DNN network를 통해 처리하는 방법을 말한다. 본 논문에서는 end-to-end 모델의 출력을 추정하기 위해 자소 단위의 출력구조를 사용한다. 자소 기반으로 네트워크를 구성하는 경우, 추정해야 하는 출력 파라미터의 개수가 11,172개에서 49개로 줄어들어 보다 효율적인 학습이 가능하다. 이를 구현하기 위해, end-to-end 학습에 주로 사용되는 DNN 네트워크 구조인 CTC와 Attention network 모델을 조합하여 end-to-end 모델을 구성하였다. 실험 결과, 음절 오류율 기준 10.05%의 성능을 보였다.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.21 no.8
• /
• pp.521-529
• /
• 2020

Forest fires are one of the most important environmental risks that have adverse effects on many aspects of life, such as the economy, environment, and health. The early detection, quick prediction, and rapid response of forest fires can play an essential role in saving property and life from forest fire risks. For the rapid discovery of forest fires, there is a method using meteorological data obtained from local sensors installed in each area by the Meteorological Agency. Meteorological conditions (e.g., temperature, wind) influence forest fires. This study evaluated a Data Mining (DM) approach to predict the burned area of forest fires. Five DM models, e.g., Stochastic Gradient Descent (SGD), Support Vector Machines (SVM), Decision Tree (DT), Random Forests (RF), and Deep Neural Network (DNN), and four feature selection setups (using spatial, temporal, and weather attributes), were tested on recent real-world data collected from Gyeonggi-do area over the last five years. As a result of the experiment, a DNN model using only meteorological data showed the best performance. The proposed model was more effective in predicting the burned area of small forest fires, which are more frequent. This knowledge derived from the proposed prediction model is particularly useful for improving firefighting resource management.

• Advances in nano research
• /
• v.13 no.3
• /
• pp.297-312
• /
• 2022

Coughing and breath shortness are common symptoms of nano (small) cell lung cancer. Smoking is main factor in causing such cancers. The cancer cells form on the soft tissues of lung. Deformation behavior and wave vibration of lung affected when cancer cells exist. Therefore, in the current work, phase velocity behavior of the small cell lung cancer as a main part of the body via an exact size-dependent theory is presented. Regarding this problem, displacement fields of small cell lung cancer are obtained using first-order shear deformation theory with five parameters. Besides, the size-dependent small cell lung cancer is modeled via nonlocal stress/strain gradient theory (NSGT). An analytical method is applied for solving the governing equations of the small cell lung cancer structure. The novelty of the current study is the consideration of the five-parameter of displacement for curved panel, and porosity as well as NSGT are employed and solved using the analytical method. For more verification, the outcomes of this reports are compared with the predictions of deep neural network (DNN) with adaptive optimization method. A thorough parametric investigation is conducted on the effect of NSGT parameters, porosity and geometry on the phase velocity behavior of the small cell lung cancer structure.

• Journal of Information Processing Systems
• /
• v.19 no.3
• /
• pp.355-369
• /
• 2023

Since the widespread adoption of deep-learning and related distributed representation, there have been substantial advancements in part-of-speech (POS) tagging for many languages. When training word representations, morphology and shape are typically ignored, as these representations rely primarily on collecting syntactic and semantic aspects of words. However, for tasks like POS tagging, notably in morphologically rich and resource-limited language environments, the intra-word information is essential. In this study, we introduce a deep neural network (DNN) for POS tagging that learns character-level word representations and combines them with general word representations. Using the proposed approach and omitting hand-crafted features, we achieve 90.47%, 80.16%, and 79.32% accuracy on our own dataset for three morphologically rich languages: Uyghur, Uzbek, and Kyrgyz. The experimental results reveal that the presented character-based strategy greatly improves POS tagging performance for several morphologically rich languages (MRL) where character information is significant. Furthermore, when compared to the previously reported state-of-the-art POS tagging results for Turkish on the METU Turkish Treebank dataset, the proposed approach improved on the prior work slightly. As a result, the experimental results indicate that character-based representations outperform word-level representations for MRL performance. Our technique is also robust towards the-out-of-vocabulary issues and performs better on manually edited text.

• Journal of The Korean Society of Agricultural Engineers
• /
• v.62 no.5
• /
• pp.93-103
• /
• 2020

In this study, we estimated soil moisture values using the Deep Neural Network(DNN) scheme at the mountainous regions. In order to test the sensitive analysis of DNN scheme, we collected the measured(at the soil depths of 10 cm and 30 cm) soil moisture and DNN input(weather and land surface) data at the Pyeongchang-gun(relatively flat) and Geochang-gun(steep slope) sites. Our findings indicated that the soil moisture estimates were sensitive to the weather variables(5 days-averaged rainfall, 5 days precedent rainfall, accumlated rainfall) and DEM. These findings showed that the DEM and weather variables play the key role in the processes of soil water flow at the mountainous regions. We estimated the soil moisture values at the soil depths of 10 cm and 30 cm using DNN at two study sites under different climate-landsurface conditions. The estimated soil moisture(R: 0.890 and RMSE: 0.041) values at the soil depth of 10 cm were comparable with the measured data in Pyeongchang-gun site while the soil moisture estimates(R: 0.843 and RMSE: 0.048) at the soil depth of 30 cm were relatively biased. The DNN-based soil moisture values(R: 0.997/0.995 and RMSE: 0.014/0.006) at the soil depth of 10 cm/30 cm matched well with the measured data in Geochang-gun site. Although uncertainties exist in the results, our findings indicated that the DNN-based soil moisture estimation scheme demonstrated the good performance in estimating soil moisture values using weather and land surface information at the monitoring sites. Our proposed scheme can be useful for efficient land surface management in various areas such as agriculture, forest hydrology, etc.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.20 no.3
• /
• pp.89-95
• /
• 2020

Recently, studies are being conducted to recognize a sketch image of a GUI (Graphical User Interface) based on a deep learning and to make it into a code implemented in an application. UI / UX designers can communicate with developers through storyboards when developing mobile applications. However, UI / UX designers can create different widgets for ambiguous widgets. In this paper, we propose an automatic UI detection method using symbol markers to improve the accuracy of DNN (Deep Neural Network) based UI identification. In order to evaluate the performance with or without the symbol markers, their accuracy is compared. In order to improve the accuracy according to of the symbol marker, the results are analyzed when the shape is a circle or a parenthesis. The use of symbol markers will reduce feedback between developer and designer, time and cost, and reduce sketch image UI false positives and improve accuracy.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.24 no.6
• /
• pp.714-721
• /
• 2020

DNN is expanding its use in embedded systems such as robots and autonomous vehicles. For high recognition accuracy, computational complexity is greatly increased, and multiple DNNs are running aperiodically. Therefore, the ability processing multiple DNNs in embedded environments is a crucial issue. Accordingly, multicore based platforms are being released. However, most DNN models are operated in a batch process, and when multiple DNNs are operated in multicore together, the execution time deviation between each DNN may be large and the end-to-end execution time of the whole DNNs could be long depending on how they are allocated to the cores. In this paper, we solve these problems by providing a framework that decompose each DNN into individual layers and then distribute to multicores through a global queue. As a result of the experiment, the total DNN execution time was reduced by 31%, and when operating multiple identical DNNs, the deviation in execution time was reduced by up to 95.1%.

• KIPS Transactions on Software and Data Engineering
• /
• v.9 no.5
• /
• pp.169-176
• /
• 2020

Recently, deep neural networks (DNNs) have brought revolutions to many mobile and embedded devices by providing human-level machine intelligence for various applications. However, high inference accuracy of such DNNs comes at high computational costs, and, hence, there have been significant efforts to reduce computational overheads of DNNs either by compressing off-the-shelf models or by designing a new small footprint DNN architecture tailored to resource constrained devices. One notable recent paradigm in designing small footprint DNN models is sharing parameters in several layers. However, in previous approaches, the parameter-sharing techniques have been applied to large deep networks, such as ResNet, that are known to have high redundancy. In this paper, we propose a parameter-sharing method for already parameter-efficient small networks such as ShuffleNetV2. In our approach, small templates are combined with small layer-specific parameters to generate weights. Our experiment results on ImageNet and CIFAR100 datasets show that our approach can reduce the size of parameters by 15%-35% of ShuffleNetV2 while achieving smaller drops in accuracies compared to previous parameter-sharing and pruning approaches. We further show that the proposed approach is efficient in terms of latency and energy consumption on modern embedded devices.

• The Journal of the Korea Contents Association
• /
• v.20 no.8
• /
• pp.10-18
• /
• 2020

Among various contents industry, this study especially focused on music industry and tried to develop a prediction model for music box office using deep learning. The deep learning prediction model designed to predict music chart-in period based on 17 variables -singer power, singer influence, featuring singer power, featuring singer influence, number of participating singers, gender of participating singers, lyric writer power, composer power, arranger power, production agency power, distributing agency power, title track, LIKEs on streaming platform, comments on streaming platform, pre-promotion article, teaser-video view, first-week performance. Additionally we conducted a linear regression analysis to sort out factors, and tried to compare the prediction performance between the original DNN prediction model and the DNN model made of sorted out factors.

• Journal of Sensor Science and Technology
• /
• v.29 no.6
• /
• pp.433-439
• /
• 2020

Recently, the need for IoT-based intelligent systems is increasing in various fields. In this study, we implemented the system that automatically measures the sugar content of fruits without damage to fruit's marketability using near-infrared radiation and machine learning. The spectrums were measured several times by passing a broadband near-infrared light through a fruit, and the average value for them was used as the input raw data of the machine-learned DNN(Deep Neural Network). Using this system, he sugar content value of fruits could be predicted within 5 s, and the prediction accuracy was about 93.86%. The proposed non-destructive sugar content measurement system can predict a relatively accurate sugar content value within a short period of time, so it is considered to have sufficient potential for practical use.